**Electrical Energy Prediction in Residential Buildings for Short-Term Horizons Using Hybrid Deep Learning Strategy**

#### **Zulfiqar Ahmad Khan, Amin Ullah, Waseem Ullah, Seungmin Rho, Miyoung Lee and Sung Wook Baik \***

Sejong University, Seoul 143-747, Korea; mzulfiqar3797@gmail.com (Z.A.K.); qamin3797@gmail.com (A.U.); kwaseem3797@gmail.com (W.U.); smrho@sejong.ac.kr (S.R.); miylee@sejong.ac.kr (M.L.)

**\*** Correspondence: sbaik@sejong.ac.kr

Received: 9 November 2020; Accepted: 29 November 2020; Published: 2 December 2020

**Abstract:** Smart grid technology based on renewable energy and energy storage systems are attracting considerable attention towards energy crises. Accurate and reliable model for electricity prediction is considered a key factor for a suitable energy management policy. Currently, electricity consumption is rapidly increasing due to the rise in human population and technology development. Therefore, in this study, we established a two-step methodology for residential building load prediction, which comprises two stages: in the first stage, the raw data of electricity consumption are refined for effective training; and the second step includes a hybrid model with the integration of convolutional neural network (CNN) and multilayer bidirectional gated recurrent unit (MB-GRU). The CNN layers are incorporated into the model as a feature extractor, while MB-GRU learns the sequences between electricity consumption data. The proposed model is evaluated using the root mean square error (RMSE), mean square error (MSE), and mean absolute error (MAE) metrics. Finally, our model is assessed over benchmark datasets that exhibited an extensive drop in the error rate in comparison to other techniques. The results indicated that the proposed model reduced errors over the individual household electricity consumption prediction (IHEPC) dataset (i.e., RMSE (5%), MSE (4%), and MAE (4%)), and for the appliances load prediction (AEP) dataset (i.e., RMSE (2%), and MAE (1%)).

**Keywords:** bidirectional gated recurrent unit; convolutional neural networks; electricity consumption prediction; hybrid deep learning model; residential load prediction

#### **1. Introduction**

The electric power industry plays an important role in the economic development of a country, and its decisive operation provides significant societal wellbeing. As reported in [1] the global energy consumption is increasing for sustainable advancement in society; therefore, the effectiveness of electricity consumption prediction needs to be improved [2]. As reported by the World Energy Outlook in 2017, the compound annual growth rate (CAGR) of electricity demand will have a global incremental rise of 1.0% in the period of 2016–2040 [3]. Another report presented in [4] described that residential buildings generally consume 27% of the total energy consumption, whereas buildings in the United States (US) consume 40% of their national energy [5]. Owing to high energy consumption levels in residential buildings, efficient management of their consumption is essential. Therefore, proper planning of energy is vital for energy saving, which is possible through effective energy consumption prediction models.

The electricity prediction strategies for short-term horizons are categorized into four types: very-short, short, medium, and long-term [6,7]. Short and very short-term predictions refer to minutely ahead predictions up to several days. Medium-term load prediction means one week ahead prediction

or up to a year, whereas annual or several years' ahead prediction is called very long-term prediction. Each electricity prediction horizon has its own applications; however, in this study, we focus on short-term and very-short-term predictions [6,8–10]. The major applications of short-term electric load prediction are power plant's reliable and secure operation, reliability and economic dispatch, and power system generation scheduling. Short-term load prediction ensures power system security, and it is an essential tool for the determination of the optimal operational state. Reliability and economic dispatch are also important applications of power systems in short-term horizons, wherein the abrupt variation of load demand fluctuates its reliability. This causes a power supply shortage if the load demand is underestimated, which makes it difficult to manage overload conditions and the quality of the overall power supply system. Another application of short-term load prediction is generation scheduling, which can be achieved through accurate load prediction to verify the allocation of operational limitations, generation resources, equipment usage, and environmental constraints. In the literature, several studies have been conducted for short-term load prediction, electricity load forecasting, electricity demand forecasting, electricity storage, and occupant behavior [11–15]. The mainstream electricity load prediction models are mainly grouped into two categories: statistical models and artificial intelligence models [16,17]. Statistical models such as Auto Regressive Integrated Moving Average (ARIMA) [18], linear regression [19], Kalman filtering [20], and clustering [21,22] etc. were used for load prediction in the early days. These models are effective in learning linear data but inadequate to learn the nonlinear complex electricity load. Besides this, the artificial intelligence models can learn nonlinear complex and linear electricity loads, which are further divided into shallow and deep structure methods. Shallow based methods include random forest [23], wavelet neural networks [24], support vector machines (SVMs) [25], artificial neural networks (ANN) [26], and extreme learning machines [27]. Shallow based methods perform well compared to statistical methods but perform poorly in feature mining. Hence, these methods require more features and selects strong features to enhance the prediction accuracy. Obtaining optimal feature extraction is a challenging task for these methods. Further, these methods have insufficient generalization ability over different datasets due to small hypothesis owing to less number of parameters. Deep structure methods have the ability to address the aforementioned concerns of shallow-based methods using multi-layer processing and hierarchical feature learning from electricity historical data. Recently, recurrent neural networks (RNNs) and convolutional neural networks (CNNs) are two powerful architecture proposed in the literature for the analysis of time series data. For instance, Amarasinghe et al. [28] developed a CNN-based methodology for electricity load forecasting and compared their results with a factored restricted Boltzmann machine, sequence-to-sequence long short term memory (LSTM), support vector regressor (SVR), and ANN. Another study presented in [29] proposed a deep CNN network for day-ahead load forecasting and compared the results with an extreme learning machine, ARIMA, CNN, and RNN. Several studies also used RNN models for electricity load prediction, whereas Tokgoz et al. [30] used RNN, gated recurrent unit (GRU), and LSTM models for electricity load prediction in Turkey and extensively decreased the error. Furthermore, the authors of [31] developed an LSTM-based model for periodic energy prediction and compared their results with other models. Another study presented in [32] developed an RNN-based model for medium- and long-term electricity load prediction.

The electricity consumption data is time-series data, which comprises spatial and temporal information. The CNN models perform well for spatial information extraction, but insufficient for temporal information, whereas the RNN models are insufficient for spatial information and can learn temporal information. Therefore, to develop an optimal model for electricity load prediction, hybrid models are introduced in the recent literature. For instance, Kim et al. [33] developed a hybrid model combining CNN with LSTM for short-term load prediction and compared their results with GRU, attention LSTM, LSTM, and bidirectional LSTM. Ullah et al. [34] also developed a hybrid model with a combination of CNN and multi-layer bidirectional LSTM and compared their results with bidirectional LSTM, LSTM, and CNN-LSTM. Similarly, another study presented in [17] integrated a CNN with an

LSTM auto-encoder and compared the final results with LSTM, LSTM autoencoder, and CNN-LSTM. Moreover, Sajjad et al. [35] and Afrasiabi et al. [36] presented the performance of a CNN-GRU based model for electricity forecasting. The performance of hybrid models is quite promising and has achieved state-of-the-art accuracy; however, further improvements are needed for optimal electricity load prediction. Therefore, in the current study, we established a two-step framework for predicting the electricity load that includes data preprocessing and proposed a hybrid model. In the first step, the historical data of electricity is refined to remove abnormalities that are then passed to the next step CNN along with MB-GRU model for learning. To extract the spatial information, we used CNN layers where MB-GRU is used to learn the temporal information. The contributions of the proposed research are summarized below:


The main goal of this work is to improve the prediction accuracy for short-term electrical load prediction in residential buildings, which reduces customer consumption and provides economic benefits. The experimental section shows the effectiveness of the proposed method, which ensures the best performance of the proposed method as compared to other baseline models.

The remainder of the paper is arranged as follows: Section 2 provides a detailed explanation of the proposed method; Section 3 includes the experimental results of the proposed method and comparison with other state-of-the-art models. Finally, the manuscript is concluded in Section 4.

#### **2. Proposed Framework**

Accurate electricity load prediction is very important for electricity saving and vital economic implications [37]. As reported by [38] a 1% decrease in the error rate of the electricity prediction model can profit 1.6M dollars and can save 10K MW of electricity annually. For accurate load prediction, an appropriate learning methodology is required. The electricity load prediction models are learned from historical data generated from smart meter sensors. However, some times, due to weather conditions, meter faults, etc., they generate some abnormal data that should be refined before training. Therefore, this work represents a two-step framework that includes data preprocessing and the proposed hybrid model. The preprocessing step refines the input raw data of electricity consumption and then passes it to the proposed model to learn it, as demonstrated in Figure 1, where the details of each step are further discussed in the following sections.

**Figure 1.** Two steps framework for electricity load prediction. (**a**) Applying different preprocessing techniques to refine the input raw dataset. (**b**) The proposed convolutional neural network (CNN) multilayer bidirectional (MB)-gated recurrent unit (GRU) architecture to learn the patterns of electricity consumption.

#### *2.1. Data Preprocessing*

For better performance of the electricity load prediction models, the training data should be analyzed before training. As previously mentioned, the historical data of electricity consumption include abnormalities that affect the model performance. In this study, we used individual household electricity consumption prediction (IHEPC) and appliances load prediction (AEP) datasets, which consist of missing and outlier values. These abnormalities are removed from the data in the preprocessing step of the proposed framework. For missing values filling, NAN interpolation techniques are used, whereas for outlier values, three sigma rules of thumb [39] are applied. After the outlier reduction and filling missing values, the datasets are normalized using the min–max normalization technique to transform the dataset into a particular range which the neural network can learn easily.

#### *2.2. Proposed CNN and MBGRU Architecture*

This work combines a CNN with a multilayered bidirectional GRU (MB-GRU) for short-term electricity load prediction, where the CNN layers are incorporated to extract features from the preprocessed input data, and the MB-GRU model is used to learn the sequences between them. CNN is a neural network architecture that learns in a hierarchical manner whereby each layer learns more and more abstract features. The first layers learn atomic/primitive representations, while the intermediate-level layers learn intermediate abstract representations, and finally, fully connected layers learn high-level patterns. Therefore, the depth of the network is defined by the number of such layers. The higher the layer count, the deeper the network and can learn tiny representations. A CNN is a particular type of deep neural network that employs alternating layers of convolutions and pooling. It contains trainable filter banks per layer. Each individual filter in a filter bank, which is called a kernel and has a fixed receptive field (window) that is scanned over a layer below it, to compute an output feature map. The kernel performs a simple dot product and bias computation as it scans the layer below it and then feeds the result through an activation function, a rectifier, for example, to compute the output map. The output map is then subsampled using sum or max pooling, the latter being more common in order to reduce sensitivity to distortions in the upper layers. This process is alternated up to some point when the features become specific to the problem at hand. Thus, the CNN is a deep neural network for learning increase and more compact features that can later be used for recognition problems. The last few layers in a typical CNN comprise a typical fully connected neural

network or support vector machines in order to recognize different combinations of features from the convolutional layers. The CNN architecture is used in different domains, such as image and video recognition [17,35,40,41], language processing [42,43], electricity load forecasting [44,45], crowed counting [46], etc. In the time series domain, the CNN layers are used to extract spatial information and then pass the output into sequential learning algorithms such as RNN LSTM and GRU.

RNN [47] is a sequence learning architecture with backward connections among hidden layers that include some kind of memory and is extensively used in several domains such as natural language processing [48], time series analysis [49], and speech recognition [50], visual data processing [51–53], etc. The RNN models generate output at each time stamp from the input data, which leads to the vanishing gradient problem. The RNN model forgets the long sequence of electricity data, such as 60-min resolution, which leads to loss of important information.

ʘ"

ʘ"

ʘ"

ʘ"

ʘ"

ʘ"

ʘ"

ʘ"

ʘ"

$$\mathbf{f}\_{\mathsf{f}} = \odot(\acute{\boldsymbol{\omega}}\_{\mathsf{f}} \cdot [\mathbb{A}\_{\mathsf{f}-1\prime} \ \boldsymbol{a}\_{\mathsf{f}}] + \boldsymbol{\theta}\_{\mathsf{f}}) \tag{1}$$

$$
\dot{a}\_{\mathsf{T}} = \odot(\acute{\omega}\_{\mathsf{i}} \cdot \left[\mathbb{A}\_{\mathsf{T}-1\prime} \ a\_{\mathsf{T}}\right] + \emptyset\_{\mathsf{i}}) \tag{2}
$$

$$c\_{\mathbb{T}} = \tan \mathbb{A} \left( \hat{\omega}\_{\boldsymbol{\epsilon}} \cdot \left[ \mathbb{A}\_{\mathbb{T}-1\prime} \ a\_{\mathbb{T}} \right] + \mathcal{J}\_{\boldsymbol{\epsilon}} \right) \tag{3}$$

$$\mathcal{C}\_{\mathsf{T}} = \mathsf{f}\_{\mathsf{T}} \times \mathcal{C}\_{\mathsf{T}-1} + \ \mathsf{i}\_{\mathsf{T}} \times \ \mathsf{c}\_{\mathsf{T}} \tag{4}$$

$$\mathcal{O}\_{\mathsf{T}} = \odot(\dot{\omega}\_{\mathsf{O}} \cdot [\mathbb{A}\_{\mathsf{T}-1\prime} \ a\_{\mathsf{T}}] + \varnothing\_{\mathsf{O}}) \tag{5}$$

$$\mathcal{H}\_{\mathsf{T}} = \mathcal{O}\_{\mathsf{T}} \times \tan \mathcal{A} \left( \odot \left( \mathcal{C}\_{\mathsf{T}} \right) . \tag{6}$$

ʘ" ʘ" ʘ" The problem of losing long sequence information is addressed by LSTM using the three-gate mechanism input, output, and forget. The mathematical representation of each gate is shown in Equations (1)–(6). In these equations the output of each gates is represented through "", "ê" and "O" where "*-*" represents the activation function. The weights of the gates are represented through "*ω*´ ", whereas" ʘ" <sup>Ʈ</sup>−1" refers to the output of previous LSTM block and "" represents the bias of the gates. The LSTM structure is complex and computationally expensive due to these gates' units and memory cells. To overcome the concern of LSTM, another lightweight architecture is developed called GRU [54] which comprises the reset and update gates. The mathematical representation of GRU gates are shown in Equations (7)–(10) where the update gate examines the earlier cell memory to remain active,

ʘ"

ʘ"

ʘ"

ʘ"

$$\varkappa\_{\mathbb{T}} = \odot(\acute{\omega}\_{\mathbb{x}} \cdot [\mathbb{X}\_{\mathbb{T}-1}, a\_{\mathbb{T}}] + \vartheta\_{\mathbb{x}}) \tag{7}$$

$$\mathcal{F}\_{\mathsf{T}} = \odot(\omega\_{\mathsf{r}} \cdot [\mathbb{X}\_{\mathsf{T}-1}, \mathsf{a}\_{\mathsf{T}}] + \mathsf{d}\_{\mathsf{r}}) \tag{8}$$

$$\mathcal{A}\_{\mathbb{L}} = \tan \hbar \left( \left. \dot{\omega}\_{\mathbb{A}} \cdot \left[ \nu\_{\mathbb{T}} \cdot \mathbb{A}\_{\mathbb{T}-1'} a\_{\mathbb{T}} \right] + \left. \theta\_{\mathbb{A}} \right) \right| \tag{9}$$

$$\mathbb{V}\_{\mathbb{T}} = (1 - \varkappa\_{\mathbb{T}}) \cdot \mathbb{V}\_{\mathbb{T}-1} + \varkappa\_{\mathbb{T}} \cdot \mathbb{V}\_{\mathbb{T}}) \tag{10}$$

and the reset gate merges the next cell input sequence with previous cell memory. In this study, we used MB-GRU, which processes the sequence of input data in both backward and forward directions [55]. The bidirectional RNN models perform better in several domains such as classification, summarization [56], and load forecasting [57]. Therefore, in this study, we incorporate bidirectional GRU layers that contain both backward and forward layers, where the output sequence of the foreword layer is iteratively calculated through input in the positive sequence. The output of the backward layer is calculated through the reverse of the input.

The electricity consumption patterns include spatial and temporal features. Some researchers deployed a solo model that is insufficient to extract both types of features at a time. Therefore, in this work, we established a hybrid model that combines CNN with MB-GRU, as shown in Figure 1b. The proposed hybrid model includes an input layer, CNN layers, and bidirectional GRU layers. Two CNN layers are incorporated after several experiments over different layers and different parameters. Finally, we select filters of 8 and 4 for the first and second CNN layers with a kernel size

of 3.1, respectively and used ReLU as an activation function in these layers. After convolutional layers, two bidirectional GRU layers are incorporated to learn the temporal information of the electricity historical data. Finally, the fully connected layers are integrated for the final output prediction.

#### **3. Results and Discussion**

In this section, we provide a detailed description of the dataset, evaluation metrics, and experimentation over the IHEPC and AEP datasets, and compare them with other baseline models. The model was trained over a GeFore GTX 2060 GPU with 64 GB RAM using the Keras framework with backend TensorFlow.

#### *3.1. Datasets*

The model's performance is assessed on two benchmark datasets, AEP and IHEPC [58,59]. The AEP dataset was recorded in 4.5 months in a residential house in 10 min resolution. This dataset comprises 29 various parameters of weather information (wind speed, humidity, dew point, temperature, and pressure), light, and appliance energy consumption, as presented in Table 1. The data samples were collected from both indoor and outdoor environments through a wireless sensor network. The outdoor data are collected from nearby airport. The building includes 9 indoor and 1 outdoor temperature sensors, 9 humidity sensors in which 7 are integrated with indoor environment and one is in outdoor environment. The outdoor pressure, visibility, temperature, humidity, and dew point are recorded nearby airport region. The IHEPC dataset includes 9 parameters which are; date, time, voltage, global-active-power (GAP), intensity, global-reactive-power (GRP) and three sub-metering as shown in Table 2. The dataset was recorded in a residential house in France during 2006 and 2010 for one-minute resolution.


#### *3.2. Metrics of Evaluation*

To evaluate the performance of the model, we used RMSE, MSE, and MAE metrics. The mathematical representation of these metrics is depicted in Equations (11)–(13). RMSE calculates the difference between all predicted data points and the actual data point, then compute the mean of these square errors and finally calculate the square root of the mean values. The MSE calculates the mean disparity between the actual and model output values. The MAE calculates the mean absolute difference between the actual and predicted values.

$$RMSE \;= \sqrt{\frac{1}{n} \sum\_{1}^{n} (y - \hat{y})^2} \tag{11}$$

$$MSE = \frac{1}{n} \sum\_{1}^{n} (y - \hat{y})^2 \tag{12}$$

$$\text{MAE} = \frac{1}{n} \sum\_{1}^{n} |y - \hat{y}| \tag{13}$$

#### *3.3. Experimentations over IHEPC, AEP Dataset and Comparison with other Models*

In this section, we evaluate the performance comparison of the proposed model with existing models for short-term load prediction (one hour ahead) over the IHEPC and AEP datasets. For the IHEPC dataset, the proposed model achieved 0.42, 0.18, and 0.29 RMSE, MSE, and MAE, respectively. The performance prediction over the test data is displayed in Figure 2a. A comparison of the proposed model over IHEPC dataset for short-term load prediction with other baseline models is shown in Figure 3. For more detail the performance of the proposed model is compared with [1,17,33–35,60,61] in the short-term horizon. In [60] the authors used deep learning methodology for residential load prediction and obtained 0.79 RMSE and 0.59 MAE, whereas [33] used a CNN-LSTM hybrid network for short-term residential load prediction and achieved 0.59, 0.35, and 0.33 RMSE, MSE, and MAE, respectively. 0.47, 0.19 and 0.31 RMSE, MSE and MAE was reported in [17] whereas [34] reports 0.56, 0.31 and 0.34 values for these metrics. In [61] the authors achieved 0.38 MSE and 0.39 MAE, whereas [1] reported 0.66 RMSE. Another strategy presented in [35] attained 0.47, 0.22, and 0.33 RMSE, MSE, and MAE, respectively. Among these results, the proposed model achieved the lowest error rate for short-term electric load prediction.

