Enhancing Transparency of Climate Efforts: MITICA’s Integrated Approach to Greenhouse Gas Mitigation

Abstract

Under the Paris Agreement, countries must articulate their most ambitious mitigation targets in their Nationally Determined Contributions (NDCs) every five years and regularly submit interconnected information on greenhouse gas (GHG) aspects, including national GHG inventories, NDC progress tracking, mitigation policies and measures (PAMs), and GHG projections in various mitigation scenarios. Research highlights significant gaps in the definition of mitigation targets and the reporting on GHG-related elements, such as inconsistencies between national GHG inventories, projections, and mitigation targets, a disconnect between PAMs and mitigation scenarios, as well as varied methodological approaches across sectors. To address these challenges, the Mitigation-Inventory Tool for Integrated Climate Action (MITICA) provides a methodological framework that links national GHG inventories, PAMs and GHG projections, applying a hybrid decomposition approach that integrates machine learning regression techniques with classical forecasting methods for developing GHG emission projections. MITICA enables mitigation scenario generation until 2050, incorporating over 60 PAMs across Intergovernmental Panel on Climate Change (IPCC) sectors. It is the first modelling approach that ensures consistency between reporting elements, aligning NDC progress tracking and target setting with IPCC best practices while linking climate change with sustainable economic development. MITICA’s results include projections that align with observed trends, validated through cross-validation against test data, and employ robust methods for evaluating PAMs, thereby establishing its reliability.

1. Introduction

The Paris Agreement [1] established the Enhanced Transparency Framework (ETF) with the aim to foster trust among Parties. Operationalised through the modalities, procedures, and guidelines (MPGs) [2], the ETF sets extensive reporting requirements, enabling effective tracking of progress toward the Agreement’s objectives. Under the ETF, countries are required to submit biennial transparency reports (BTRs) every two years, with the first due by 31 December 2024. BTR content includes different information pieces related to greenhouse gases (GHG), including a national GHG inventory, GHG projections, information on mitigation policies and measures (PAMs), and information to track progress of the Nationally Determined Contribution (NDC). Parties are expected to report ‘with measures’ (WM) and may report ‘with additional measures’ (WAM) and ‘without measures’ (WOM) projections, allowing to assess impact of national PAMs into the future GHG emission profile [3].

In addition to BTRs, Parties are required to submit successive updated NDCs every five years, representing progression compared to the previous NDC and reflecting the highest possible ambition, especially in the form of mitigation targets. These submissions, initially due by 2020 and every five years thereafter, require information necessary for clarity, transparency, and understanding (ICTU), encompassing quantifiable details on reference points, time frames, scope, methodological approaches, and fairness of NDCs [4]. Despite synergies between reporting elements such as national GHG inventories, PAMs, projections, NDC updates, and NDC tracking, consistency issues and primary difficulties in periodically producing and reporting on these elements are anticipated. This is particularly attributed to data collection, lack of national expertise and weak institutional systems [5].

To further elaborate on the consistency issues and primary difficulties mentioned earlier, several key aspects need consideration. These include the inconsistency observed between national GHG inventories and projections, as highlighted by [6]. Additionally, a critical issue is the disconnection between PAMs and mitigation scenarios, as discussed by [7]. Furthermore, the utilisation of inconsistent methodological approaches across different sectors leads to a lack of clarity regarding aggregated emissions and mitigation targets, resulting in increased uncertainty—a concern emphasised by [8,9].

The literature also identifies a significant challenge related to the insufficient capacity in developing countries to employ complex modelling approaches for producing GHG projections in different mitigation scenarios, exacerbated by the absence of common guidelines to produce GHG emission projections [10,11,12,13].

To effectively address these challenges in developing mitigation scenarios for NDC design and tracking requires careful consideration of specific key elements. Firstly, there is a necessity for consistency between various GHG elements, namely national GHG inventories, PAMs, mitigation scenarios, and mitigation targets [13,14]. Ensuring alignment with the Intergovernmental Panel on Climate Change (IPCC) methodologies and nomenclatures [15,16] is crucial for maintaining standardised and comparable reporting practices, while adopting coherent and informed policy decisions.