**Table 2.** Individual household electricity consumption prediction (IHEPC) dataset variables description and units.


**Figure 2.** Predictions results of our framework over test data (**a**) actual and predicted values for IHEPC dataset (**b**) actual and predicted values for AEP dataset.

**Figure 3.** Performance comparison of the proposed model with the other state-of-the-art models over IHEPC dataset.

Furthermore, the effectiveness of the proposed model is evaluated over the AEP dataset for a short-term horizon. For the AEP dataset, the proposed model attained 0.31, 0.10, 0.33 RMSE, MSE, and MAE, respectively, whereas the prediction results are shown in Figure 2b. Similarly, the effectiveness of the proposed model over the AEP dataset is also compared with other baseline models, as shown in Figure 4. For instance, the results are compared with [34,62–64]. For further details, Ref. [62] achieved 0.59 RMSE and 0.26 MSE, Ref. [63] achieved 0.35 RMSE and 0.66 MSE, and [64] achieved a 0.59 RMSE. In [35], authors achieved 0.31, 0.09, and 0.24 scores for RMSE, MSE, and MAE, respectively. Compared to these models, the proposed model performed better in reducing the RMSE and MAE error rate, while only the results of Sajjad et al. [35] in terms of MSE is better than the proposed model in the short-term horizon.

**Figure 4.** Performance contrast of the proposed model with the other baseline models over AEP dataset.

#### **4. Conclusions**

In this study, we established a two-step methodology for short-term load prediction. In the first step, we performed data preprocessing over raw data to refine it for training. The refinement of the data is important because the historical data of energy consumption is generated from smart meter sensors, which include abnormalities such as outliers, missing values etc. These abnormalities from the raw data are extracted in this step, and finally, the normalization technique is applied to transform the data into a specific range. The second step is the hybrid model, which is a combination of CNN and multilayered bi-directional GRU (MB-GRU). The CNN layers are incorporated to extract important features from the refined data, while the MB-GRU layers are used to learn the temporal information of electricity consumption data. The proposed methodology is tested over two challenging datasets and achieves better performance when compared to other methods, as demonstrated in the results section.

**Author Contributions:** Conceptualization, Z.A.K.; methodology, Z.A.K.; software, Z.A.K.; validation, W.U. and A.U.; formal analysis, M.Y.L.; investigation, A.U.; resources, S.W.B.; data curation, Z.A.K.; writing—original draft preparation, Z.A.K.; writing—review and editing, Z.A.K., W.U. and A.U.; visualization, W.U.; supervision, S.W.B.; project administration, M.L.; funding acquisition, S.W.B. and S.R. All authors have read and agreed to the published version of the manuscript.

**Funding:** This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korean government (MSIT) (No. 2019M3F2A1073179).

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Article* **Evaluating the Impact of the COVID-19 Pandemic on Residential Energy Use in Los Angeles**

**Michael J. Klopfer 1, Joy E. Pixley 1, Armen Saiyan 2, Amir Tabakh 2, David Jacot <sup>2</sup> and Guann-Pyng Li 1,\***


Armen.Saiyan@ladwp.com (A.S.); Amir.Tabakh@ladwp.com (A.T.); david.jacot@ladwp.com (D.J.) **\*** Correspondence: gpli@uci.edu; Tel.: +1-(949)-824-9073

#### **Featured Application: This work is potentially useful for updating residential energy usage models, considering presented energy impacts of the COVID-19 pandemic.**

**Abstract:** The 2020 COVID-19 pandemic provided an opportunity to assess energy use during times of emergency that disrupt daily and seasonal patterns. The authors present findings from a regional evaluation in the city of Los Angeles (California, USA) with broad application to other areas and demonstrate an approach for isolating and analyzing residential loads from community-level electric utility feeder data. The study addresses effects on residential energy use and the implications for future energy use models, energy planning, and device energy standards and utility program development. In this study we review changes in residential energy use during the progression of the COVID-19 pandemic from four residential communities across Los Angeles covering approximately 6603 households within two microclimate sub regional areas (Los Angeles Basin and San Fernando Valley). Analyses address both absolute and seasonal temperature-corrected energy use changes while assessing estimated changes on energy usage from both temperature-sensitive loads (e.g., air conditioning and electric heating) and non-temperature-sensitive loads (e.g., consumer electronics and major appliance use). An average 5.1% increase in total residential energy use was observed for non-temperature sensitive loads during the pandemic period compared to a 2018–2019 baseline. During mid-spring when shelter in place activity was highest a peak monthly energy use of 20.9% increase was seen compared to a 2018–2019 composite baseline. Considering an average of the top five warmest summer days, a 9.5% increase in energy use was observed for events during summer 2020 compared to summer 2018 (a year with similar magnitude summer high heat events). Based on these results, a potential trend is identified for increased residential load during pandemics and other shelter-in-place disruptions, net of any temperature-sensitive load shifts with greater impacts expected for lower-income communities.

**Keywords:** residential energy modeling; COVID-19; coronavirus pandemic; temperature sensitivity; energy security

#### **1. Introduction**

In 2020, changes in energy use and emissions were seen worldwide as a direct effect of the COVID-19 pandemic [1–6]. Mandatory stay-at-home periods globally reduced jet and aviation fuel by 50%, gasoline by 30%, and electricity (on average) about 10 percent during the early pandemic where shelter-in-place (SIP) orders were widespread across many regions. This reduction was followed by partial rebounds for all mentioned energy types later in 2020 [2,7–11]. While commercial transport and mobility to support commercial activities (e.g., commuting for work) were greatly reduced by a curtailment in overall business activities, the impact on residential energy use is harder to directly assess from

**Citation:** Klopfer, M.J.; Pixley, J.E.; Saiyan, A.; Tabakh, A.; Jacot, D.; Li, G.-P. Evaluating the Impact of the COVID-19 Pandemic on Residential Energy Use in Los Angeles. *Appl. Sci.* **2021**, *11*, 4476. https://doi.org/ 10.3390/app11104476

Academic Editor: Matti Lehtonen

Received: 12 April 2021 Accepted: 7 May 2021 Published: 14 May 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

publicly available electrical grid regional operator data. Preliminary results from studies early in the pandemic suggest increased residential energy use, but results vary [12–14]. Further, little attention has been paid to understanding the mechanisms leading to this change in energy use during both the early pandemic SIP periods and periods following, in addition to regressive periods due to regional re-closures due to increased COVID-19 cases.

Analysis of total energy use for a given region provides conclusions for macro trends. However, analyzing data comprised of heavily mixed sectors (residential and commercial loads) and as a combined set across all day types (weekends and weekdays) provides limited utility for sector-based analysis, and complicates actionable model adjustments for energy planning and conservation efforts. Approximately 21% of energy use nationwide is from residential customers [15]. Residential energy efficiency is a substantial focus for utility programs, but sector changes can be obscured within direct regional load figures. While a general decrease in energy use was broadly observed across most regions worldwide during the 2020 COVID-19 pandemic, modeling and planning difficulties when predicting future demand led to service disruptions. Most notably, poor forecasting models for pandemic-related changes in energy use directly led to widespread rolling blackouts in California in mid-August of 2020 during a substantial heatwave [16,17]. The 2020 pandemic period exhibited increased reliance on non-dispatchable, low carbon energy sources, with increases of 22.3% solar production and 13.5% wind production in the US compared to 2019 [18]. Understanding sector-focused changes in energy use helps improve demand predictions for future widespread lockdown events in an era of increasing effects of climate change and increased reliance on non-dispatchable and distributed generation.

Residential electric load is primarily comprised of the following major load categories: electricity-driven space conditioning (air conditioning, ventilation/forced air circulation, and electric heaters), lighting, major appliances, miscellaneous (plug) loads, constant building loads, and electric transportation. Of these categories, only space conditioning is directly temperature sensitive. Demand from three other categories—lighting, major appliances, and plug loads—is largely driven by occupancy without substantial regard to ambient temperature. With 42% of US residential use due to space conditioning, ambient temperature is a primary driver of residential electricity use, especially with high air conditioning penetration [15,18,19]. Despite the mild climate in Southern California, Chen et al. assessed a substantial (69% estimated) regional household penetration for air conditioning [19,20]. This includes residential air conditioning systems in different form factors and cooling capacities. For temperature-sensitive loads, both increased occupancy and the reaction of occupants to change in the ambient temperature affect energy use. For the remaining categories, changes in daily occupancy rates (occupied by none versus one or more individuals) and resulting changes in device use behavior (i.e., which loads or devices are used and how they are used) are the main considerations.

Residential occupancy shifted substantially for much of the population during the pandemic, particularly early in the pandemic timeline. While exact assessments of stayat-home rates are difficult, general trends show higher rates of SIP compliance early-on following the first COVID-19 case wave with proportional compliance (SIP compliance compared to present active COVID-19 cases) generally dropping during the following COVID-19 case waves throughout 2020. In Los Angeles County, mobility data indicates estimated stay-at-home rates of 50.6% of individuals on April 11 and dropping to 35.5% of individuals by September 1 (compared to approximately 25% during mid-February) [21]. Similarly, in a national Gallup study, 49% of respondents reported being likely to shelter in place if asked to during a third surge in late 2020, compared to 67% in early April 2020 during the first surge [22]. SIP restrictions reduced leisure activities in evenings and especially on weekends, but primarily impacted weekday occupancy through three mechanisms: a shift toward working from home, reduced access to educational facilities for students and educators, and increased unemployment [23]. During 2020, Los Angeles experienced a maximum unemployment rate of 18.8% in May 2020 with a recovery to 12.3% by December 2020 compared to a pre-pandemic level of 4.9% in February 2020 (non-

seasonally adjusted) [24]. The majority of jobs lost across the USA (as in other countries) were in leisure, hospitality, entertainment, manufacturing, and food services sectors, with pandemic-related job loss disproportionally impacting women, younger workers, and workers with less education [25]. Minor shifts in population impacting household size also occurred during the early pandemic: in a June 2020 Pew Research Center study 6% of respondents reported gaining a household member and 3% reported moving because of the pandemic. Of those who moved, 61% of respondents reporting moving into a family member's home. The shutdown of college campuses (25%), the desire to be with family (20%), and financial related reasons (18%) were major relocation catalysts, and relocations were highest among young adults (ages 18–29) [26].

The current study analyzed energy use data from distribution station feeder loads, specific to defined geographic areas in the city of Los Angeles, accessed using generalized utility supervisor control data acquisition (SCADA). Such grouped load data is often the only measure available. Prior investigations have identified limitations in using it in standard linear regression-based energy prediction models due to autocorrelation and homoscedasticity. There are also limits when relying on temperature data at high time scale resolutions (e.g., per day), given the shifts in energy use corresponding to behavior variation over the course of the day. However, for certain use cases comparing daily average energy use to daily temperature data has been demonstrated to provide satisfactory estimation figures [27,28]. Here, the authors demonstrate an approach for analyzing grouped load data and daily temperature values to provide insight into how energy use changes due to widespread emergency conditions such as the COVID-19 pandemic.

With a diverse population and a warm, dry climate and typically temperate spring, Los Angeles provides a near-ideal environment to assess the impact of the pandemic on residential utility customers, especially assessing non-temperature sensitive load contribution to total residential energy use. In addition, the city of Los Angeles provides a useful case study because it was substantially impacted by the COVID-19 pandemic in both number of COVID-19 cases in addition to state and local restrictions on business, services, and travel. California's aggressive stay-at-home order was initiated on 19 March and was followed by a relaxation in June, a partial reinstatement in July (following the start of a second wave of COVID-19 cases), a relaxation in September and an amended limited stay at home order issued in late November following through the end of the year (in response to a third wave of COVID-19 cases). Los Angeles County (the major regional health reporting resource covering the city of Los Angeles) suffered three successively increasing waves of COVID-19 case peaks in 2020, occurring on 8 April, 22 July, and 27 December with this one county representing 32% of all cases statewide (note that approximately 25.5% of California's population lives in Los Angeles County) [29,30]. As of 31 December, 7.7% of the LA county population had been infected with COVID-19. The city of Los Angeles regularly maintained stricter controls on business activities to reduce population movement compared to both state and county COVID-19 guidelines [30,31]. A follow-up SIP order to the one issued in spring focused on Los Angeles, beginning 30 November and continuing through 31 December, this was the strictest order in the state of California, effectively banning most outdoor gatherings, restricting employment travel, and reducing retail capacity. Accordingly, the city of Los Angeles provides a rich opportunity to draw transferrable lessons on energy responses to major behavioral shifts.

#### **2. Materials and Methods**

#### *2.1. Feeder and Data Selection*

This study used Los Angeles Department of Water and Power (LADWP) municipal electrical substation 4.7 kV customer distribution feeder net loads (reported in hourly average kW load values) servicing a designated geographic territory within the city of Los Angeles to provide serviced population load data. Comparison baselines were created from composites of 2018 and 2019 load data from individual feeders, while the evaluation period was initiated with the California SIP order on 19 March 2020 and continued through the end

of reporting on 31 December 2020 [31,32]. For comparison as needed an evaluation period beginning on 1 January 2020 is used to report energy use change with respect to calendar year. To compare heatwave-related events, period to period comparisons between single years were used. From a pool of all available feeders the authors performed a two-tier screening process. The first tier selected for feeders that serviced primarily residential customers (greater than 90% residential customers with largely negligible pre-pandemic observable commercial loading patterns). The second tier selected for diversity across the city considering community location, community income, and community building types. Feeders were then excluded for substantial service interruptions or for major changes from 2018 through 2020 rendering those periods non-comparable, including substantial changes in customer base, major new construction, building demolition, or zoning changes. Feeders were also excluded for exhibiting great heterogeneity of income across neighborhoods served by the same feeder, with the exception of Feeder A providing service to Section 8 subsidized low-income housing in Watts. Four feeders were ultimately selected for the current analysis, serving distinct communities across Los Angeles with a total residential customer base of approximately 6603 combined residential customers covering areas with a range of median incomes (see Table 1).


**Table 1.** Summary of sampled feeders including feeder service area and service demographic information.

Refer to Supplementary Table S1 for additional details. <sup>1</sup> Presented with community name, reference zip/postal code tabulation area (ZCTA) inclusive of served feeder area; note that Los Angeles-Long Beach Census tract codes inclusive of feeder service area are presented in the supplementary extension of this table. <sup>2</sup> Los Angeles (LA) Basin or San Fernando (SF) Valley; NWS weather station (WS), ICAO airport code used for identification; temperature reference and corresponding microclimate region. <sup>3</sup> Median income of service area inclusive ZCTA, and specific feeder service area median income. <sup>4</sup> Feeder service area includes single and multi-family homes, small apartment complexes. <sup>5</sup> Feeder service area includes public housing. <sup>6</sup> This feeder corresponds to a service area bordering North Hollywood (Los Angeles) and Burbank and is served by LADWP. <sup>7</sup> Single and multi-family homes, small apartment complexes near commercial district. <sup>8</sup> Feeder service area includes a large apartment community, low rise with numerous common facilities. <sup>9</sup> Pair of mid-rise buildings mixed retail, mercantile, offices, buildings in LA Jewelry District.

> Feeders were evaluated across the period of investigation from 2018 through 2020. Customer construction permit records indicate <6% mid-day solar load contribution total, with slow growth, and 2017 motor vehicle records showed <5% average customer EV penetration average across all primary evaluated feeder service areas (see Table S1). Both factors suggest a low overall impact such that the change between the pandemic and pre-pandemic periods for the evaluated feeders and accordingly the differential impact from solar and EV loads are treated as negligible. The majority of the building types represented were a mixture of single-family homes and small multi-family properties hosting several units, with a smaller proportion of low-rise apartment complexes. The communities assessed represented two microclimates: the Los Angeles Basin and San Gabriel Valley. The California Energy Commission designates two of these feeders (A and

B) in climate zone (CZ) #8 and two (C and D) in CZ #9 [33]. Typically, temperature data would be collected from a weather station in the same CZ. However, the neighborhoods served by Feeders A and B are on the border of CZ #9 and exhibit more similarity with the weather station in CZ #8 than with the closest weather station in CZ #9, which is farther away and on the coast. For this reason, the closest weather station is used for all analyses, regardless of designated CZ.

Along with the residential distribution feeder data (listed as primary feeders) used in these analyses, two example feeders are provided for additional context in the discussion section, representing a large apartment complex and a commercial zone.

System-wide net power load (NPL) was sourced directly from LADWP. Reported NPL summarizes full system net load (not including customer onsite co-generation) on an hourly basis for 2018 through 2020. All power data was analyzed with ambient temperature data, which was sourced from local National Weather Service (NWS) weather stations via a third party sourcing utility, MesoWest/SynopticLabs [34]. Load data was temporally correlated with weather data interpolated to the nearest hour using Universal Translator 3 (UT Online, Pacific Energy Center, Pacific Gas and Electric Corp., San Francisco, CA, USA) [35] and Easy Data Transform (Oryx Digital Ltd., Swindon, Wiltshire, England, UK) [36] software packages. When city data is not available with respect to COVID-19 caseloads and stayat-home rates, data scoped at the inclusive Los Angeles County or California state level is used.

#### *2.2. Load Evaluation*

First, individual feeders' average loads were compared on a monthly or weekly basis (using an ISO 8601 defined week—see Table S4) across the period of study without temperature normalization or restriction. Major holidays were excluded from categorization. Analysis of input data and calculations were performed in kW and kWh.

For temperature analyses, hourly average temperature values were used along with monthly degree day values, which were assessed from local NWS observation weather station monthly reports (see Table 1 for data source information). Interpolative re-sampling was used to correlate temperature data to load data. Temperature data and derivative units were converted from ◦F to ◦C for final reporting and rounded to the nearest 0.1 ◦C for reported values.

Two effects of temperature on load must be distinguished in these analyses. First, the expected effect of temperature on electricity use (particularly cooling on hot days) must be considered when comparing across periods with different temperature patterns. Second, higher residential occupancy rates can increase households' response to temperature, making the effect of hot days stronger during the stay-at-home periods than otherwise.

Temperature models to assess sensitivity to load change due to temperature change were created using 2018 and 2019 daily average load data (for counterfactual models) discretely processed with ambient temperatures corresponding to feeder weather station source. As individual household loads are not available, reporting is performed in percent change compared to the counterfactual model used as a baseline for 2020 observed data. Depending on the specific application, temperature data was used as average period temperature or relative to heating or cooling degree days with a customary balance point of 18.3 ◦C (65 ◦F). In average-temperature regression models, the mean static temperature (MST) temperature was used rather than the customary degree day balance point value in calculation. Processing was performed as an average daily load considering daily average temperature (computed average of all periods as opposed to average of minimum and maximum daily observed temperature approach, which is used with degree-day calculation). Hourly models were used for direct comparison of specific, short-term periods. Being more stable, daily models were used in the linear modeling methods used in this report, consistent with similar observations in previous method comparisons [19,27].

For the relationship of energy usage to ambient temperature, piecewise regression corresponding to ASHRAE RP-1050 type linear change-point regression [37,38] was used to determine the 5-parameter models used (corresponding to three segments marked by two change points (CPs)–representing three regressed periods of load versus temperature), see Equation (1):

$$\begin{array}{l} y = m\_1 \mathbf{x}\_1 + b\_1 \text{ (inf. to low CP bound, } \mathbf{t}\_1\text{)},\\ y = b\_2 \text{ (low CP bound, } \mathbf{t}\_1 \text{ to high CP bound, } \mathbf{t}\_2\text{)},\\ y = m\_2 \mathbf{x}\_2 + b\_3 \text{ (high CP bound, } \mathbf{t}\_2 \text{ to inf)},\end{array} \tag{1}$$

where *y* is an average feeder load (kW), *m*<sup>1</sup> is a regressed constant (kW/temp), *x*<sup>1</sup> is a period average temperature value (below the low CP bound), *m*<sup>2</sup> is a regressed constant (kW/temp), *x*<sup>2</sup> is a period average temperature value (above the high CP bound), *b*<sup>1</sup> and *b*<sup>3</sup> are a set of regressed intercept values corresponding to load at the CP bounds (kW) and *b*<sup>2</sup> is average constant load (kW) across temperature range between CP bounds.