Moreover, the adopted approaches should have the flexibility to accommodate limited data availability, recognising the resource constraints often faced by developing country Parties [17]. This adaptability is essential for enabling a broader spectrum of countries to effectively participate in both climate action and the ETF reporting process. Additionally, the utility of these approaches extends beyond mere reporting; they should facilitate the establishment of mitigation targets and streamline the tracking of mitigation efforts [18]. A robust framework that meets these requirements would not only enhance the transparency and comparability of reporting but also empower developing countries to actively contribute to the global efforts outlined in the Paris Agreement.

The current landscape of models and tools for developing mitigation scenarios falls short of meeting these criteria. Consequently, the proposed Mitigation-Inventory Tool for Integrated Climate Action (MITICA) aims to bridge these gaps by leveraging existing IPCC methodologies and expertise in developing national GHG inventories. MITICA’s objective is to align national GHG inventories with GHG projections and PAMs, thereby facilitating NDC progress tracking, and harmonising mitigation planning. Through its innovative framework, MITICA endeavours to empower stakeholders with the tools and insights necessary to navigate the complexities of climate mitigation and contribute meaningfully to the objectives outlined in the Paris Agreement. This paper presents MITICA as a comprehensive solution to the identified challenges. It outlines the framework’s structure, emphasising its role in reducing inconsistencies, enhancing transparency, and empowering countries to align their mitigation targets effectively. By integrating national GHG inventories, projections, and mitigation scenarios, MITICA provides a systematic and harmonised approach to mitigation planning that not only enhances the accuracy and reliability of reporting but also fosters a more coordinated and effective approach to climate action on a global scale.

The paper is organised in the following sections. Section 2 describes the material and methods of the study, before the results and discussion are presented in Section 3. Finally, in Section 4 the conclusions of the paper are discussed, delineating the main insights and avenues for future work.

2. Matherials and Methods

An extensive literature review in Appendix A discusses challenges in developing mitigation scenarios, approaches used by developed and developing countries for elaborating such scenarios, and relevant studies on GHG forecasting. From the assessment of developed Parties’ submissions outlined therein, it is observed that most adopt sector-specific bottom-up models built from national GHG inventory methodologies and use them in conjunction with macro top-down models incorporating exogenous drivers that characterise their respective national economies. A notable drawback identified in this approach is the substantial resources, including time, personnel, and budget, required for generating distinct mitigation scenarios for each IPCC category and sector. This is attributed to the considerable human interventions necessary in model production. As a result of this observation, the objective of MITICA is to standardize a framework that enables any country, with a particular focus on developing nations, to formulate specific bottom-up mitigation scenarios specified at the IPCC category level by the country, combined with a top-down specification of their national economy. Additionally, MITICA draws inspiration from the main modelling alternatives utilised by developing countries, offering an extensive list of possibilities for PAMs, and developing sector-specific modelling approaches. Considering various alternatives discussed in the literature [19,20], the most suitable option to meet the study’s requirements is statistical frameworks with the flexibility to accommodate diverse sector-specific models tailored to different circumstances and data availability, while maintaining overall consistency across sectors, scenarios, and time periods. Under these considerations, MITICA aims at establishing both a framework and a tool to create consistent mitigation scenarios that can be tracked against historical GHG emission trends and used as a benchmark to define and implement relevant interventions when needed. The following subsections describe MITICA’s general framework, its forecasting approach, the accountability for PAMs, and the software characteristics.

2.1. General Framework

MITICA’s conceptualization and development adhered to specific requirements. Aligned with its objectives, it followed IPCC nomenclatures, ETF definitions, and UNFCCC inventories as primary data inputs. The framework is designed to be universally applicable, offering a standardised methodology to overcome identified challenges effectively. It utilises national GHG inventories at the highest disaggregation level, mirroring their detailed structure to enhance model specification. MITICA employs a consistent modelling framework for all IPCC sectors to minimise inconsistencies, while still being emission source and country specific.