This model accounts for temperature effects on energy use for heating and cooling as well as temperature ranges where load is not substantially affected by temperature. Regression analyses were performed using the open-source Energy Charting and Metrics (ECAM) (ECAM v.6.6, Bonneville Power Administration, Portland Oregon, OR, USA) calculation engine for Microsoft Excel (Excel v.14.0 (32-bit), Microsoft Corp., Redmond, WA, USA) [39] with an 80% confidence interval (CI) used for both temperature change point determination and data boundary determination. Testing showed that an 80% CI provided a balance between valid model calculation convergence and data inclusivity for all feeders analyzed. The calculated midpoint temperature between the determined change points corresponds to the MST. As data is analyzed, CI boundaries are similarly passed through calculations to provide error estimation for multi-step calculations. Temperature-based correction was used to normalize the influence of temperature such that all data sets are corrected to a value representing MST on a daily or hourly basis (as previously discussed) and compared. This approach estimates non-temperature-sensitive load. In addition, a reporting-period basis calculation provides estimates of post-period energy difference considering pre-period basis. This calculation used ECAM's internal engine implementing modified ASHRAE Guideline 14, model guidelines [28,40]. Analyses compared 2020 energy use to baseline data in 2018–2019 (for either the 2020 calendar year period or 2020 COVID-19 pandemic subset period) to normalize the impact of temperature between the evaluation and baseline periods in comparison. By removing this substantial factor, this provides a means to assess differences in load due to the changed factors (namely occupancy) during the COVID-19 pandemic period compared to the baseline period.

A separate two-term linear regression (see Equation (2)) was performed to model the impact of temperature on load as a function of heating degree day (HDD) or cooling degree day (CDD) values on a daily basis. Raw calculated values were limited such that values with CDD or HDD values less than 1.1 ◦C (2 ◦F) were removed from the model to reduce the bias from non-temperature related load variance. A CDD or HDD value would be mutually exclusive for a given day. Analyses were performed using a multiple linear regression in Origin Pro 9.0 (Origin Lab Corp., Northampton, MA, USA). Regression results were modeled for impact across an inclusive range of HDD and CDD values for both the baseline and evaluation period and presented as a simple percent difference for change comparing the differences between evaluation and comparison period with the same change in simulated CDD and HDD values:

$$y = m\_1 \mathbf{x}\_1 + m\_2 \mathbf{x}\_2 + b\_\prime \tag{2}$$

where *y* is the total feeder daily energy use (kWh), *m*<sup>1</sup> is a regressed constant (in kWh per HDD), *x*<sup>1</sup> is the HDD (single day) value, *m*<sup>2</sup> is a regressed constant (kWh per CDD), *x*<sup>2</sup> is the CDD (single day) value and *b* is regressed energy independent of HDD or CDD change (kWh).

#### *2.3. Temperature Restricted Load Calculation*

Temperature restriction is an approach used to filter values outside a pre-defined temperature range where limited correlation exists between temperature and elevated energy usage for each hourly temperature value. This method is appropriate when ambient temperatures largely remain near 18.3 ◦C (65 ◦F), which is the conventional degree-day calculation reference value customarily used by the US NWS. In the current analyses, a 4 ◦C range above and below the balance-point temperature was used for the restriction cut-off. Temperature-restricted 2020 evaluation period load data was compared to the combined 2018 and 2019 composite counterfactual model on a monthly or weekly basis considering day-type scope (all days, weekdays, or weekends/weekend days) or illustratively to a 2018 or 2019 single year baseline. Calculations of energy usage change were performed in the same manner as that used in the previously discussed temperature normalization process.

#### **3. Results**

Stay-at-home behavior generally tracks early public directives and provides the framework for interpreting shelter in place (SIP) response and the impact on energy usage. An LA County state of emergency was declared on 3 March while a California-wide state of emergency was declared on 4 March in response to rising regional case numbers. An SIP executive order was initiated in California on 19 March, and modified for provisions for essential workers on 4 May [32]. A follow-up tightening of restrictions followed on 2 July. Estimates of SIP response rates based on smartphone data (reported from early February through early September) show approximate alignment with LA County first wave COVID-19 reported case values (see Figure 1).

**Figure 1.** Comparison of LA County and all of California for shelter-in-place response and COVID-19 diagnosed cases over time. Data sources: California Department of Public Health [41], SafeGraph, Inc. [21]. The SIP Index represents the change (as a difference) in the % of people staying home compared to pre-pandemic baseline. The index ranges from −100% to 100%, where 0 (zero) is no change from a pre-pandemic baseline.

SIP response for the observed period peaked on 12 April [21,41], and decreased through late June. SIP response, measured as stay-at-home rate, is designated as no commuting or transit observed via mobile phone tracking. A pre-pandemic baseline rate of approximately 25% stay-at-home corresponds to a SIP index of 0. On 13 July commerce was restricted during the second case wave. Compared to the initial SIP response and despite the severity of the second wave (July through August), at nearly an order of magnitude higher than the first wave (mid-March through April), the population reaction was weaker, with less than a 5% increase in SIP response as compared to California and LA County at the pandemic onset, with a nearly a 15% decrease comparing the peak of COVID-19 case count during the second wave to that of the first wave. The magnitude increase of successive COVID-19 case peaks for each wave is so substantial that Figure 1 uses a y-axis logarithmic plot scaling to present this. Comparatively, SIP data is presented with a y-axis linear plot scaling. This smartphone based measure of SIP response over time closely resembles other indicators of stay-at-home behavior, such as keyword search histories for topics related to baking and home improvement, providing anecdotal evidence on activities performed by individuals with more available time and resources during the peak SIP period [42,43].

#### *3.1. Unnormalized Load Comparison*

The first set of load analyses use gross energy use data, not normalized for temperature. Energy use for Feeders A, B, C, and D for the pandemic period compared to the comparison period was higher by 10.0% for all days of the week considered together and by 10.4% during weekdays alone (see Table 2).

**Table 2.** Change in energy use for 2020 compared to a comparison baseline for Feeders A-D showing monthly energy use. Values are not normalized for temperature. Positive values indicate higher 2020 energy use compared to the counterfactual model constructed using the 2018–2019 baseline during the comparable monthly period. See Figure S2 for the yearly summary of individual feeders and Figure S3 for a weekly summary chart for individual feeders.


Evaluating temperature differences while considering occupancy differences for the same period helps differentiate the causes of energy use change (see Figure S1 for monthly summarized temperature information for the LA Basin feeders). As shown earlier, stayat-home rates for LA County rose swiftly in late March, peaked in April, reduced but remained high in May and June, and fell to a lower plateau for the rest of the summer. As shown in Figure 2, average temperatures were fairly similar in the 2020 period as in the 2018–2019 comparison period. Energy use was 2.6% higher for the whole month of March, but 8.6% higher for the second half of the month, after the initial SIP order (see Figure 2). Average temperatures were somewhat higher in April (1.8 ◦C, not significant) than in the composite 2018–2019 comparison period.

**Figure 2.** (**a**) Energy usage across all evaluated communities (simple composite average, not temperature normalized) with 2020 observation compared to a 2018–2019 comparison baseline, showing higher energy use. (**b**) Average monthly temperature as measured at weather data source KCQT in downtown Los Angeles observed for 2020 and comparison periods. Energy use strongly follows temperature change in relation to the mean static temperature (MST).

However, during most parts of the day and night temperatures were near the 18.3 ◦C (65 ◦F) nominal balance point, where the load is least impacted by temperature. Temperatures were much higher in May: a weighted average of 20.9% warmer (4.2 ◦C) with an average 2020 temperature above the balance point of 18.3 ◦C, indicating cooling-related energy use as a driver for the increase of 13.4% in average load that month. June 2020 had an average temperature within 1 ◦C of the counterfactual (weighted), but an average of 6.2% increase for 2020 against the counterfactual, suggesting increases in non-temperaturesensitive loads. Summer 2020 had generally reduced stay-at-home rates compared to spring with a substantially cooler July compared to the same period in the counterfactual. During August 2020, an extended warm period mid-month increased the average monthly temperature, which would have otherwise been a month cooler than the comparison monthly period. During this month, yearly record-high energy use in California was recorded. Increased occupancy compared to the comparison period with extended periods of high temperature led to increased energy use during these extreme heat events.

In fall and early winter, October and November both had monthly averages for 2020 within 1 ◦C of the monthly comparison periods but have 18% and 5% respective increases in energy use over the comparison periods for each month. December, with <1 ◦C of the monthly comparison period, despite the high COVID-19 cases had an energy usage increase within 2% as compared to the comparison period.

In general, monthly average load correlates with temperature change, consistent with expected temperature-driven load increases in hotter periods, particularly if higher occupancy rates lead to stronger response to ambient temperature. However, higher energy use in March provides a tell-tale indicator of increased load in these residential neighborhoods due to SIP activity during a period of relatively consistent temperature. By comparison, the overall LADWP NPL decreased during March and April in large part due to a reduction in commercial activities, which use a higher proportion of total energy load than residential customers (see Figure 3, top portion).

#### *3.2. Temperature Normalization*

Temperature normalization compensates for the impact of temperature on energy use, to better estimate the impact of non-temperature sensitive loads. However, as temperatures can vary across larger measured areas that combine residential and commercial loads, use of this technique on highly distributed loads such as NPL can lead to poor correlation (see Figure 3, bottom portion). Correlations between temperature and commercial loads are generally weaker than for residential because commercial buildings tend to have a higher proportion of temperature-insensitive process loads and large scheduled or sensed ventilation loads regardless of ambient temperature.

Residential energy use presented as a total for the evaluated feeders is shown in Figure 4 and Table 3. Total load yearly average difference against the baseline is 3.6% for 2020 for a scope of all days and 5.1% for the pandemic period against the comparison baseline. During the pandemic period, the average increase due to non-temperature sensitive loads is estimated at 5.6% for weekdays and 4.8% for weekend days. During the spring months of March through June, when SIP response was the highest, average total loads for these residential feeders were higher by 5.2% for all days, with a much higher increase for weekdays (6.2%) than for weekends (3.6%). When the 80% CI regression coefficients are evaluated for temperature and normalized for each MST value, a general pattern develops in the 2020 pandemic period of a smaller static temperature range with a higher comparable static load (greater temperature insensitive load proportion) compared to the baseline. Energy use is higher at low temperatures for all 4 feeders for temperatures adjacent to the upper temperature boundary for 2020 weekdays compared to counterfactual model values for weekdays. The nature of the data shows a distribution for 2020 with a large spread and bias to high load shifts in early spring compared to the comparison data considering the same sub-periods of evaluation. With lower temperatures in July 2020 compared to the counterfactual baseline, temperature range under-sampling occurred, resulting in low temperature data biasing the 2020 data. The limited number of days with high average temperatures in July 2020 compared to the baseline period results in variability as low temperature data is substantially influencing average daily the temperature-toload relationship.

**Figure 3.** Monthly net load (NPL) including residential and commercial customers for Los Angeles Department of Water and Power for the pandemic compared to a counterfactual model using a 2018–2019 baseline. (**a**) Presented without temperature correction, and (**b**) presented with normalization to MST against a corresponding monthly counterfactual value, showing 80% CI boundaries in error bars as a result of temperature normalization.

**Figure 4.** Total residential estimated non-temperature sensitive energy change for 2020 compared to a counterfactual using a 2018–2019 baseline, presented on a monthly basis as a simple composite average of feeders.

**Table 3.** Change in energy use for 2020 compared to a 2018–2019 baseline for Feeders A–D showing monthly energy use after temperature normalization. Positive values indicate higher 2020 energy use compared to the counterfactual model values.


ECAM's native engine was used to generate a predictive model of total load change for the entire pandemic period against a counterfactual model of the comparison period (Figure 5). Energy use change reported is consistent with the temperature normalization method and within 2% for all individual feeders across the evaluation period. Results show relatively constant non-temperature load for the COVID-19 pandemic period in 2020

compared to the counterfactual in the 1–5% range considering all days (weekends and weekdays) (see Figure 5).

**Figure 5.** Total estimated non-temperature sensitive energy change during the COVID-19 pandemic period compared to a 2018–2019 comparison baseline for each analyzed feeder in addition to system wide LADWP NPL load. Error bars represent 80% CI bounds propagated.

Comparing change in energy use to median household income for each feeder (Figure 6), a weak trend develops suggesting higher impacts for temperature-insensitive loads for feeders in communities with lower median income. This may be due to disproportionate impact within this population of unemployment or population shift due to the pandemic. The Burbank feeder (Feeder D), while servicing primarily residential buildings, has a business artifact from an auto dealership on the periphery of the feeder territory which caused a small reduction in load early during the early COVID-19 pandemic period in 2020 compared to the counterfactual baseline.

**Figure 6.** Energy change compared to feeder service community median income with a consistent 5% shown in the vertical error bars and the 80% CI shown in the horizontal error bars. The analysis scope was the COVID-19 pandemic period of Mid-March through December.

Estimation of energy use as a function of heating and cooling use change showed modest changes in the impact of load as a function of average HDD and CDD compared to the counterfactual period considering only the COVID-19 pandemic period as well as all of 2020 considering weekends and weekdays separately or combined (Figure 7).

**Figure 7.** *Cont*.

**Figure 7.** (**a**–**d**) Modeled normalized load change for 2020 compared to the baseline period for both calendar year periods (2020 to a 2018–2019 baseline) and subsets of mid-March through December for all periods comparing change in load relative to the baseline for a range of CDD and HDD values for each of the four feeders (**a**) Feeder A, (**b**) Feeder B, (**c**) Feeder C, (**d**) Feeder D. See Figure S2/Table S3 for a similar presentation of this data using normalized MST values and average daily temperatures as opposed to HDD and CDD values.

The HDD impact from heating loads decreased in all cases as presented. As noted earlier, 2020 was warmer in early spring leading to potential model bias during the period where SIP would have had the greatest impact on energy use. Electric heating (primarily portable space heaters) is a minor heat source in the region, with natural gas heating being predominant. Another region with higher heating requirements may provide better data for impact analysis. As expected, cooling loads for most scopes increase as temperatures rise from moderate to high, but plateau at very high temperatures, after air conditioning use is saturated. With this said, high heat events did distinctly show an increase in load for a given CDD value; this is especially apparent in the feeders in the LA Basin. For the Burbank feeder, a leveling off of increasing load is observed as the result of limited reserve cooling capacity–all available cooling having already been activated and in use (see Figure 8). Per Chen et al., warmer areas in Southern California, such as the San Fernando Valley, are less temperature sensitive compared to cooler areas. The current results suggest this phenomenon similarly carries over to a more limited change in energy use during extreme heat events during the COVID-19 pandemic period as compared to other more temperature sensitive areas.

**Figure 8.** *Cont*.

**Figure 8.** Comparison of feeder daily energy use for 2018, 2019, and 2020 for observed CDDs for (**a**) the month of July for Feeder A; and (**b**–**e**) presenting the month of August for Feeders A–D.

#### *3.3. Temperature Restriction*

Estimation of non-temperature sensitive loads on an hourly basis provides indication for granular energy use change based on changes in behavioral patterns that can only be observed at an hourly (versus a daily) level. Removing heating and cooling loads by restricting points when these loads are likely active reduces the temperature variability and helps present impact due to behavior change during SIP and the impact on nontemperature sensitive loads. Mid-day energy use is increased on weekdays (Figure 9) for most feeders. Weekend data is typically noisier than weekday data given relative under sampling compared to weekdays. Early evening peaks are moderately higher and weekday morning peaks are reduced. The values found via this direct analysis (Table 4) are largely similar to the estimated change due to non-temperature sensitive loads (Table 3).

**Figure 9.** Baseline peak normalized (separately for weekdays and weekends) energy use change comparing 2020 to 2019 baseline for mid-March through April using a temperature restriction. Examples presented for (**a**) Feeder A and (**b**) Feeder B within the LA Basin microclimatic region.


**Table 4.** Energy use difference for mid-March through April comparing 2020 against a counterfactual baseline of 2018–2019. The all-day average for Feeders A–D was 5.3%.

#### **4. Discussion**

While major fuel and energy sources were observed to show a net decrease in use early in the pandemic, the opposite was largely observed for residential energy use. These findings were consistent with that of earlier studies such as those performed by Pecan Street [14] in Austin, TX, with 113 panel-instrumented homes: study results showed an approximate 42% (~300 W) mid-day increase in April 2020 for non-temperature sensitive loads such as consumer electronics, appliances, miscellaneous electric loads (plug loads), and lighting, compared to a baseline of the previous year, reflecting increased occupancy with increased load during both weekdays and weekends. Full-day energy use increase is likely closer to ~14%, estimating from Pecan Street provided figures. Similarly, this Pecan Street study identified an increase in temperature sensitivity across March and April identified by average home kWh/cooling degree day (CDD) of the evaluated period with a value of 0.7 in April 2020 compared to a value of 0.56 for the average of April 2017, April 2018, and April 2019, a comparative 25% increase in load for each CDD change [14]. These results match the general trends observed in our study, albeit with higher magnitude changes between 2020 observations and past baselines. Much of this difference is likely related to Pecan Street's use of instrumented single-family, higher-income housing combined with regional climatic variance (e.g., impact of humidity and higher regional temperatures on cooling behaviors). Also, days with potential heating and cooling activity in shoulder periods (often with low CDD or HDD values) can incur bias from the dominant space conditioning energy load used during the period, as previously mentioned. Energy use for this scenario can increase for low HDD or CDD values; our tests showed that using a threshold value of 2 CDD or HDD substantially reduced this impact. The temperature in Los Angeles in April rarely requires air conditioning usage, whereas Austin, Texas experienced a warm and humid spring during the highest SIP period.

Load impact from non-temperature-sensitive loads during the early pandemic were estimated from sampled feeders through both temperature restriction (Table 4) and temperature normalization (Figure 4) resulting in estimated increases of 5.3% and 5.7%, respectively (mid-Mar through April, all days), less than that reported by Pecan Street. With the exception of Feeder C, change in weekday load was more impacted than weekend load compared to the 2018–2019 baseline during the early pandemic (Table 4). Non-temperature loads were a substantial component of energy used which is evidenced by the similarity in total load change (Table 3) to temperature restricted load change (Table 4). Heavy mixtures of both HDD and CDD during this period complicate regression analyses (of the type used in Figure 7). This is because the nature of the degree day metric is not exclusive to heating or cooling, but is the balance point difference computed between the range from daily highs and lows. When temperature fluctuates enough over a 24-h period to require both heating and cooling, that day may be labeled with a low value for HDD, CDD, or both. This effectively skews energy use per HDD or CDD when using multiple regression models. Temperature normalization based on average daily temperature regression performs marginally better with respect to these temperature variations.

When temperatures increase, increased occupancy (even at lower levels compared to the mid-April peak) drives loads higher. This is clearly illustrated in Figure 8b representing Feeder A. The load events with CDD values between 9 ◦C and 11 ◦C required 9.2% more load compared to similar events in this same temperature range in 2018, consistent with the idea of higher home occupancy rates driving higher demand for cooling on hot days. The effect of SIP response can differ for weekend and weekday loads. This is illustrated with the highest heat day in this figure, which has substantially less load than the second highest load event: note that this day falls on a weekend (for which occupancy shifts due to SIP should be reduced) versus higher impacts on adjacent weekdays during this extended extreme heat event. Mixing weekdays and weekends for analysis results in model variance challenges due to substantially different activities for these two day types. This is especially true during typical, non-SIP periods such as the baseline. Clearly, increased occupancy drives up cooling requirements during extreme heat events. Capturing a representative spectrum of temperatures and loads for each month while occupancy was varying due to SIP response to allow direct calculation is challenging. For example, as illustrated in Figure 8a, the high heat events observed in July 2018 and 2019 were not replicated in July 2020, which weakens any comparison across these months to assess 2020 SIP response effects on energy use.

Daily energy use patterns were strongly impacted early in the pandemic. Compared to the counterfactual model, energy use was slower to rise in the early morning and was higher during mid-day hours, with a moderate increase in daily peak energy use across all feeders (see Figure 9). Assessed with restricted temperature analysis, the impact of these features decreased with a slow resumption toward baseline energy use as SIP response reduced.

Energy usage impacts for large multi-family apartment complexes is likely different from that for the single family and small multi-family residences studied above. Figure 10 shows results for an additional example, Feeder E, representing a large apartment complex. During the early pandemic period, energy use for this case largely tracked other residential loads. By summer, the shutdown of many shared-use areas within these buildings to reduce potential community spread of COVID-19 reduced the cooling burden to these buildings, resulting in a net drop compared to the baseline during the period when the cooling burden is the highest (mid-summer). This effect, plus the centralization of cooling and heating, are likely substantial divergence points comparing large apartment complexes and high-rises to low- and medium-density homes and low-rise apartments, which have limited shared facilities and individual heating and cooling supplies.

Commercial energy use, a major component represented in the NPL figure, is illustrated by a single mid-rise building source (see Feeder F in Figure 10). This example is included as a contrast to the residential feeders analyzed above, as an approximate indicator for impacts of SIP on non-essential business activity (jewelry manufacture and distribution). For this commercial feeder, a major drop in energy use occurred during week 12 of 2020 (16–22 March), corresponding to the initiation of SIP restrictions, which is when residential energy use increased. By mid-June (Week 21) energy use in the commercial feeder had greatly increased. This follows a weakening of SIP response, previously discussed. The second-wave restrictions did not substantially reverse the increase in energy use, which showed continued growth until early fall. The lower energy use in November and December of 2020 compared to the 2018–2019 counterfactual composite baseline may reflect the reduction of typical high-intensity holiday shopping during those months, including extended hours.

**Figure 10.** Direct change in energy use (non-normalized for temperature) for Feeder E (large apartment complex) and Feeder F (commercial building).

The current findings show limited evidence of a higher increase in non-temperature sensitive load over the COVID-19 period for lower income areas than higher income areas. The expected effect of SIP response on energy by income is not clear, as various factors predict mixed results. For instance, more highly educated, higher-income professionals were more likely to be able to shift to working from home, while less-educated workers were more likely to either continue working outside the home (e.g., in essential service or manufacturing) or lose their jobs. Lower-income households tend to have more members to use devices if everyone is at home, but higher-income households have more square footage and more devices to be used per person; furthermore, lower-income households spend less money (and time) on entertainment and dining outside the home than higher-income households normally, and would thus experience less change. Residential use of portable space heaters and window AC units, more common among older housing stock in lowerincome areas, also adds to electricity use. Given the limited number of neighborhoods sampled here and the small observed effect, this result is considered questionable, but suggestive of further consideration; additional research would be required to clearly ascertain the income differences in effect of SIP response on energy use. However, it is worth noting that even the same or lower increase in energy use is a greater hardship for lower-income households, as they already experience a significantly higher energy burden (that is, the proportion of their income spent on energy bills) and have little or no discretionary income to cover unexpected expenses.

Overall, SIP compliance was initially strong, but this effect was temporary. Approximately one month elapsed between the rapid ramp-up of SIP response and a long-term decrease and eventually leveling out of SIP compliance. This occurred even with daily briefings from health experts and government officials reporting increasing caseloads in the LA area. Energy models considering change in occupancy must expect a ramp-up, peak, and an extended dynamic equilibrium for general change in occupancy. Considering the near future of the COVID-19 timeline, stay-at-home rates will continue to subside into

an extended equilibrium that is likely higher than pre-pandemic levels. This suggests that increased telecommuting from home will continue to raise the energy burden during high heat events. Mid-day energy use, compared to a pre-COVID-19 baseline has had a modest increase–this can help offset the increasing glut in solar energy mid-day during normal conditions. However peak conditions, especially in late afternoon when solar is switching to spinning reserves can still impact energy supplies during this critical ramp-up and source switching period.

#### **5. Conclusions**

This research adds to the growing body of knowledge on how the COVID-19 pandemic has affected human behavior and the resulting impact on energy usage. Increased residential occupancy has impact on energy use. Over the course of the 2020 pandemic period, fatigue with SIP compliance led to a rebound toward earlier pre-pandemic occupancy rates (reduced SIP response) and a substantial rise in regional COVID-19 active cases. It is reasonable to assume that in future pandemic events, similar behaviors are to be expected. The potential for extended SIP activity for extended periods has limits. The timing of an SIP period can strongly affect energy use change. During temperate periods, limited heating or cooling impact will likely be observed with a constant increased non-temperature sensitive load increase. Even with occupancy patterns trending more toward normal, impacts on energy used for cooling during heat events was observed. As the current analysis examined only electricity, and space heating in this region is largely fueled by natural gas, observed stay-at-home impacts on heating were minimal. However, as electrification development continues, increased reliance on electric heating should be reflected in larger impacts of residential occupancy on electrical energy use. As long-term work at home activity continues, increased residential energy use during weekdays will continue for applicable households. Modeling this change is outside the scope of this study but relevant to future expected household energy change and population impacts. The results suggest the possibility of a higher impact of stay-at-home behavior on energy change for communities with lower median income level, however, evidence is weak and further research would be necessary to confirm such a relationship.

Continued efficiency measures for miscellaneous electric loads can help reduce nontemperature sensitive loads. Focus on reducing wasteful energy use (i.e., devices not properly entering low-power mode when not in use) is a major potential area of research. The analysis this study has provided on residences is also applicable to businesses, to highlight opportunities for better managing plug and process loads, especially while not in use, and may be a fruitful area for follow-up study. Follow-up studies using similar approach methodology with data for areas with substantial heating and cooling loads would help draw the maximum impact of stay-at-home behaviors when considering temperature sensitive loads as a major energy load contributor.

**Supplementary Materials:** The following are available online at https://www.mdpi.com/article/ 10.3390/app11104476/s1, Numeric tools, calculation scripts, and calculation workbooks as well as extended data subsets used within this report are available at: https://github.com/CalPlug/ CovidResEnergyAnalysis2021. Supplementary Report Contents: Table S1: Extended feeder summary table showing solar and EV installation penetration within the bounding ZCTA for each feeder, Table S2: Change in energy use for 2020 compared to a counterfactual 2018–2019 comparison baseline showing change in energy usage by communities, Figure S1: Monthly temperature summary for Los Angeles across all periods used for analysis, Table S3: Summary of regression segments (80% CI) and temperature change point values for both comparison and pandemic evaluation periods. Table S4: Specific dates for the ISO weeks for 2018, 2019, and 2020. Figure S1: Monthly temperature summary for Los Angeles across all periods used for analysis. Data sourced from KCQT (Downtown LA), NWS. Figure S2: Comparison model of the effect of ambient temperature on feeder energy use—MST normalized to 100%. Figure S3: Non-temperature normalized energy use for all feeders presented on a weekly basis. Table S5: Report abbreviation and acronym list.

**Author Contributions:** Conceptualization, G.-P.L., M.J.K. and A.T.; methodology, M.J.K., J.E.P. and A.S.; software, M.J.K.; validation, M.J.K. and J.E.P.; formal analysis, M.J.K. and J.E.P.; inves-tigation, M.J.K. and J.E.P.; resources, A.T., G.-P.L. and D.J.; data curation, M.J.K., A.S. and J.E.P.; writing original draft preparation, M.J.K., A.S. and J.E.P.; writing—review M.J.K. and J.E.P.; visualization, M.J.K.; supervision, M.J.K.; project administration, G.-P.L., A.T. and D.J. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Institutional Review Board Statement:** Not applicable.

**Data Availability Statement:** Not applicable.

**Acknowledgments:** The authors wish to thank Miguel Malabanan, Mehdi Shafaghi, Jeremiah Valera, and Luke Sun from LADWP for technical support and Sabine Kunrath, Mahejabeen Kauser, and Katie Gladych from CalPlug for their support with background research, data preparation, and manuscript formatting/copy-editing assistance.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


### *Article* **Solar Energy in Urban Planning: Lesson Learned and Recommendations from Six Italian Case Studies**

**Matteo Formolli 1, Silvia Croce 2,3, Daniele Vettorato 2, Rossana Paparella 3, Alessandra Scognamiglio 4, Andrea Giovanni Mainini <sup>5</sup> and Gabriele Lobaccaro 6,\***


**Abstract:** This paper presents the results of the analysis conducted on six case studies related to solar energy integration in urban and rural environments located on the Italian territory. The analysis has been carried out within the Subtask C—Case Studies and Action Research of the International Energy Agency Solar Heating and Cooling Program Task 51 "Solar Energy in Urban Planning". Three different environments hosting active and passive solar energy systems (existing urban areas, new urban areas, and agricultural/rural areas) have been investigated to attain lessons learned and recommendations. Findings suggest that (a) it is important to consider solar energy from the early stages of the design process onwards to achieve satisfactory levels of integration; (b) a higher level of awareness regarding solar potential at the beginning of a project permits acting on its morphology, achieving the best solution in terms of active and passive solar gains; (c) when properly designed, photovoltaic systems can act as characterizing elements and as a distinctive architectural material that is able to valorize the aesthetic of the entire urban intervention; (d) further significant outcomes include the importance of supporting the decision strategies with quantitative and qualitative analyses, the institution of coordinating bodies to facilitate the discussion between stakeholders, and the need for deep renovation projects to fully impact existing buildings' stock; (e) when large solar installations are planned at the ground level, a landscape design approach should be chosen, while the ecological impact should be reduced by carefully planning the adoption of alternative solutions (e.g., agrivoltaics) compatible with the existing land use.

**Keywords:** solar energy; urban planning; active solar systems; passive solar strategies; Italian case studies; agrivoltaics

#### **1. Introduction**

Cities represent the place where 80% of the gross domestic product is generated and where around half of the world population lives, with expected growth to two-thirds by the middle of the current century [1]. The combination of these two factors demonstrates why urban areas are considered responsible for up to 75% of the global energy consumption and more than 70% of energy-related carbon dioxide (CO2) emissions [2].

**Citation:** Formolli, M.; Croce, S.; Vettorato, D.; Paparella, R.; Scognamiglio, A.; Mainini, A.G.; Lobaccaro, G. Solar Energy in Urban Planning: Lesson Learned and Recommendations from Six Italian Case Studies. *Appl. Sci.* **2022**, *12*, 2950. https://doi.org/10.3390/ app12062950

Academic Editor: Salvatore Vasta

Received: 13 February 2022 Accepted: 9 March 2022 Published: 14 March 2022

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2022 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

In this context, the decarbonization of the energy sector, in line with the Paris Agreement's goals to keep the temperature increase below 2 ◦C, requires a substantial contribution from cities. A growing number of cities, mainly located in the European Union (EU) and North America, have set renewable energy targets to support the green energy transition [3]. At the municipal level, the adoption of regulations, economic incentives, and the direct financial involvement of municipalities and public authorities, combined with the storytelling of successful practices, proved to be crucial in raising public awareness and in steering private developers toward sustainable strategies [4,5]. Similarly, in the past decades, national level measures constituted a strong enabler for the exploitation of renewable energy sources (RES) [6].

Solar photovoltaic (PV) and solar thermal (ST) had respectively the first (36%) and the fourth (10.5%) highest expansion rate among the RES in the past 30 years [7]. In one of the preconfigured future scenarios, the International Energy Agency (IEA) is expecting that PVs would cover almost one-third of the new electricity demand within 2030, with an average annual growth rate of 13% [8]. Utility-scale ground-mounted systems dominate the market [9]; however, several potential issues related to their impact on the ecosystem, landscape, and competition with agricultural land are emerging [10,11].

The major alternative is distributed rooftop systems (i.e., applied and integrated systems), representing a market share of 40% in 2020 [12]. This technology has the highest potential use in the built environment, especially for photovoltaic systems fully integrated into the building envelopes (i.e., facades and roofs), which perform the double function of energy producer and building cladding. The advantages of building-integrated photovoltaics (BIPV) and solar thermal (BIST) are manifold: unused surfaces can be turned into active energy generators [13], losses associated with transmission and distribution of electricity are reduced thanks to on-site production [14], and higher energy flexibility toward extreme weather conditions is guaranteed [3]. On the other hand, the installation of such systems in urban areas, especially on vertical surfaces (i.e., facades), requires considering complex phenomena such as inter-building solar reflections and overshadowing effects [15–17], and other energy and climate-related issues, such as high surface temperature [18] and fire hazard [19]. The visibility and the socio-cultural sensitivity impact of the systems should also be considered [20–23]. In Europe, BIPV accounted for a cumulative installed capacity of 6.9 GWp at the end of 2019, with Italy covering about 38% of the total, due to the early campaign of statal incentives to boost the adoption of PV technology [24].

Other two viable alternatives, representing small market niches in rapid expansion, are floating solar and agrivoltaics [12]. These technologies allow to reduce land use for solar installations and through appropriate design can provide a series of benefits. Floating solar has the advantage of reducing the evaporation of water reservoirs. Furthermore, the presence of water helps to keep the panels' temperature low, stabilizing the efficiency of the system [25,26]. Similarly, agricultural PV can improve crop yield, limit evaporation, provide shade to livestock or crops, and protect against extreme weather conditions and soil erosion [12,27]. The potential for agrivoltaic systems in Europe is immense; if solar would be deployed on 1% of the arable land, its technical capacity would amount to over 900 GW, which is more than six times the current installed PV capacity in the EU [28].

#### *Framework and Aim of the Work*

This paper presents part of the findings related to Italian case studies within the *Subtask C—Case Studies and Action Research* (STC) framed in the International Energy Agency (IEA) Solar Heating and Cooling Program (SHC) Task 51 "*Solar Energy in Urban Planning*" [29–31]. During the whole duration (2013–2017), the international experts working in Task 51 (i.e., architects, urban planners, public authorities, researchers, etc.) promoted through their work the integration of active and passive solar solutions in the built environment. Furthermore, the collection of international case studies through a common template carried out within STC allowed to create an overview of solar energy in urban planning by highlighting potentialities and fragilities and by summarizing lessons learned for urban stakeholders, public authorities, and researchers. The developed template is here introduced and used to present six Italian case studies, which make use of solar energy in different ways and are characterized by different built environments. Five of them deal with consolidated and new urban areas, with photovoltaic or solar thermal panels fully integrated or applied to the buildings' envelopes, while one case provides an example of an agricultural PV system and its relationship with the landscape. Different from [31] where an international overview is given, in this paper, the Italian case studies are presented more in detail, and specific conclusions are drawn for solar energy implementation in the Italian context. Additionally, a comparison between the different built environments through similarities and differences is provided.

In the next section, an overview of the Italian energy planning legislation is given, followed by the presentation of the case studies and their discussion. Finally, the limitations of the study are presented, together with a conclusive summary of the main lessons learned and recommendations.

#### **2. Background: Italian Legislative Framework**

The framework of Italian legislation on urban and energy planning is characterized by a hierarchical approach stretching from the national level down to the regional, provincial, and municipal levels. According to the Constitution, urban planning and energy-related topics are a shared task between the State, the Regions, and the autonomous Provinces. Consequently, regional authorities may implement autonomous legislations as long as they do not contradict the general principles and requirements provided by national and EU regulations.

The Italian national legislation promoting energy efficiency and the diffusion of RES has been developed as an implementation of the major European directives. In 2011, the National Renewable Energy Action Plan [32] transposed the Directive 2009/28/EC setting the targets for renewable energy production by 2020. The plan also defined a minimum quota of production from RES for all new buildings and buildings subjected to major renovation. The minimum share of RES quota for domestic hot water (DHW) was set to 50% of the primary energy consumption. Furthermore, a calendar with a progressive higher RES share was established for the sum of primary energy consumptions for DHW, heating, and cooling. In December 2018, the EU targets have been revised by the European Directive 2018/2001/EC on renewable energy [33] and by the Regulation 2018/1999/EU on the governance of the energy union and climate action [34]. These regulations were part of the "Clean Energy for all Europeans Package", which promoted a 40% reduction in greenhouse gas (GHG) emissions, a goal of 32% final consumption from RES, and an energy efficiency target of 32.5%. Finally, at the beginning of 2020, the European Parliament adopted the European Green Deal, which set the objective "to increase the EU's GHG reductions target for 2030 to at least 50% and towards 55% compared with 1990 levels in a responsible way" and to achieve climate neutrality for the continent by 2050 [35]. Building on the EU legislation, and in line with the European Regulation 2018/1999/EU, which required all EU countries to develop 10-year National Energy and Climate Plans (NECPs) for the period 2021–2030, the Italian Integrated National Plan for Energy and Climate (INECP) was adopted in January 2020. The INECP set clear targets to 2030: (i) 30% of energy from RES in the final gross energy consumption, (ii) 43% reduction in primary energy consumption compared to the Price-Induced Market Equilibrium System (PRIMES) 2007 scenario, and (iii) 38% overall reduction in GHG emissions compared to 1990 [36].

#### *2.1. National Standards*

The INECP sets some growth targets for power and thermal energy from RES at the national level. In the case of solar energy, targets are set to 28,550 MW for 2025 and 52,000 MW for 2030 (in 2017, the production amounted to 19,682 MW).

With regard to the thermal sector, the targets are 590 ktoe in 2025 and 751 ktoe in 2030 [36]. Furthermore, the "Clean Energy for all Europeans Package" was implemented in Italy by the "Renewables Decree" (D.Lgs 28/2011 [32]), which also led to the definition of guidelines for energy performance certification of buildings (D.M. 26 June 2015. Specifically, Article 11 of Decree D.Lgs 28/2011 introduces the obligation to integrate renewable energy sources in new buildings or buildings subject to major renovations. Indeed, in case of new construction, refurbishment, or demolition and reconstruction, RES should cover 50% of DHW consumption, and 50% of the sum of consumptions for DHW, space heating, and cooling. For public buildings, the share is increased to 55%; while in the case of private buildings located in historic city centers, the share is reduced to 25% (27.5% for public buildings) [32]. In 2021, the share of RES has been further increased, with the introduction of the D.Lgs 199/2021. From June 2022, in case of new construction and major renovations, RES should now cover 60% of the consumptions for DHW, space heating, and cooling. Conversely, a 65% coverage should be demonstrated for public buildings [37]. Concerning passive solar, minimum daylight levels in buildings are regulated by national laws (D.M. 190/1975, C.M. LL. PP. 22 November 1974 n. 1301 and D.M. n. 26/1975). These set minimum levels of the daylight factor depending on the type of building:


#### The Italian National Status of Ground-Mounted PV

The national energy target for 2030 foresees an increase in ground-mounted photovoltaics of about 35 GW. However, the actual installation trend of 1 GW/yr is far below the 6.5 GW/yr needed to fulfill the target on time [38]. The installation of PVs on the ground is currently hampered by barriers along the authorization process due to landscape preservation and land use concerns. The adoption of innovative solutions such as agricultural PV aims at overcoming these barriers by combining a dual use of land and establishing a synergy between energy production and agriculture, which can be seen as a driver. In that regard, the current legislation is in constant evolution. In April 2021, the National Recovery and Resilience Plan (RRP), part of the Next Generation EU (NGEU) program, allocated 1.1 billion euros investment for the "agrivoltaic development" with the specific objective of installing 2 GW of agrivoltaic capacity and improving the competitiveness of the agricultural sector [39]. Additional elements regarding the definition of agrivoltaics have been introduced by the National Law 29 July 2021 [40] where these systems are described as "innovative integrative solutions, with PV modules raised from the ground, also including tracking systems, which do not compromise the continuity of the agricultural activities on ground". This law introduces simplifications in the process, which would lead to a rationalization of the authorizations for PV toward the national energy targets. Despite a specific definition for agrivoltaics not being currently available in Italy, an official document is expected in a short time. Meanwhile, the National Agency for New Technologies, Energy and Sustainable Economic Development (ENEA) has launched the National Network for Sustainable Agrivoltaics [41], while the Italian Electrotechnical Committee (CEI) has initiated a working group on the topic.