MITICA’s goal is to address GHG emission sources and sinks comprehensively resulting from the implementation of various PAMs within user-defined macroeconomic and sectoral frameworks. The modelling approaches therefore consider the evolution of proxies, encompassing macroeconomic, demographic, and sectoral drivers across various scopes, influencing country-level GHG inventory methodologies. While MITICA’s outcomes are not predictive, they serve to scientifically assess policy alternatives and derive potential mitigation targets. This aids both developed and developing countries in designing and tracking NDCs within the ETF of the Paris Agreement, as well as to assist in reporting to the UNFCCC.

Considering these requirements, MITICA develops mitigation scenarios starting with the estimation of a WOM scenario. This scenario represents projected national GHG emissions considering a set of projected proxies ceteris paribus; only the proxies change in the projected years, being the technology mix, consumer behaviour as well as the GHG accounting methodologies the same of the latest historical year; these elements will only change as a result of the implementation of PAMs. Indeed, MITICA uses the WOM as a benchmark for developing mitigation scenarios (WM and WAM, in line with ETF definitions). In these scenarios the only difference concerns the PAMs implemented and their impact on GHG emissions. Further information on the design of the forecasting approach selected for projecting WOM emissions in MITICA is provided in Section 2.2.

The input data required by MITICA for estimating the WOM scenario is the national GHG emission inventory and a set of projected proxies. MITICA’s estimations are made at the highest disaggregation level available in the national GHG emission inventory. Inventory estimations can start from year 1990 and are provided on an annual basis. However, the flexibility provisions of the ETF allow developing Parties to develop and submit national GHG inventories with starting years other than 1990 [2]. Countries facing higher capacity constraints may only be able to develop and report limited time series, as can be observed in the biennial update reports (BURs) submitted until 2023 [21]. Despite shorter national GHG emission inventory time series, developing countries are required to develop and submit NDCs including mitigation targets, and also report on the tracking of progress of the NDC every two years. Furthermore, the completeness and the level of detail of national GHG emission inventories may vary due to the different methodological approaches that are allowed by the 2006 IPCC Guidelines. In this context, MITICA develops mitigation scenarios even with limited inventory time series and sectoral disaggregation. However, the quality of results is substantially improved with longer time series and higher sectoral granularity levels.

The proxies needed by MITICA to estimate the WOM scenario GHG emission consist of two layers: a first layer of national-level proxies needed to develop GHG projections, and a second layer of sector-specific proxies aimed at refining sectoral model specifications. Despite the forecasting approach is common for all sectors, different sectoral proxies enable MITICA to define national-specific sectoral models to project WOM emissions.

These proxies have been selected considering: (i) data availability, prioritising proxies with generally accessible and comprehensible data; and (ii) the theoretical relationship between sectoral emissions and the proxy, grounded in relevant research.

Table 1. Sectoral proxies in MITICA.