#### *2.2. Municipal Standards*

The targets set at the national level are pursued by Italian Regions, which can enhance the minimum quota or specify the renewable energy systems to use. Therefore, depending on regional laws, municipalities can set their own energy-related regulations in the building code. Generally, the building codes consider two major categories of built environment: historical city centers and non-historical urban areas [42].

Historical city centers are subjected to strict regulations about solar system applications due to the need to preserve the visual effect of their views and panoramas. Concerning this category, buildings codes set either prohibition of installing solar systems or restrictions on the visual effect and location of the systems (e.g., modules have to be located on roof surfaces not visible from main streets and to be building integrated). In addition to these restrictions, the approval of the Superintendence for Architectural Heritage and Landscape is required for installing solar systems on constructions subjected to landscape or heritage protection [43]. In the second category—non-historical urban areas—solar systems are generally permitted as building-integrated. For new buildings or buildings subjected to major renovations, it is required that PV or ST plants are installed integrated or adherent to the roofs, following their same orientation and inclination.

Further regulations on solar energy systems are sometimes included in municipal building codes, deriving from national laws on building energy efficiency and the promotion of renewable energy production. However, the standards regard only the design of buildings at the architectural scale. Energy performance certification is mandatory for new constructions and for selling or renting a dwelling in existing buildings [44].

#### **3. Materials and Methods**

This section illustrates the methodology used to analyze the case studies. The method was developed within the *Subtask C*—*Case Studies and Action Research* of the IEA SHC Task 51 to provide experts in the field of urban planning with an exhaustive set of information and examples on the integration of solar energy in heterogeneous environments. A total of six cases located on the Italian territory are presented here.

#### *3.1. Classification of the Environments*

As a first step, the case studies were divided according to the three types of environments presented in Table 1.


**Table 1.** Classification of the environments.

#### *3.2. Template Definition and Description*

The second step consisted of the creation of a template to systematically organize the information of the case studies within a homogeneous framework. Despite the adoption of a standard structure, the template has a certain degree of flexibility. A total of ten sections are available, six of which are common among all the investigated cases, while the others can be compiled according to the type of environment described and the available information. A schematic representation of the template structure is visible in Figure 1, followed by the sections' description.

**Figure 1.** Representation of the different sections of the template used. In full color, the sections compiled for all the cases, while the optional ones are shown in white.


**Figure 2.** (**a**) Architectural integration quality matrix; (**b**) Criticality matrix developed in relation to (**c**) context sensitivity and (**d**) system visibility.

6. *Solar landscape*. This section is specific to the landscape environment. It includes the definition of the solar system used, its functional features (i.e., modules' pattern, presence of edges—Figure 3), and the energy production. This is defined according to the classification proposed by ENEA on solar landscape plan [10].

**Figure 3.** The spatial system as a whole (pattern) (on the **left**), the photovoltaic space (in the **middle**), and the "pore" space (on the **right**).


An overview of the six analyzed case studies is visible in Table 2.


**Table 2.** Overview of the analyzed case studies. (CCHP: Combined Heating and Cooling; DH: District Heating; DC: District

 Cooling).


**Table 2.** *Cont.*

#### **4. Results and Discussion**

In this section, the six case studies are presented and discussed, starting from the ones in existing urban areas, followed by new urban areas, and concluding with the integration of solar technologies into the landscape.

#### *4.1. Photovoltaic Village in Alessandria*

#### 4.1.1. Overview

The case study is a residential neighborhood constituted of several multistorey buildings, for a total of 192 flats, located in the southwest fringe of Alessandria (Piemonte). The project was developed within the context of a complex initiative undertaken by the Alessandria Municipal Council after the flood of 1994. It aims to regenerate a social housing community using a sustainable, environmental, and social approach through the realization of new buildings, urban facilities (i.e., a community center, recreational areas, sheltered sits, and parking lots) and the refurbishment of existing residential blocks (Figure 4) [48]. The extensive use of photovoltaic technology characterizes the project.

**Figure 4.** The (**a**) aerial view of new buildings in the eastern part of the Photovoltaic Village (source: © Municipality of Alessandria) and a site plan of the entire area indicating the position of facilities, new and existing buildings (**b**) (source: © Municipality of Alessandria).

#### 4.1.2. Challenges, Issues, and Decision Strategies

The challenge encountered during the realization of the project was the definition of an innovative planning process to meet the interests of both public and private actors and the promotion of energy-efficient solutions and RES. In addition, the existing urban plan was substituted due to the adoption of new legislative procedures. The decision strategy was based on three principles: (i) active participation and collaboration between the actors; (ii) a pivotal role of the municipal authorities in the whole design process; (iii) the project conception as a pilot for town planning and suburban requalification.

#### 4.1.3. The Planning Process

The project began in 1996 with the enforcement at the regional level of the Environmental, Building and Urban Requalification Program [49], with energy saving as one of the main objectives for buildings and urban projects. This stimulated the adoption of solar energy and sustainable practices and led to create the Building Operators Council in 1997 to coordinate the actors involved in the process. The construction phase lasted five years (2000–2005). The most important phases were the large scale of comprehensive/strategic planning, represented by the Environmental, Building and Urban Requalification Program, and the architectural design stage of the existing buildings. The Municipal Town Council, the Building Operators Council constituted by private and cooperative builders from the Province of Alessandria, architects, and various other private stakeholders took part in developing this initiative. Furthermore, researchers from Politecnico di Torino contributed to the design and monitoring of solar systems [48].

#### 4.1.4. Energy Concept

The buildings are designed to reduce their impact on the environment through energysaving measures, and the use of RES. PV modules, with an efficiency of up to 15%, are installed on roofs with a tilt angle of 30◦, on the southern facades in front of the stairways of two buildings in the community center (indicated with letters B and C in Figure 4b) and as part of a photovoltaic pergola for the public space (Figure 5). PV systems cover a total area of 3000 m<sup>2</sup> (i.e., 1600 m2 panel net surface). The overall power is 163 kW with estimated energy production of 674–830 kWh/kWp per year. This is expected to cover 100% of the electricity consumption for common areas and up to 70% for the flats [48].

**Figure 5.** Ground (**a**) and aerial (**b**) view of the pergola equipped with PV panels in the public area (source: © Municipality of Alessandria).

#### 4.1.5. Architecture, Visibility, Sensitivity, and Quality

The LESO–QSV method illustrated in the methodology is used in this section to evaluate the integration quality of the installed solar systems. As visible from the colored segments of the ring of the matrix in Figure 6, the integration has been judged partly coherent for all the three investigated aspects. Since the modules do not perform the double function of replacing a building component and simultaneously producing energy, being just overlaid to the roofs and facades, they cannot be considered as fully integrated into the building envelope from an architectural point of view. Moreover, the blue color of the solar cells contrasts with the finishing color of the facades. The systems' visibility from a close point of view is high, especially for the facades' elements, and it remains medium from a remote distance. Nevertheless, the context can be classified as of medium sensitivity due to the low historical value and the absence of any relevant elements or monuments.

**Figure 6.** Architectural integration matrix for the PV systems in the Photovoltaic Village of Alessandria (left); (**a**) PV panels installed on facade in front of the stairways (source: © PierFranco Robotti); (**b**) aerial view of the systems installed on the flat roofs (Source: © Municipality of Alessandria).

#### 4.1.6. Environmental, Economic, and Social Impacts

The project follows eco-sustainability and bio-compatibility principles. Particular attention was reserved to open spaces where green areas, a water pond, and urban furniture were carefully designed to create different bioclimatic zones for the users to stroll, relax, and socialize. The project is financed by a mix of public and private sources. The region financed part of the investments for its environmental sustainability implications, while approximately 70% of the total cost for the installation of photovoltaics was covered by public funds through the program "10,000 Photovoltaic Roofs" coordinated by the Italian Ministry of the Environment [48,50].

#### 4.1.7. Approaches, Methods, and Tools

A monitoring campaign on the PV systems was carried out by researchers of Politecnico di Torino for a period of 12 months (September 2004–August 2005) to measure the electricity generated during the buildings' operation phase and correlate it to the actual energy consumption. Data were collected every 15 min and function curves for daily, weekly, monthly, and annual periods were generated. The analysis shows that PV systems can fulfill the electricity demand for seven months, but an electricity supply through the grid is still needed. In addition, in the absence of regional, national, or community incentives, the investment can be paid off in a long-term period (more than 30 years).

#### 4.1.8. Lesson Learned and Recommendations

The Photovoltaic Village in Alessandria illustrates the limits of retrofitting existing buildings with solar technologies in the absence of a deep renovation opportunity. Despite the overall positive results, the designers were forced to consider a limited range of design options since they had to comply with existing buildings' masses and materials. A different outcome may have been obtained through an in-depth renovation process involving replacing/recladding facades and roofs. Furthermore, the integration of solar systems in public spaces has only been marginally explored. However, the institution of a Building Operators Council as a coordinating body for the different involved actors was a successful practice replicable in similar contexts.

#### *4.2. SINFONIA Bolzano*

#### 4.2.1. Overview

The SINFONIA case study is the result of a five-year European project aiming to implement large, scalable, and integrated energy solutions in mid-sized cities. Bolzano, the capital of South-Tyrol province (Trentino Alto Adige), was selected together with Innsbruck (Austria) as a testbed for this initiative to understand the replicability of the proposed solutions in other European cities [51]. The project aimed to achieve the following:


In this framework, refurbishment interventions were undertaken for six residential complexes in Bolzano to improve their energy performance and indoor comfort. The initiative is in line with the municipal goal to reduce energy use and increase the share of RES. The presented case study focuses on one of these interventions, consisting of two buildings blocks of 36 flats each, located in the south part of the town (Figure 7).

**Figure 7.** North building before (**a**) and after (**b**) the intervention (source: © Ivo Corrà).

#### 4.2.2. Challenges, Issues, and Decision Strategies

The location of the building complex has been one of the main challenges in the development of the project. Its proximity to the side of a mountain represented an accessibility issue during the refurbishment phase. Other challenges included the limited surface for the integration of solar energy systems, which was also due to the proximity to the mountain slope toward the south and the presence of overshadowing due to four stairwell towers above the roof levels. Finally, the intervention should have the minimum possible impact on the tenants, as all refurbishment activities have been carried out with lived-in flats. Regarding the decision strategy, targets for the energy concept were set at the beginning to reach: (i) a final energy balance ≤ 22.5 kWh/(m<sup>2</sup> yr); (ii) DHW from renewable energy sources ≥ 9 kWh/(m2 yr); and (iii) PV installation ≥ 53 kWp for the entire complex. Furthermore, the optimization of the solar systems and the improvement of interior daylight conditions were also investigated. An Integrated Design Process (IDP) was followed to assure good stakeholders' involvement.

#### 4.2.3. The Planning Process

The refurbishment was carried out in three phases. The first was the design phase, which was developed through an IDP by a multidisciplinary team of experts from the Municipality of Bolzano, the design team, research institutes, and Agenzia Casaclima (the local certification body for energy efficiency in constructions). The two following phases were the construction phase and the monitoring phase, which lasted one year. All the spatial scales were investigated during the five years of the realization (2014–2019).

#### 4.2.4. Energy Concept

The buildings' complex was originally realized in the 1990s without energy-saving measures, apart from a thin insulation layer (4 cm) on the external walls. Hence, the project involved the refurbishment of the existing facades by using timber prefabricated multifunctional elements to improve the energy performance and the indoor thermal comfort as well as aesthetically rehabilitate the buildings. Renewable energy strategies were also adopted with the installation of PV and ST systems on the roof and a geothermal heat pump serving centralized heating. PV modules with 15.5% efficiency, covering an area of 337 m2, are installed on the roof of the northern building (Figure 8). The system has a power of 20 kW, and the modules are mounted horizontally (0◦ tilt angle) to avoid mutual overshadowing effects. Conversely, the roof surface of the southern building has been used for the installation of a 477 m2 ST system with a 10◦ tilt angle and with an estimated production of 11.2 kWh/(m2 yr) for DHW. This choice was dictated by the need to comply with the national legislation, requiring to cover at least 35% of the energy need for DHW with RES [32]. Both PV and ST cannot be considered architecturally integrated into the building, since they constitute free-standing elements placed on the roofs.

**Figure 8.** Technical plan of PV system on the roof of the north building (source: © EQ Ingegneria).

#### 4.2.5. Environmental, Economic, and Social Impacts

The SINFONIA project intended to have an impact on the entire municipality of Bolzano by improving the quality of life of the citizens, turning the city into a smart city. The installation of over 100 smart points and three multifunctional interactive totems provides services such as charging electric vehicles, monitoring air quality and weather, and improving the lighting of public spaces. The installation of this diffuse technological network runs in parallel with the refurbishment of six residential housing complexes that enhanced the interior comfort and energy performance of 455 dwellings for a total of around 15 million euros of investments. From a social perspective, the tenants of the refurbished apartments have been directly involved through informative meetings and questionnaire-based surveys during the realization of the project. Additionally, displays have been installed in part of the apartments; data on energy performance and indoor conditions are shown in real-time, enabling the inhabitants to take actions to reduce their consumption. This was completed to raise awareness of the role that their behavior plays in the energy performance of buildings (i.e., ventilation rate, air temperature control, energy use) [52].

#### 4.2.6. Approaches, Methods, and Tools

During the IDP, retrofit solutions were assessed before adopting a mixed approach combining the use of traditional insulation for the lodges and prefabricated multifunctional facades (MFF) for the rest of the building envelope. This choice did not require the use of scaffolding, simplifying and speeding up the construction phase, ensuring a low impact on the site's inhabitants. To define the best solutions for the MFF, prototypes were constructed and tested in laboratory tests. Moreover, the design focused on defining the optimal solution for solar active systems, assessing the potential energy production, and preserving the indoor visual comfort. To this aim, solar potential and daylight analyses were conducted using Diva for Rhino [53] and LadybugTools [54]. To assess the solar potential, close shading elements were modeled in the 3D environment of Rhinoceros [55], while the distant obstacles, such as the mountains, were considered by a horizon line generated from data measured with a Solmetric SunEye. The optimization process was carried out using a genetic algorithm to maximize the total annual and average irradiation and to minimize overshadowing effects. As the installation of the new prefabricated MFF caused a reduction in interior daylight availability, daylight analyses were performed to verify the compliance with the Italian national legislation (i.e., daylight factor above 2%). Details such as the windowsills' materials and loggias' colors were investigated to improve the solar mutual reflection effects. The simulations and the laboratory tests were conducted during the design phase, which allowed reducing the risks and uncertainties during the construction phase.

#### 4.2.7. Lesson Learned and Recommendations

The main lesson learned from the SINFONIA case study is the benefit that an IDP can have on energy-related measures applied to the building.

Fundamental is an early collaboration between the different experts and stakeholders involved in the project (i.e., energy consultants, designers, owners, researchers, technicians). Furthermore, the case study demonstrates the importance of conducting solar potential and daylight analyses during the initial phase of the project by supporting the design team in the most important choices (i.e., finishing materials and colors to improve the interior daylight level, localization of active solar systems to avoid overshadow effect). Finally, it is recommended to consider how the occupants' behavior can influence the energy use of a building and try to engage and inform the users accordingly.

#### *4.3. Le Albere in Trento*

#### 4.3.1. Overview

Le Albere is a district in the city of Trento (Trentino Alto Adige) that is constructed on a former industrial site enclosed between the railway and the river Adige. The aim of the project, designed by the Italian architect Renzo Piano, was to reconnect the site with the city center and to re-establish a relationship with its natural context. The result of the intervention is a mixed-use district that includes offices, residential buildings, a science museum (MUSE), a library for the University of Trento, and a park (Figure 9) [56]. The main feature of the project is the extensive integration of active solar systems on roofs and facades of the buildings of the whole complex. In that regard, the choice to adopt this energy strategy was autonomously taken by the actors involved since, at the time of the design, the Italian national legislation on energy and urban planning did not set any specific target for solar energy integration, despite supporting the use of RES.

**Figure 9.** (**a**) Aerial view of the Le Albere district (source: © Google Earth) and (**b**) view from the public park (photo: © Silvia Croce).

#### 4.3.2. Challenges, Issues, and Decision Strategies

The main challenge for the case study has been the formal integration of the photovoltaic systems in architectural design. High requirements were set by the architect in terms of the color, appearance, and dimension of PV modules due to the choice of having them as a key and visible element of the project. The strategy adopted by the municipality of Trento was to create a high-quality district, showcasing the best practices in terms of energy efficiency and sustainability, with extensive integration of RES.

#### 4.3.3. The Planning Process

The planning process encompassed different spatial scales, spanning from the provincial down to municipal, local, and detail scales. Le Albere was a former industrial area occupied since 1937 by the tire manufacturing company Michelin. When the factory ceased its activity in 1998, a group of public and private investors, under the name of "Initiative Urbane", purchased the site to regenerate and reconnect it to the rest of the city.

In 2000, the Province of Trento included the renewal of the district into its Program for Urban Regeneration and Sustainable Development of the Territory and into the Trento Strategic Masterplan. Two years later, the architectural firm Renzo Piano Building Workshop was designated to realize the project, and a final master plan was approved in 2004. The construction phase started in 2008 and lasted for five years when the MUSE, the mixeduse district, and the public park were inaugurated. Conversely, the university library located in the southern part of the parcel opened at the end of 2016.

#### 4.3.4. Energy Concept

Several RES were used in the Le Albere district, such as BIPV systems integrated on many of the buildings that are present on site. The systems have a nominal power of 279 kW and cover a total area of approximately 3200 m2. Tedlar–glass and glass–glass modules with different orientations and inclinations were specifically designed depending on their location. They are integrated into the buildings' facades and roofs or used as PV shading devices. The electricity produced by the BIPV modules covers the electrical demand of offices, common areas, pump rooms, and the lighting in the basement. The MUSE is served by a geothermal plant while a combined cooling, heating, and power plant, outside of the district's boundary, covers the heating and cooling demand for the whole area.

#### 4.3.5. Architecture, Visibility, Sensitivity, and Quality

Following the LESO–QSV methodology, the integration of PV modules has been judged as fully coherent for all three categories of geometry, materiality, and pattern (Figure 10). Despite the different heights, inclinations, and shapes of the buildings, PVs act as unifying elements for the entire complex, further enhancing and characterizing it. The level of criticality, as a function of context sensitivity and system visibility, can be considered as medium. In fact, the case study is bordered by urban areas without any particular architectural or historical value, with the only exception of a renaissance villa/fortress on the north side. Regarding the systems' visibility, it is high from internal streets, open areas of the park, and roads bordering the site when PVs are installed on facades or integrated into glass surfaces constituting the tilted roofs. On the contrary, low perception levels for an observer located in the urban canyons of the district are achieved when the system is integrated on the opaque galvanized roofs. From an observer in the courtyards or in the park, the photovoltaics are perceived as a geometrical pattern. Finally, the remote visibility is low due to the overall height and dimensional scale of the complex, which is comparable with those of the historical center and the existing industrial structures.

**Figure 10.** Architectural integration matrix for the PV systems in the Le Albere district in Trento (left); (**a**) the two typologies of BIPV systems: tedlar–glass modules on the galvanized roof and the glass–glass modules in the background (photo: © Silvia Croce); (**b**) detail of the glass–glass PV modules installed on the facades (photo: © Silvia Croce).

#### 4.3.6. Environmental, Economic, and Social Impacts

Sustainability was central in the Le Albere project since the first phases of the design, starting from the construction materials sorted locally whenever possible to limit the environmental impact. The master plan was developed to preserve the natural features of the area by concentrating all the new constructions toward the east side of the plot and

creating a large public park on the west side. The existing street, running parallel to the river, was transformed into an underground passage to allow the continuity of the park and directly interact with the river Adige. Regarding mobility, the realization of two-story underground parking greatly reduces the number of cars on the surface, while the traffic in the district is limited to residents, taxis, and public transport. Great attention was also placed on the design of numerous walkways and bikeways. Furthermore, the MUSE and the university library have obtained the LEED Gold certification, while residential buildings and offices, built according to passive standards, received the level B label of CasaClima (i.e., energy-efficiency classification), corresponding to a heating energy requirement below 50 kWh/(m<sup>2</sup> yr). The project carries a high social value thanks to the restitution of the area to the public realm with a new identity. The district has now a strong cultural and recreational vocation due to the presence of two social attractors (MUSE and university library) at the borders of the plot and a five-hectare park with residential and commercial buildings in the middle. The renewal intervention, which cost around 350 million euros, was entirely financed by a pool of banks, insurance companies, and industry associations in collaboration with the Municipality and the Province of Trento.

#### 4.3.7. Approaches, Methods, and Tools

The approach used for the design of the case study has been strongly oriented toward sustainability. Building information modeling (BIM) tools were used along the entire process to integrate the buildings' design with energy modeling and considerations on internal and external comfort conditions.

#### 4.3.8. Lesson Learned and Recommendations

The Le Albere case study strongly highlights the potential of solar systems integration not only as energy generators but also as architecturally valuable elements that can provide identity and enforce the aesthetic of the urban intervention. This is realized by declaring the presence of the photovoltaic modules and giving them high visibility instead of hiding or camouflaging them. When this approach is pursued, the quality of the system's integration becomes of primary importance and has to be carefully designed. In addition, the project poses sustainability, under its various declination of social, economic, and environmental, as the pivotal element to guide the design process, successfully resewing a large portion of the city with the rest of the urban fabric.

#### *4.4. Violino District in Brescia*

#### 4.4.1. Overview

The Violino district is a new residential area located in the outskirt of the city of Brescia (Lombardia). It is constituted of 112 terraced houses and 2 multi-family houses five floors high (143 housing units), which are organized in lots defined by an orthogonal grid (Figure 11). It is the result of a new urban plan for social housing followed by an architectural competition announced by the Brescia Municipal Council [57] to design a district according to bioclimatic principles and with the extensive use of solar strategies. The area required a northeast/southwest oriented grid, which was judged to be not optimal for the full exploitation of the thermal and energy potential of the sun. The issue was overcome by adapting the terraced house' typology to the rigid grid with a partial rotation of the buildings' masses, ensuring adequate solar accessibility to all accommodations.

**Figure 11.** (**a**) Aerial view of the terraced houses of Violino District (source: © BAMSphoto—Basilio) and (**b**) the two multi-family houses (photo: © Fabio Cattabiani).

#### 4.4.2. Challenges, Issues, and Decision Strategies

The main challenges were the need for an adequate urban quality in social housing, elevated technical performances to reach sustainability and energy targets, and the rigid urban grid. Most of these challenges were highlighted in the call for tenders, and the participants were asked to submit proposals with quantifiable quality requirements. This allowed adopting a decision strategy based on a scoring system and conveying information on the conformity of each proposal to performance and quality requirements (i.e., thermal, acoustic, hygrometric insulation, daylight, energy production) as well as understanding the economic impact of these measures compared to standard social housing.

#### 4.4.3. The Planning Process

The origin of the district dates back to 1980 with the Piano Regolatore Generale (PRG) of Brescia drawn up by Secchi, Viganò and Scarsato, and it emerged in its current configuration in 2002 when the Municipality of Brescia purchased the land and issued a tender for the development of a new social housing intervention in the southwest sector. Architects, installers, and consultants collaborated in the planning process, while the construction phase (from 2004 to 2006) was carried out by local cooperatives and construction companies. The intervention was financed with funding from the Lombardia Region and from Azienda Lombarda Edilizia Residenziale (ALER). All the spatial scales were investigated during the realization of the project.

#### 4.4.4. Energy Concept

The entire district has been designed according to bioclimatic principles to benefit from climatic conditions. Therefore, each building is oriented to maximize daylight and passive solar gain as well as the generation of electricity through photovoltaic systems. A characteristic element is represented by the greenhouses integrated into the terraced houses and acting as solar collectors and thermal buffers. The sun rays that penetrate the windows are transformed into heat in contact with a massive surface: the heat is retained, thus counteracting the excessive temperature ranges. The terraced houses are all equipped with photovoltaic systems of 1.3 kWp mounted flat or with a 30◦ southwest inclination, while multi-family houses have systems ranging from 5 to 20 kWp with a 30◦ angle.

#### 4.4.5. Architecture, Visibility, Sensitivity, and Quality

The integration of PV modules in the terraced houses can be described as fully coherent regarding system geometry and materiality, while they can be considered partially coherent for the modular pattern (Figure 12). Many gaps are visible between the panels, giving the overall impression of an uneven surface. Regarding the systems installed on the two multifamily houses, they cannot be considered as coherent with any of the listed aspects, since they are mounted inclined on a flat roof, using a metal structure. The area is characterized

by a medium sensitivity level due to the absence of historical buildings, monuments, or meaningful elements. The close visibility of PV panels is also classifiable as medium, being the systems either installed on tilted roofs, visible from the street canyons, or on flat roofs. Concerning the remote visibility, the plain terrain of the site does not provide any vantage point from where the PV systems are visible, with the only exception represented by the top stories of the two multi-family houses.

**Figure 12.** Architectural integration matrix for the PV systems at Violino District in Brescia (left); (**a**) system modular pattern (photo: © Fabio Cattabiani); (**b**) view from the roof of one of the multifamily houses (photo: © Fabio Cattabiani).

#### 4.4.6. Environmental, Economic, and Social Impacts

The environmental dimension of the project resides in the bioclimatic approach utilized for the design of the entire intervention, where great attention was placed on the exploitation of solar radiation. From the economic and social point of view, the project was fully financed by public funds with the scope to provide an affordable and high-standard social housing solution. Some accommodations were reserved for specific age groups to create a more heterogeneous and diversified environment. For instance, 26 dwellings are for elderly people, while 12 are for young couples. The rationalization of local roads, with the interposition of three public pathways, provides a protected connection to the main area of the neighborhood and fosters public aggregation and socialization.

#### 4.4.7. Approaches, Methods, and Tools

The approach adopted for designing the terraced houses aims to guarantee the "right of sun" to each residential unit. The interior distribution of spaces, reflected in the exterior composition of volumes, is based on their function. The south orientation is dedicated to the most used spaces, namely the living room on the ground floor and individual rooms at the upper level. Services such as bathrooms and vertical distribution are on the north side, while the west side is occupied by the kitchen and the master bedroom. The greenhouse, painted in dark hues to further enhance heat gains in winter, is integrated into the south facade. Colors have also played a significant role in the project, as a study has been carried out to diminish the perception of the repetitiveness of the terraced houses' typology using different shades of colors. Furthermore, this choice highlights the volumetric composition of the residential units and guarantees different perceptions during the day according to the various light conditions (Figure 13). Since 2007, the project was also subjected to a monitoring campaign to evaluate the performance of PV systems, their energy production, and the amount of electricity fed into the grid. Measurements show that the PV systems installed in housing units produce, on average, 4.54 kWh per day. Moreover, between 2014 and 2015, the Smart Domo Grid Project [58,59] was carried out by the energy company A2A S.p.A., the Politecnico di Milano, and the company Whirlpool.

**Figure 13.** Volumes' composition (**top left**); section of terrace house with passive strategies (**bottom left**); color plan by Jorrit Tornquist (**right**) (source: © Boschi + Serboli Architetti Associati).

The project involved 21 families and consisted of testing a smart system to enhance energy efficiency and reduce electricity costs for the users. Using an energy management application that measures consumption and provides insight on how to minimize energy electricity costs, users could avoid overload and plan the use of appliances in the most convenient moment of the day.

#### 4.4.8. Lesson Learned and Recommendations

The Violino District underlined the importance to work simultaneously on volume composition, internal space distribution, and technical systems to maximize the exploitation of passive strategies and achieve hygrothermal comfort conditions. The sun was used as an important bioclimatic element in the design process to determine buildings' orientation and facades' exposure. The greenhouses were used as natural temperature regulators to trap and slowly release solar heat gains, especially in winter.

#### *4.5. CasaNova District in Bolzano*

#### 4.5.1. Overview

CasaNova is a large social housing project in Bolzano (Trentino Alto Adige). It has been developed by the municipality to provide affordable rental solutions while guaranteeing high living standards and a low ecological footprint. It is located in the southern parcel of the city, on an area of 10,000 m<sup>2</sup> delimited on the northern side by an existing neighborhood and the countryside, and on the southern side by the railway and the Isarco river. Groups of three to four buildings each form eight urban blocks arranged around green courtyards for a total of 950 apartments (Figure 14), which mostly comply with the highest level of CasaClima energy classification.

**Figure 14.** (**a**) Aerial view of CasaNova district showcasing the buildings' arrangement around courtyards to form urban blocks (source: © altoadige.it); (**b**) view of the block designed by CDM Architetti Associati in the north part of the plot (photo: © Andrea Martiradonna).

#### 4.5.2. Challenges, Issues, and Decision Strategies

Despite being the project established on a greenfield (former agricultural area), the municipality chose the lot for its position in continuity with an already urbanized area to avoid excessive decentralization. The focus was placed on creating a strong relationship between the new development and the countryside through extensive use of greenery, both at ground level and on the roofs, as well as in the design of pedestrian paths and buildings' alignments. On the other hand, several issues were encountered during the realization of the district such as the absence of a well-established connection system with the rest of the city, the presence of rural buildings within the plot, and hydrogeological risks related to the proximity to the river and the high level of the water table in the area. Furthermore, the inhabitants of the adjacent neighborhoods and the farmers firmly opposed the project due to its possible future expansion [60].

#### 4.5.3. The Planning Process

The project started in 2001 when the Bolzano municipality acquired 10,000 m2 of farmland and turned it into buildable land. After an architectural competition held at the European level [61], the realization of a master plan was assigned to the Dutch firm Frits Van Dongen. During the urban design phase, several workshops were carried out together with the political and technical representatives of the municipality, representatives of the owners, the neighborhood council, and other actors involved in the project. Successively, the eight lots were designed by different architectural firms after another competition organized by the Bolzano municipality. The construction phase started in 2007 and terminated in 2014 with the realization of a public space with a mix of functions in the center of the plot. Finally, a cycling path and a train station were built to improve the connectivity of the neighborhood with the rest of the city.

#### 4.5.4. Energy Concept

Before starting the architectural urban design, the energy plan of the neighborhood was developed. It set the compliance with the highest CasaClima certification energy standard (i.e., heating index of 30 to 50 kWh/m2y depending on the surface/volume ratio of each building) [62]. The energy concept included passive strategies of bioclimatic planning approach and active systems to produce electricity and DHW (Figure 15).

The compact shape of the buildings aims to maximize solar exposure and reduce overshadowing effects by placing the higher buildings in the north part of each block with sloped roofs in a unique direction. The PV and ST systems installed on most of the buildings cover almost completely the DHW need for the neighborhood. Space heating during winter and the pre-heating of the DHW is guaranteed by geothermal heat pumps and by the district heating network, which are connected to a cogeneration plant. Space cooling in summer is provided by a controlled ventilation system where the air is pre-cooled by the geothermal pumps. These solutions allow a reduction of 65% of the energy consumption compared to a traditionally built neighborhood.

#### 4.5.5. Environmental, Economic, and Social Impacts

The environmental impact of the CasaNova district is kept low thanks to the design approach oriented toward a strong reduction in energy consumption. An example is in the construction method, which used reinforced concrete only for the main structure and in the slabs, and prefabricated elements for the external walls. The prefabricated elements allowed to cut the construction time nearly in half, and save electricity and water in situ due to the minor use of cranes and machines. Another environmental feature is the reuse of the rainwater thanks to the installation of four water tanks at the corners of each urban block. The collected water is used to irrigate the green areas at the ground and roof level and for toilets. The outdoor thermal comfort is addressed by arranging the buildings to exploit the fresh breeze in summer and act as a protection against cold drafts in winter and by the extensive presence of green spaces. To reduce the impact of traffic and to guarantee good air quality standards, only one road crosses the districts, and all parking spaces are located underground. Finally, the large offer of affordable rental solutions and the good accessibility achieved after the realization of a new train stop, bus connections, and cycling paths represent a relevant economic and social impact achievement of the project.

**Figure 15.** Design concept of CasaNova district. At the neighborhood level: (**a**) urban mobility and accessibility to the area, (**b**) social public spaces organization, (**c**) shadow cast and buildings' morphology to reach the highest energy class according to the Italian energy certificate system. At the building level: (**d**) morphology of the building block, (**e**) internal connections and social spaces, (**f**) study of the views from the building stock, (**g**) study of the solar exposure and orientation, (**h**) green roofs and water management system, and (**i**) use of RES; (Source: © Wienerbeger).

#### 4.5.6. Approaches, Methods, and Tools

The case study aims to reduce energy consumption, maximize solar exposure, and create a permeable urban fabric able to establish a dialogue with the countryside. The architects' approach reflects these objectives in the arrangement of compact-shaped buildings around open courtyards, with taller ones located on the north side of the blocks to avoid overshadowing effects. The same attention is reserved for the composition of the single housing units, where large glazed areas are used to exploit solar heat gains and every apartment has a loggia oriented toward south or west. Finally, in 2009, Eurac Research was appointed as technical and operational support for building monitoring. A campaign was carried out to verify the energy consumption and internal comfort conditions, and to compare them to the design values. The results show that seven out of nine blocks consume more than what was calculated during the design phase. The probable causes have been identified both on technical management and user behavior [62].

#### 4.5.7. Lessons Learned and Recommendations

The shape, orientation, and reciprocal arrangement of the buildings demonstrated how the preliminary planning allows reaching high energy and living standards. This approach contributes to the energy saving of the entire district, and it lasts for its entire lifetime. Aspects such as internal layout, choice of materials, rainwater management, use of district heating and cooling, and integration of active solar systems are considered. It is important for architects and city planners to have the right instruments to evaluate these factors since the early design phases. A continuous process of verification and optimization allows to solve technical and management problems and reach the planned energy performance. The case study shows the importance of infrastructural connections and recreative spaces: the district's attractiveness is guaranteed by a new train stop, bicycle lane, green spaces, and a multifunctional area with services.

#### *4.6. Similarities and Differences among the Different Built Environments*

Table 3 presents the similarities and differences among the built environments.

**Table 3.** Similarities and differences among the different built environments.


**Table 3.** *Cont.*


*4.7. Agrovoltaico*

4.7.1. Overview

Recent years have seen the development of solutions called "agrivoltaic" combining the dual use of food production at the ground level with electricity generation through PV modules located at an upper layer (Figure 16) [10,63]. The case study presented, the Agrivoltaico ® [64], is a patented system developed by the private company REM (Revolution Energy Maker) and tested for the first time in Pianura Padana (Emilia Romagna), a vast alluvial plain corresponding to the river Po's basin in Northern Italy. The area has a strong agricultural vocation thanks to a capillary irrigational system and a flat homogeneous morphology. These characteristics make the territory suitable for the installation of solar systems, which can however compete with the production of food.

**Figure 16.** View of the dual use of land with the PV track system on the upper layer and agricultural works at ground level (photo: © REM).

The solution proposed by REM exemplifies how a double use of land can be achieved. The system was built in a period of strong opposition toward the realization of utility-scale photovoltaic systems on agricultural land (Figure 17).

**Figure 17.** (**a**) View of the PV track system during the first crop (photo: © REM)*;* (**b**) aerial view of the solar plant in landscape (photo: © REM).

#### 4.7.2. Challenges, Issues, and Decision Strategies

The main challenges encountered during the development of the project were related to the need to overcome the national legislation that did not allow for the installation of large PV systems (above 1 MWp) in agricultural areas [65]. This has been possible thanks to an innovative solution, where the PV modules are suspended 5 m above the ground using a metal structure. Only 2% of the land is occupied by the punctual supports holding the metal structure (in compliance with the regional regulation requiring a soil occupation for PV installation < 10%), leaving great freedom of movement to the harvest machines operating below. To balance the low density of the system compared to a standard one mounted on the ground, double-axis sun-tracking devices were used for the panels. The different configurations of the system during the day and its low density allow a minimal impact on the total radiation falling to the ground. Regarding the decision strategy used for the project, REM collaborated with several industrial partners, a university (the Institute of Agronomy, Genetics and Field crops of the Università Cattolica del Sacro Cuore of Piacenza), and local authorities to assure the design's requirements.

#### 4.7.3. The Planning Process

The project was realized in four phases between 2010 and 2012 and addresses landscape and architectural planning scales. The scope was to overcome legislative and public acceptance barriers for the installation of PV systems in agricultural areas through an innovative solution allowing food and energy production with the regional regulation on land occupation. The private investor REM initially invested 2.5 million euros in the pilot project before replicating it in two other Italian locations for a total of 30 million euros in value including the satellite activities and nearly 700 people involved.

#### 4.7.4. Solar Landscape

The notion of a "solar landscape" was proposed in [10], and it consists of a different approach to the design of PV systems on the ground. The focus is not anymore on the integration of the system in the landscape but rather the design of the system as a landscape. This different perspective requires shifting the focus of the PVs' planning from the sole energy performance to a new set of paradigms, which are more in line with the ecological burden that the creation of a new landscape implies. The pattern of the system (dense or porous) is determined by the spatial arrangement of the modules, the type and grain of the patch, the pattern type, and its edges (Table 4). Other important aspects are the function of the space and its connectivity. The first aspect is related to the porosity of the system that determines the amount of solar radiation reaching the ground and consequently its ecological potential. No changes in the land's function or type of cultivation were needed for Agrivoltaico due to the high porosity of the system. The second aspect depends primarily on the distance of the PV modules from the ground and if any fence or enclosures

are present at the boundary of the area. The case study has a great level of connectivity with the surrounding landscape, allowing animals, people, and harvesting machines to easily cross the area. The installed nominal power is 3.2 MW, and it is composed of 11,535 polycrystalline PV modules tiltable on two axes using a wireless control.


**Table 4.** Formal functional features of the solar landscape for Agrivoltaico.

#### 4.7.5. Site Potential

The site has a cultural value and requires the preservation of its features, being the product of a long human transformation of a humid area into one suitable for agriculture. Nowadays, the area is intensively exploited for the cultivation of crops thanks to the efficient network of irrigation and flat morphology. Despite the PV system suspended above the ground does not perform any other function than producing energy, the site represents an example of double use of land (i.e., agriculture/food production and energy production from PV system).

Moreover, the PV double-axis tracking system controlled by a specific algorithm allows for optimal control of the dynamic shading on the crops, aiming at optimizing agricultural production. The assessment of the sensitivity (low or high) of several selected landscape factors is visible in Table 5.


**Table 5.** Sensitivity of landscape factors for Agrivoltaico.

#### 4.7.6. Environmental, Economic, and Social Impacts

The environmental impact of the system can be considered low, since the original use of the land is preserved and the system does not constitute a barrier in the territory. Furthermore, it is entirely made of safe, non-polluting, and recyclable materials (as the recycled aluminum that the trackers are made of) and its construction technique makes it easy to disassemble at the end of the lifecycle (25–30 years) [66]. Despite the visibility of the system changes during the day because of the tracking, no mitigation strategy was required due to the porous pattern of the modules. Finally, the collaboration established with local authorities since the early phases of the planning process helped to achieve an optimal design and raised public awareness and participation.

#### 4.7.7. Approaches, Methods, and Tools

The company REM Tec has patented the Agrivoltaico technology worldwide. Two solutions of sun-tracking systems are currently available, having in common the length (12 m) and the height (4 to 5 m) of each tracker but differing in the system's power and the number of modules. In the first one, ten panels are installed on each tracker for a peak power spanning between 2.5 and 4.35 KW, while the second solution presents a denser configuration with 32 panels and peak power from 8.64 to 10.46 KW. Furthermore, the cooperation with the Università Cattolica del Sacro Cuore of Piacenza permitted us to understand the impact of the system on different crop species and develop optimal design solutions combining energy and food production.

#### 4.7.8. Lesson Learned and Recommendations

The Agrivoltaico case study demonstrates that the ecological impact of large PV systems installed on the ground can be greatly reduced with the adoption of innovative solutions allowing a double use of land. Barriers posed by legislation and regulations can often be overcome through a conscious design approach and collaboration with local authorities, industrial, and scientific partners. In the final analysis, the installation of large PV systems has to be considered as a matter of landscape design, where the paradigm is shifted from the sole energy production toward a more holistic approach, taking into account the various ecological aspects of a landscape transformation.

#### **5. Limitations of the Study**

This study does not pretend to be seen as an exhaustive illustration of the status of solar energy in the Italian urban context but it rather wishes to give a satisfactory overview of it through exemplary cases selected and analyzed by experts in the field. The selection of the six case studies reflects the knowledge and available information provided by the experts during the IEA SHC Task 51 "Solar Energy in Urban Planning". The geographical distribution of the cases, all located in the northern part of Italy, can be considered as the major limitation of the study. The inclusion of projects from southern regions could have further enriched the study and investigated different climatic conditions. Furthermore, three out of six cases are located in a single region (i.e., Trentino Alto Adige), which has special administrative conditions compared to the rest of the country. More landscape PV case studies should be analyzed to provide a more complete overview of this environment as well as a critical analysis of the similarities and differences. It is also worth mentioning that IEA SHC Task 51 was terminated in 2017; therefore, potentially relevant cases realized after that year were not included in this paper. The ongoing IEA SHC Task 63 "Solar Neighborhood Planning" [67,68] will partially address this issue.

#### **6. Conclusions**

In this paper, six case studies located in the Italian territory were presented and discussed following a common template developed within Subtask C of the IEA-SHC Task 51 "*Solar Energy in Urban Planning*". The integration of solar energy in three different environments—namely existing urban areas, new urban areas, and landscape—has been investigated. General considerations can be outlined, allowing for a partial generalization of the results for the entire country or at least for the administrative territory in which the cases are located. The main lesson learned is the importance of including solar energy since the beginning of the design process to achieve a higher level of integration.

When this basic rule is not observed and a lack of communication between experts and stakeholders occurs, the risk of encountering pitfalls during the following phases exponentially increases, and only sub-optimal results can be obtained. An additional advantage of the early analysis of solar energy potential is the possibility to optimize the urban morphology and the buildings' form to maximize active and passive solar strategies. Several design aspects can be controlled in the early design phases, although they tend to crystallize rapidly as the project progress. Furthermore, an optimized design of buildings' masses is a passive measure that can have positive and long-lasting impacts. This aspect is particularly visible when dealing with existing buildings, often offering limited design solutions if not subject to deep renovation interventions. Finally, an important lesson is the need to consider solar systems beyond their electricity generation function. This is true when the integration occurs into buildings as well as with utility-scale landscape systems. A set of recommendations specific for the three different analyzed environments can be summarized as follows.

	- Deep renovation processes involving morphological and material changes represent potential successful solutions for solar systems integration as demonstrated by the case studies SINFONIA Bolzano and Photovoltaic Village in Alessandria.
	- The use of solar systems in public areas (e.g., shading devices on pergola as in the Photovoltaic Village in Alessandria) has a high unexploited potential.
	- The institution of coordinating bodies and the adoption of an integrated design process can have a significant role in the application of energy-related measures while improving interdisciplinarity and collaboration among stakeholders as happened in both SINFONIA Bolzano and Photovoltaic Village in Alessandria.
	- It is important to use simulation software since the early-design phases and to focus on final-user behavior as in SINFONIA Bolzano.
	- Photovoltaics can be utilized as a distinctive architectural element and material, which can enforce the identity and aesthetic of urban interventions as in the case studies of the Le Albere district, Violino District in Brescia, and CasaNova.
	- The shape, orientation, reciprocal arrangement of volumes, materials, internal layout, and opening distribution are long-term passive strategies that can maximize the contribution of solar energy to building efficiency and comfort, as visible in the case studies of the Violino District in Brescia and CasaNova in Bolzano.
	- The last two points listed in the existing urban areas are also applied for new urban areas as demonstrated in all the presented case studies such as Le Albere district, Violino District in Brescia, and CasaNova.
	- The ecological impact of ground-mounted PV can be greatly reduced using innovative solutions that combine a dual use of land.
	- Barriers represented by regulations and legislation can be often overcome with a conscious design and an early collaboration between the different involved actors.
	- PV installation at the ground level should be considered as a landscape design matter, where the pattern, patch, grain, and borders of the system are carefully planned.

**Author Contributions:** Conceptualization G.L., M.F. and S.C.; methodology G.L., M.F. and S.C.; investigation M.F.; resources G.L., S.C., D.V., R.P. and A.S.; writing—original draft preparation G.L., M.F. and S.C.; writing—review and editing M.F., G.L., S.C., D.V., R.P., A.S. and A.G.M.; supervision G.L. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by The Research Council of Norway (NFR) and the Norwegian University of Science and Technology (NTNU) through the Zero Emission Neighbourhoods in Smart Cities Research Centre (FME ZEN) (project No. 257660) and the research project HELIOS enHancing optimal ExpLoitatIOn of Solar energy in Nordic cities through digitalization of built environment (FRIPRO FRINATEK) (project No. 324243), the Department of Civil, Environmental, Architectural Engineering—University of Padua, the Department of Architecture Built Environment and Construction Engineering—Politecnico di Milano, the Italian Ministry of Economic Development in the framework of the Operating Agreement with ENEA for Research on the Electric System, and the European Union's Seventh Programme for Research, Technological Development and Demonstration

(grant agreement No. 609019). The European Union is not liable for any use that may be made of the information contained in this document.

**Acknowledgments:** The authors wish to thank all the experts within the research project IEA SHC Task 51 "Solar Energy in Urban Planning", who contributed to the successful collection and analysis of the Italian case studies within the subtask C "Case studies and actions research" and for the copyright of graphics, images, tables, and figures. The authors also wish to express their gratitude to the IEA SHC Executive Committee for supporting Task 51 and in this way strengthening the international collaboration among researchers and practitioners. This work has been written within the FME ZEN Research Centre and the research project HELIOS. The authors M.F. and G.L. gratefully acknowledge the support from the FME ZEN Research Centre, its partners, and the NFR. The author R.P. wishes to thank the Department of Civil, Environmental, Architectural Engineering—University of Padua. The authors S.C. and D.V. acknowledge the European Union's Seventh Programme for Research, Technological Development and Demonstration. The author A.S. thanks for the support received by the Italian Ministry of Economic Development in the framework of the Operating Agreement with ENEA for Research on the Electric System.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


### *Article* **Numerical Validation of the Radiative Model for the Solar Cadaster Developed for Greater Geneva**

**Benjamin Govehovitch 1,\*, Martin Thebault 2, Karine Bouty 2, Stéphanie Giroux-Julien 1, Éric Peyrol 3, Victor Guillot 4, Christophe Ménézo <sup>2</sup> and Gilles Desthieux <sup>4</sup>**


**Citation:** Govehovitch, B.; Thebault, M.; Bouty, K.; Giroux-Julien, S.; Peyrol, É.; Guillot, V.; Ménézo, C.; Desthieux, G. Numerical Validation of the Radiative Model for the Solar Cadaster Developed for Greater Geneva. *Appl. Sci.* **2021**, *11*, 8086. https://doi.org/10.3390/ app11178086

Academic Editors: Tiziana Poli, Andrea Giovanni Mainini, Gabriele Lobaccaro, Mitja Košir, Juan Diego Blanco Cadena and Constantinos A. Balaras

Received: 24 June 2021 Accepted: 25 August 2021 Published: 31 August 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

**Abstract:** The achievement of the targets for reducing greenhouse gas emissions set by the Paris Agreements and the Swiss federal law on the reduction of greenhouse gas emissions (CO2 law) requires massive use of renewable energies, which cannot be achieved without their adoption by the general public. The solar cadaster developed as part of the INTERREG G2 Solar project is intended to assess the solar potential of buildings at the scale of Greater Geneva—for both industrial buildings and for individual residential buildings—at a resolution of 1 m. The new version of the solar cadaster is intended to assess the solar potential of roofs, as well as that of vertical facades. The study presented here aims to validate this new version through a comparison with results obtained with two other simulation tools that are widely used and validated by the scientific community. The good accordance with the results obtained with ENVI-met and DIVA-for-Rhino demonstrates the capability of the radiative model developed for the solar cadaster of Greater Geneva to accurately predict the radiation levels of building facades in configurations with randomly distributed buildings (horizontally or vertically).

**Keywords:** solar cadaster; solar potential modeling; numerical validation

#### **1. Introduction**

The massive use of renewable energies and, more particularly, solar energy is necessary in order to meet the objectives of the Paris Agreements and the Swiss federal law on the reduction of greenhouse gas emissions (CO2 law) [1]. As buildings are some of the main contributors to climate change, one of the main challenges is, therefore, turning buildings from energy consumers into energy producers. This cannot be done without the development of tools dedicated to the evaluation of solar potential.

This massive deployment involves the evaluation of solar potential on a large scale. In addition, the urban environment has many specific features that are to be taken into consideration when evaluating this potential. This can be achieved with solar cadasters.

However, these tools are usually limited to the evaluation of the solar potential on rooftops [2–5]. This limitation can be an obstacle to the development of the use of solar energy because the potential of vertical surfaces in an urban environment can prove to be predominant [6–9].

Thus, there is a need for large-scale tools that are able to take vertical facades into account, in addition to roofs. From a physical point of view, the effects of surrounding surfaces (the ground or surrounding buildings) on the vertical radiative balance are not well known and may appear to be non-negligible. From the point of view of solar production potential, the issues of the variability linked to shading effects are added. Tools that are able to consider these issues are being developed and need validation.

A solar cadaster tool is being developed at the scale of Greater Geneva, which is a trans-border agglomeration around the city of Geneva between France and Switzerland, and it totals an area of 2000 km2. A first solar cadaster was developed for the Canton of Geneva only (280 km2), as presented by Desthieux et al. [10]. The goal now is to update and extend this solar cadaster for the whole of the Greater Geneva area in the framework of INTERREG G2 Solar. Its aim is to intensify solar energy production in the agglomeration by mobilizing a large panel of stakeholders from public institutions, municipalities, energy providers, the private sector, and universities. The difficulty lies in the fact that the solar cadaster must cover a large area (2000 km<sup>2</sup> for Greater Geneva) while providing accurate information at a small spatial resolution and with a reasonable computation time.

The main objective of this study is to validate a solar cadaster tool by using a comparison with the results from other numerical tools. To this purpose, the results from the solar cadaster are compared with those of two widely used tools: DIVA-for-Rhino and ENVI-met. The latter, thanks to its holistic model, makes it possible to deepen the assessment of the buildings' photovoltaic potential by evaluating their surface temperatures.

The validation of the solar cadaster involves three different tools based on different models and methods and with a varied field of applications. Thus, these tools are presented first, prior to introducing the methodology used. The results in terms of predicted shortwave irradiance on both the roofs and the facades are then detailed and discussed.

#### **2. Materials and Methods**

#### *2.1. Modeling Approaches for Radiative Transfers*

Two main methods are commonly used for the evaluation of radiative transfers: raytracing-based models or radiosity methods. Their principles, as well as their advantages and drawbacks, are described in the following.

#### 2.1.1. Ray-Tracing Methods

Ray-tracing methods consist of following the paths taken by electromagnetic rays. There are two types of ray-tracing approaches. Forward ray-tracing methods, also called light tracing, consist of launching rays from light sources in a set of directions. Backward ray-tracing methods consist of following the light path backward. In this latter case, rays are launched in a set of directions from the element of interest (the eye of an observer or an irradiated surface). A comparison between forward and backward ray tracing is illustrated in Figure 1.

**Figure 1.** Illustration of backward and forward ray tracing.

For both of these methods, the directions can be random (probabilistic approach) or set (deterministic approach). After rays have been launched, an intersection test is carried out between each of these rays and the objects in the scene; then, new rays are cast from the points of impact while considering the properties of the materials encountered (reflection, transmission, absorption). This method makes it possible to take the effects of light into account, such as refraction of light (for example, in the case of a scene comprising glasses). The behavior of a ray at the point of intersection then depends on the refractive properties

of the material encountered. This feature is particularly useful in the context of image synthesis or the study of transparent media with different refractive indexes.

This method also has the advantage of making the calculation of specular reflections possible (see Figure 2). On the other hand, it is not suitable for a calculation of diffuse reflections (which are usually the case for urban surfaces), as these require a significant calculation time because of the number of rays to be launched (in all directions) at each new intersection. This method has to be coupled with a diffuse radiation model.

**Figure 2.** Illustration of specular and diffuse reflection.

#### 2.1.2. Radiosity Method

The radiosity method is based on the calculation of the balance of radiosity, which is the total radiative flux exiting a surface (W m−2). This method allows the calculation of diffuse and reflected components of the irradiation only by solving a system of equations involving all the objects of a scene. Taking specular reflections (see Figure 2) into account is not directly possible with this method. However, unlike ray tracing, the calculations are independent of the position of the possible observer (surface of interest)—the calculation is generalized over the whole scene.

The radiosity method is based on solving the equation introduced by Kajiya [11] and detailed in the work of Sillion and Puech [12], where the exiting power is expressed as:

$$B(\mathbf{x}) = E(\mathbf{x}) + \rho \int\_{\Omega} L\_i(\mathbf{x}, \theta, \phi) \cos \theta d\omega \tag{1}$$

where *x* is a point in space, *B*(*x*) is the exiting power (emitted, reflected, and transmitted) at point *x* in space (in W m<sup>−</sup>2), *E*(*x*) is the power emitted by point *x* (source term, in W m<sup>−</sup>2), *ρ* is the coefficient of reflection of the surface, *Li*(*x*, *θ*, *φ*) is the incident power at point *x* from the solid angle determined by the angles *θ* and *φ*, Ω is the set of directions (*θ*, *φ*) of the hemisphere, and d*ω* is the elementary solid angle (sr).

This equation can be discretized:

$$B\_i = E\_i + \rho\_i \sum\_{j=1}^{N} F\_{ij} B\_j \tag{2}$$

where *Bi* and *Bj* are the power exiting the *i*-th cell and *j*-th cell, respectively, or the radiosity of the *i*-th or *j*-th cell (in W m<sup>−</sup>2); this term takes into account both the term emitted and the effect of the reflections. *Ei* is the power initially emitted by the *i*-th cell, *ρ<sup>i</sup>* is the reflection coefficient of the *i*-th cell, i.e., the fraction of the energy received that the cell returns (unitless value between 0 and 1), *N* is the number of meshes of which the scene is composed, and *Fij* is the form factor between the *i*-th and the *j*-th cells.

Solving the radiosity equation requires one to first know the source term *Ei* (known if the source emits directly), the reflexivity *ρi*, which is generally known as one of the characteristics of the materials, and the matrix of the form factors *Fij*, which must be calculated because it is dependent on the studied configuration.

#### *2.2. Tools Considered*

Three different tools were considered in this study: the solar cadaster (CadSol) for Greater Geneva (which is the one to be validated) and two widely validated tools—DIVAfor-Rhino (DIVA) and ENVI-met (EM). Their models and applications are described in the following.

#### 2.2.1. Solar Cadaster

As summarized by Desthieux et al. [10] and Freitas et al. [13], the solar cadaster tool is a geographic information system (GIS) tool. Such tools enable one to process large amounts of data and spatial analyses. These tools also provide automatic or systematic environmental analyses of urban areas, such as solar radiation calculations. These are different from tools that are classified in the category of computer-aided design (CAD). The latter process more accurate spatial and weather data in terms of spatial and time scales, but they require much more computing time and, thus, address the local scale (limited sets of buildings).

The irradiance received from the direct component *Ib* on an element of a surface at location *x* at a given time *t* is given by:

$$I\_b(\mathbf{x}, t) = BNI(t) \times r\_b(\mathbf{x}, t) \times S\_b(\mathbf{x}, t) \tag{3}$$

in which *BNI* corresponds to the direct normal component of the irradiance (also called the beam), *rb* is the transposition factor, and *Sb* corresponds to the shadow cast from a neighborhood building (*Sb*(*x*, *t*) = 0 if, at time *t*, the surface located at *x* is shaded by another building, or *Sb*(*x*, *t*) = 1 otherwise). The transposition factor, *rb*, depends on the solar elevation *h* and the slope of the considered surface *β* (*β* = 0 corresponds to a horizontal surface and *β* = 90◦ corresponds to a vertical surface; see Figure 3). It is calculated as *rb*(*x*, *t*) = *sin*(*h*(*t*))/*cos*(*β*(*x*)). The irradiance over the element of the surface is then considered as the average of the calculated irradiances at its edges. The calculation differences between the three tools, as described below, are illustrated in Figure 4.

**Figure 3.** Shadow casting of the solar cadaster, adapted with permission from [10,14].

**Figure 4.** Differences in terms of irradiance calculations for a mesh.

In order to model the contribution of the diffuse component to the received irradiance, the Hay model is used. This model considers two components: a circumsolar (anisotropic) and an isotropic component. Similarly to the direct component, the circumsolar component is calculated at each time step by considering the sun's position and the shadowing. For the isotropic component, the sky-view factor is computed. In the solar cadaster, a sky model of 580 light sources is used.

The diffuse component, *Id*, of the solar radiation is then calculated as follows:

$$I\_d(\mathbf{x}, t) = DHI(t) \times \left( \frac{GHI(t) - DHI(t)}{I\_0} r\_b(\mathbf{x}, t) + SVF(\mathbf{x}) \left( \frac{GHI(t) - DHI(t)}{I\_0} \right) \right) \tag{4}$$

where *GHI* is the global irradiation on a horizontal surface, *DHI* is the diffuse irradiation on a horizontal surface, *I*<sup>0</sup> is the hourly extraterrestrial irradiation, *rb* is the transposition factor as defined in (3), and *SVF* is the sky-view factor.

Finally, the reflected component is simply considered at the current stage as isotropic and is estimated based on Iqbal (1983) [15] as follows:

$$I\_r(\mathbf{x}, t) = 0.5 \times GHI(t) \times \rho(1 - \cos \beta(\mathbf{x})) \tag{5}$$

where *GHI* is as defined in (4), *ρ* is the coefficient of reflection of the surface, as defined in (1), and *β* is the slope of the surface.

More details about the calculation method and the computing tools used are available in [10,16].

#### 2.2.2. DIVA-for-Rhino

DIVA-for-Rhino is a highly optimized daylighting and energy modeling plug-in for Rhinoceros. This software uses ray-tracing and light-backwards algorithms based on the physical behavior of light in a 3D volumetric model. For hourly solar radiation, the Daysim interface is used. Daysim has been validated by several studies to be accurate in modeling visible-wavelength natural light for multiple sky conditions [17].

The daylight coefficient approach and the all-weather sky luminance model according to [18] are used here. In this approach, the irradiance received on an element of surface *x* is calculated as the sum of all sky segments visible from this element of the surface. For the diffuse component, 145 sky segments are used [19] concomitantly with three ground segments [20]. For the direct component, Daysim uses 65 sun positions; therefore, at a specific time *t*, Daysim will use one of the 65 positions that is closest to the real sun position at time *t*.

To calculate the complete set of daylight coefficients, two ray-tracing runs are performed:


The default Daysim simulation parameters were chosen. Up to two reflections from direct solar irradiation and one reflection from diffuse sky irradiation from the environment were considered.

Similarly to the solar cadaster, the irradiance calculated by Diva-for-Rhino for each surface element corresponds to the average of those calculated at its vertices.

#### 2.2.3. ENVI-Met

ENVI-met is a software aiming at simulating the urban microclimate by taking into consideration all of the phenomena that occur in an urban environment. It is based on coupled balance equations (including those of mass, momentum, and energy). This involves taking the built and natural environment into account.

The ENVI-met model has been widely used in numerous studies dealing with different issues, including the evaluation of outdoor thermal comfort [21,22], mitigation of the urban heat island effect [22,23], or assessment of buildings' solar and photovoltaic potential [24].

Regarding radiative transfers, ENVI-met uses a hybrid approach based on a radiosity method (see Section 2.1.2) for the evaluation of the irradiance with a deterministic raytracing method for the calculation of the view factors. As ENVI-met is under a proprietary license, access to its code is limited. However, some information is available. Regarding the calculation of the diffuse solar radiation, the model is isotropic. This means that there is no difference between the different sky segments and no dependency on the actual

position of the sun. Concerning the sky, it is divided into 414 segments (207 upward and 207 downward) [25].

Unlike the two other tools, the irradiances predicted by ENVI-met are given as that calculated at the center of the mesh.

#### *2.3. Case Study*

The present study focuses on two fictitious districts composed of three rows and three columns of buildings, i.e., a total of nine buildings. This ability of this kind of fictitious district has already been proven with respect to the study of the solar potential of neighborhoods [26–28]. In addition, it makes it possible to study the same district at different location, depending on the weather conditions used as input for the simulations.

These two districts, which are called the homogeneous and the heterogeneous district, have different arrangements, but they share some common urban morphology indicators, which are listed in Table 1.


**Table 1.** Urban morphology indicators of the homogeneous and heterogeneous districts.

Regarding the thermo-radiative properties of the different surfaces, the coefficient of reflection is set to 0.2 for both the ground's and the buildings' envelopes (vertical facades and roofs). These values are taken as a compromise between the typical values for concrete and soil.

#### 2.3.1. Study of the Influence of the Mesh Resolution on the Results

The mesh sensitivity was tested on a simple configuration—presented in Figure 5 which consisted of two buildings that were 25 and 35 m high and separated by 10 m. Three grid sizes of 0.5, 1, and 2 m square meshes were chosen. The daily cumulative irradiation on the south facade of the highest building is given in Figure 6. It appears that the cumulative irradiation predicted by the solar cadaster increased along with the coarseness of the resolution.

**Figure 5.** Geometry used for the study of the influence of the mesh.

**Figure 6.** Cumulative irradiation for different mesh resolutions.

The influence of the mesh resolution on the results is detailed in Table 2 for the months of February and August. The mesh resolution had an influence on the irradiation predicted for every tool considered. Indeed, the coarser the resolution, the higher the discrepancy in terms of cumulative irradiation. Nonetheless, DIVA, which is based on a ray-tracing method, was less influenced by the mesh resolution, unlike ENVI-met and the solar cadaster, which are based on radiosity methods.

**Table 2.** Divergence of the cumulative irradiation according to the resolution of the mesh.


On the other hand, the influence of the mesh resolution was more important in February than in August. This was due to the sun's lower course, which led to more shadowing.

In the particular case of the solar cadaster tool, a resolution of 2 m is, therefore, coarse and is not recommended for an analysis of irradiation in an urban environment. A resolution of 1 m remains acceptable for large areas, and the deviation from 0.5 m is small. This difference is explained by the fact that the obstacles' mutual shadows between the two plots will be better detected and considered in every facade point with a finer resolution.

A mesh composed of square cells with a resolution of 1 m was retained for the following study. This constituted a good compromise between the accuracy of the results and the computation time.

#### 2.3.2. Homogeneous Neighborhood

The homogeneous neighborhood was composed of nine identical buildings, which were 20 m wide and 30 m high, and each was separated by 20 m, as shown in Figure 7. In this case, the widths of the streets were the same, and so was the height-to-width ratio (Table 3) for all of the buildings. This configuration also made it possible to have the same sky-view factor regardless of the orientation of the vertical facade, as well as a sky-view factor that was equal to 0.5 for the roofs. The homogeneous neighborhood provided a simple configuration and made it easier to study the different radiative phenomena that occurred on the buildings' facades.

**Figure 7.** Sketch of the homogeneous neighborhood.

**Table 3.** Urban morphology indicators specific to the homogeneous district.


#### 2.3.3. Heterogeneous Neighborhood

As a counterpart used to make the study of radiative phenomena easier, the homogeneous neighborhood (see Figure 7) could not be fully representative of an actual district, as actual districts are generally composed of buildings of different heights or random positions. To bridge this gap, a heterogeneous neighborhood was considered as well. Its layout is given in Figure 8.

This district is representative of a common type of actual district because of the non-homogeneous spatial distribution of the buildings, as well as their different heights. The buildings' footprints and their total volume were kept constant, as were the urban morphology indicators given in Table 1. The urban morphology indicators specific to the heterogeneous district are given in Table 4.

**Figure 8.** Sketch of the heterogeneous neighborhood.


**Table 4.** Urban morphology indicators specific to the heterogeneous district.

#### *2.4. Weather Conditions for the Study*

The present study focused on two different days: 15 February, which had a short daylight period, and 16 August, which had a longer one. These two days are representative of a winter and a summer day for a location such as Geneva (cold and not very sunny for the first, hot and more sunny for the second). Furthermore, the sun's path corresponding to these two days was halfway between the solstices and the equinoxes.

The two days were actually those for which the sun's path was closest to the mean paths for the months. Indeed, in order to reduce the number of simulated days (two days to be simulated for the months of February and August instead of 59) while avoiding the specific conditions of a given day, this study considered two representative average days for the two considered months [10]. A representative average day (RAD) is defined, according to Equation (6), as a day for which the meteorological conditions (including irradiation level, temperature, and wind velocity and direction) for each hour are equal to the average of these conditions over all days of the month (*Nday*). Carrying out the study for two different months allowed us to evaluate the influence of the irradiation level on the accuracy of the predicted results.

$$X\_{RAD}(t) = < X\_i(t) >\_{N\_{day}} \tag{6}$$

where < ··· > stands for the time-average operator.

The two considered months had a double advantage. First, they allowed the validation of the values predicted by the solar cadaster under low and high irradiation levels (February and August, respectively). Second, they made it possible to evaluate the influence of the evolution of the sun's path over the year on the irradiance profiles.

Since the solar cadaster was developed for Greater Geneva, the fictitious districts studied were located at this place. The data used as input for the simulations came from the METEONORM® version 7.3 database for this city. The hourly averaged values issued from the METEONORM database were then averaged by month according to Equation (6) in order to reduce the computation time. Since the values used as input for the simulations were averaged by hour, the level of irradiation was considered constant between two consecutive hours for both the input and output data.

In this study, only the shortwave radiation (SW) was considered, as it is the most important energy content in the solar spectrum. Thus, the waveband considered was 2500 nm. The daily profiles of the different components of the solar radiation over Geneva for the representative days of the months of February and August are given in Figure 9. The direct shortwave radiation represents the amount of energy that comes straight from the sun, while the diffuse part is the amount of solar radiation reflected by the atmosphere prior to hitting the ground. The total shortwave radiation is the sum of the direct and the diffuse parts.

Although the overall evolution of the shortwave radiation is similar between February and August, differences are noticeable. Indeed, the maximum in terms of GHI is two times higher in August. This is mainly due to the direct part of the solar radiation, which exceeds 350 W m−<sup>2</sup> in August, in comparison with 150 W m−<sup>2</sup> in February.

**Figure 9.** Shortwave radiation used as input.

#### **3. Results**

The validation of the solar cadaster radiation model is a multi-step process. In the first step, the focus is on the horizontal surfaces of the homogeneous district because they are subject to a smaller number of phenomena. In the second step, the mean irradiance values over the vertical facades of the central building (see building 5 in Figures 7 and 8) are studied. Finally, the spatial distribution of the irradiance is analyzed.

#### *3.1. Mean Irradiance over the Unshaded Horizontal Surface*

The mean irradiance level over the roofs is given in Figure 10. For the homogeneous district (see Figure 7), the irradiance level is the same for the roofs of the nine buildings, since they are not subject to shading effects. Regarding the heterogeneous district, the irradiance level is the that of buildings 6 and 7 (see Figure 8), which are the highest buildings.

This comparison demonstrates the ability of the solar cadaster to accurately reproduce the basics of the solar conditions (including the beam horizontal irradiance (BHI) and the diffuse horizontal irradiance (DHI)). This first preliminary analysis allows us to ensure that the results of the modeling of the global horizontal irradiance (GHI) are the same for all tools. In other words, the difference that will be observed in what follows will be the consequence of the presence of shading and the inclination of the facades.

**Figure 10.** Spatial average of the irradiance received on the unshaded roofs (homogeneous neighborhood or buildings 6 and 7 of the heterogeneous neighborhood).

#### *3.2. Mean Irradiance over the Vertical Facades*

The goal of this study is to demonstrate the ability of the solar cadaster to provide accurate results for vertical facades. The results for the homogeneous district are given in Figures 11 and 12, while the results for the heterogeneous district are given in Figures 13 and 14. It appears that the values predicted by the solar cadaster are in good accordance with the values predicted by DIVA-for-Rhino and ENVI-met.

**Figure 11.** Spatial average irradiance received on the facades for the homogeneous district in February.

**Figure 12.** Spatial average irradiance received on the facades for the homogeneous district in August.

The different orientations of the facades make it possible to evaluate the prediction of the irradiance level for different solar conditions. Indeed, the north facade is mainly irradiated by the diffuse and reflected parts of the solar radiation, while the proportion of direct radiation is higher for the south facade. The east facade is directly irradiated in the morning, while the west facade faces the sun later in the day.

Regarding the level of irradiation, it does not have an influence on the concordance of the predicted values. Indeed, the results for the summer day are, in general, as accurate as those for the winter day. The differences between the results of the tools—quantified here with the normalized root mean squared error (NRMSE)—are not significantly impacted by the month of the year considered (see Table 5). The NRMSE values are calculated from differences between the daily profiles of the spatial average irradiance received on the facades and those predicted by the solar cadaster and ENVI-met on the one hand or by the solar cadaster and DIVA-for-Rhino on the other hand.

In both cases, the NRMSE remains lower for the roofs (less than 1 % for the roofs versus 4 % to 25 % for the vertical facades; see Table 5). This demonstrates the good accordance between the three tools with respect to the evaluation of the incident radiation on the roof, which are without a mask or reflection here. The observed differences for the vertical facades, which result in an increase in the NRMSE in the last four lines in Table 5, are then due to a difference in the consideration of masks and inter-building reflections.

**Table 5.** Normalized root mean squared error of the predicted level of irradiance (%) for the homogeneous district.


Nevertheless, the time of the year does not influence only the level of irradiance, but also the sun's path. Indeed, regarding the north facade (see Figures 11–14), the results from the solar cadaster are close to the those from DIVA-for-Rhino and ENVI-met (although a bit overestimated). Nonetheless, one peak can be observed in the results at 6:00 a.m. in the month of August (see Figure 12), which is not present for the same facade in February (see Figure 11). This difference in terms of the profile over time for the irradiance level is due to the evolution of the sun's path over the year. Indeed, during the month of August, the sun rises in the north-east and sets in the north-west, while it rises in the east and sets in the west in February.

The sun's path has an influence on the level of irradiation as well. Indeed, in the month of February (Figure 11), the south facade shows a slight decrease around noon, which is not present in the GHI (see Figure 9). This decrease is actually due to the shadow cast by building 2 on the central building (see Figures 7 and 8). This does not occur in August because the sun's path is high enough for the shading effect not to occur.

The results for the heterogeneous district make it possible to evaluate the ability of the solar cadaster to accurately predict the irradiation level for a more complex city. Indeed, the buildings are randomly located in this case (see Figure 8). The results for the two months considered are given in Figures 13 and 14.

It appears that the morphology of the neighborhood, although it is more random and complex than that of the homogeneous neighborhood, does not have a significant negative impact on the accuracy of the results predicted by the solar cadaster, which are in good accordance with those of DIVA-for-Rhino and ENVI-met (see Tables 5 and 6). Nonetheless, the level of solar irradiance over the different facades is lower in the case of the heterogeneous district than in the case of the homogeneous one. This can be explained by the fact that the central building is smaller than those around it and is thus subject to more of a shading effect.

**Figure 13.** Spatial average irradiance received on the facades in the heterogeneous district in February.

**Figure 14.** Spatial average irradiance received on the facades in the heterogeneous district in August.


**Table 6.** Normalized root mean squared error of the predicted level of irradiance (%) in the heterogeneous district.

#### *3.3. Irradiance Maps*

Although the results presented so far show a good accordance between the three tools, a consideration of the spatial mean value alone is not sufficient. Indeed, as shown in Figures 15 and 16, the range of the level of irradiance received over the facade can be very important.

The range of the predicted values may be due to the method of modeling radiative phenomena (ray-tracing or radiosity) or the shading effect. The latter may have an important impact on the level of irradiance. Thus, the study of the mean irradiance over the facades needs to be complemented with the study of irradiance maps.

**Figure 15.** Distribution of the irradiance over the vertical facades for the homogeneous district in February.

**Figure 16.** Distribution of the irradiance over the vertical facades for the homogeneous district in August.

The level of irradiance on unshaded roofs appears to be perfectly homogeneous and equal for the three tools considered, and it corresponds to the sum of the BHI and the DHI when given as the input of the simulation.

Nevertheless, the level of irradiance on the vertical facades depends on the orientation of the facade, the time of the day, and the time of the year. The levels of irradiance for the south facade of the central building of the homogeneous district are given in Figure 17. Figure 17a–c show the results for the month of February, while Figure 17d–f show the results for the month of August.

Regarding the data, not all are used for comparisons. All of the facades of the buildings are 20 m wide. However, the data at the edges of facades were not analyzed. Indeed, these data are difficult to compare because of the difference in terms of the calculation method for the three tools considered (depending on the tool, irradiance is considered as that at the center of the mesh or as the average over it); see Figure 4. This source of difference, due to the schemes specific to the tools, does not lead to a significant difference in the results, except at the edges of the facades.

**Figure 17.** Irradiance map of the homogeneous district for the south facade over 12 h.

The irradiance levels appear to be more homogeneous on the south facade for the month of August than for the month of February. This can be explained by the sun's higher path at this time of year. The greater range in terms of value (see Figures 15 and 16) is then due to the shading effect, which is more intense in February and leads to a range of irradiance of 50 W m−<sup>2</sup> over the south facade in August in comparison with 350 W m−<sup>2</sup> in February.

Regarding the level of irradiance predicted by the solar cadaster, which is shown in Figure 17a,d, the values are similar to those predicted by ENVI-met and DIVA-for-Rhino.

Nevertheless, although the predicted irradiance levels are close to each other, there are two particular points. First, the shape induced by the shading is not exactly the same for the solar cadaster and ENVI-met on the one hand and for DIVA-for-Rhino on the other hand. The latter predicts a smoother evolution of the predicted irradiance level. This smoothness is due to the method used for the evaluation of the radiative phenomena; DIVA-for-Rhino uses ray-tracing while ENVI-met and the solar cadaster use the radiosity method. The second point is how the irradiance is evaluated. ENVI-met considers the irradiance to be that at center of the mesh, while DIVA-for-Rhino and the solar cadaster consider it to be the average of the irradiance over the mesh. In the case of DIVA-for-Rhino, it reinforces the smoothness of the evolution of the irradiance over the facade. Concerning the solar cadaster, this has an impact at the edge of the shade (the horizontal purple line at the height of 17 m in Figure 17a).

#### **4. Discussion**

The results presented so far show a general good accordance between the three tools considered in terms of both the mean values the irradiance maps. Nevertheless, two points must be noted:


Regarding the sensitivity of the predicted irradiance, the highest discrepancies between the results of the tools appear at the first or the last hour of sunshine (see the north and the west facades in Figure 12). These discrepancies are due to a lack of precision between the calculation of the sun's path and the level of irradiation. This shift has only a slight incidence on the prediction of the irradiance during the day, but sees its influence strongly increase when the sun is close to the horizon. A sunrise that is predicted too early in the morning results in an overestimation of the predicted irradiance at this time of day, while a sunset predicted too late in the evening results in a peak of the predicted irradiance before the dusk.

Figure 17 shows the existence of a time lag between the three tools studied. Indeed, the shapes of the shadow on the south facade are different among the results of the solar cadaster (Figure 17a), ENVI-met (Figure 17b), and DIVA-for-Rhino (Figure 17c). This offset has only a slight influence on the results. Nonetheless, this impact may increase incrementally, especially early in the morning or late in the evening, as mentioned before (see the north facade in Figure 12).

#### **5. Conclusions**

The solar cadaster has proven its ability to provide accurate results in terms of the level of irradiance on both roofs and vertical facades. Indeed, the results predicted by the new version of the solar cadaster developed for Greater Geneva are in good accordance with those predicted by ENVI-met and DIVA-for-Rhino.

Regarding the concordance of the results, the non-shaded horizontal surfaces match almost perfectly. This is not the case for the shaded facades and vertical facades, but the results are very satisfactory. The differences observed between the tools come from the differences between the models in terms of diffuse radiation, the evaluation of reflections, or shading.

Finally, the solar cadaster provides precise results at a low spatial resolution while keeping the computation time low. It ranges from a few hours for the solar cadaster and DIVA-for-Rhino to a few days for ENVI-met. It can therefore be used on a large scale and can prove to be a reliable tool for promoting and intensifying the use of solar energy on the scale of Greater Geneva. However, work is in progress to improve the modeling of reflected components and, thus, the reliability of the tool for more complex geometries.

**Author Contributions:** Conceptualization, B.G., M.T., G.D., C.M. and S.G.-J.; Methodology, B.G., M.T. and G.D.; Software, V.G. and G.D.; Resources, G.D.; Writing—original draft preparation, B.G.; Writing—review and editing, B.G, M.T., K.B., É.P., S.G.-J., C.M. and G.D.; Visualization, B.G.; Supervision, C.M. and G.D.; Funding acquisition, C.M. and G.D.; All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by EU-INTERREG V France-Suisse program (G2-SOLAIRE Ref: 4610), and received support by the French National Research Agency, through Investments for Future Program (ref. ANR-18-EURE-0016—Solar Academy).

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** The data that supports the findings of this study are available from the corresponding autho upon reasonable request.

**Acknowledgments:** The authors would like to thank the INTERREG V Suisse–France program for providing financial support for conducting this study in the framework of the G2 Solar project, which aims to extend the solar cadaster to Greater Geneva and intensify solar energy production at this level. This work was supported by the French National Research Agency through the Investments for Future Program (ref. ANR-18-EURE-0016—Solar Academy). The research units LOCIE and FRESBE are members of the INES Solar Academy Research Center.

**Conflicts of Interest:** The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.

#### **References**