Granularity 1	Proxy	Theoretical Relationship
All sectors	Gross Domestic Product (GDP)	GDP involves activity levels in the different inventory emission sources. Increasing GDP generally involves increasing emissions.
All sectors	Population	Increased population levels generally lead to increasing emissions.
Energy	Energy demand	Energy demand is directly related to increased fossil fuel emissions in the absence of technological changes.
Energy	Fuel prices	Increase in fuel prices generally result in a reduction in fuel consumption in the medium term.
Energy	Energy supply	The amount of energy supplied is directly correlated with sectoral emissions. An increase in energy supply contribute to higher emissions.
Transport	Fleet	A larger fleet, particularly if dominated by vehicles with higher emission profiles, tends to contribute to increased emissions.
Transport	Vehicle kilometre travelled	The total distance travelled by vehicles is positively associated with sectoral emissions. Higher vehicle kilometre travelled result in increased fuel consumption and emissions.
Fugitive emissions	Solid fuel production activity levels	The level of activity in solid fuel production is directly linked to emissions from the use of solid fuels.
Fugitive emissions	Oil production levels	Oil production levels have a direct impact on sectoral emissions.
Fugitive emissions	Natural gas production levels	The levels of natural gas production are positively associated with sectoral emissions.
Industrial Processes and Product Use—IPPU	Industrial activity index	Industrial activity levels lead to increased emissions in the absence of changes in technologies.
Industrial Processes and Product Use—IPPU	Income indicator	Income levels are correlated with consumption patterns, particularly on products use.
Agriculture	Crop activity index	An increase in crop activity levels lead to increased emissions from agriculture in the absence of changes in practices or technologies.
Agriculture	Livestock activity index	Increased livestock population produce increased emissions from agriculture.
Land Use, Land-Use Change and Forestry – LULUCF	Forest land cover growth	Increased forest land involves increased CO2 removals, therefore reduced net GHG emissions.
Land Use, Land-Use Change and Forestry – LULUCF	Degree of conservation	Increased forest trends lead to enhanced biomass growth and subsequent CO2 removals, therefore reduced net GHG emissions.
Other sectors	Service activity index	Increased service activity may contribute to higher energy consumption and emissions associated with the provision of services.
Other sectors	Households	Increasing households’ size leads to increased household energy consumption and emissions.
Other sectoral proxies	-	-

shows the main proxies considered, describing the theoretical relationship between variables building from [22,23].

Starting from an initial WOM projection, MITICA develops generic methodologies, building from [24], to estimate the impact of relevant PAMs on main emission sources and sinks. The approach and foundations of PAMs estimations are further elaborated in Section 2.3. WM and WAM scenarios are then constructed by considering the impact of user selected PAMs. By considering different sets of PAMs, policy makers can visualise the potential impact of implementing policies of interest into the national emission profile. This type of mitigation assessment can support in establishing informed mitigation targets and assessing their potential evolution under specified mitigation scenarios and a given macroeconomic framework. Figure 1 illustrates the generalised procedural steps of MITICA, which are subsequently elaborated upon below.

A more detailed graphic workflow of MITICA is provided in Appendix C for further reference.

2.2. Methodology for Projecting GHG Emissions in the WOM Scenario

The methodology selection for projecting the WOM scenario within MITICA builds from the revision of the literature of time series forecasting.

Table 2. Selection of similar studies * for time series forecasting.

Study	Modelling Approach	Application
[25]	Autoregressive integrated moving average (ARIMA)	Energy consumption and GHG emissions from pig iron manufacturing in India.
[26]	Seasonal Autoregressive Integrated Moving average with Exogenous Factors (SARIMAX)	Forecast of short-term hourly electricity generation.
[27]	ARIMA & SARIMAX	Forecasting natural gas production and consumption in United States until 2025 on monthly basis.
[28]	Random Forest Regression model.	Forecasting CO2 emissions at city level in China.
[29]	Random Forest Regression model.	Generation capacity forecasting of cascade hydropower stations
[30]	Random Forest Regression model with Slime Mould Algorithm.	Forecasting of CO2 emissions from road transport.
[31]	Long-short Term Memory (LSTM) neural network compared with Least Squares Support Vector Machine and recurrent neural network.	Forecasting of NOx emissions from thermal power plant.
[32]	Three methods are applied, the ARIMA model, the SARIMAX model and the LSTM model.	Forecast of CO2 emissions in India.
[33]	Least Squares Support Vector Machine	Projection of thermal comfort, CO2 emission and economic growth.
[34]	Empirical mode decomposition and evolutionary least squares support vector regression.	Carbon price using EU ETS for years 2013–2016.
[35]	Least Squares Support Vector Machine.	CO2 emissions of Hebei using a time series for years 1990–2016
[36]	6 different machine learning models, including GBR	Forecasts for solar radiation in daily and hourly timescales.
[37]	GBR tree and principal component regression models.	Forecast of electricity prices in Spain.
[38]	GBR combined with Random Forest Regression.	Prediction of net ecosystem carbon exchange using data from two sites for years 1997, 2010, 2012 and 2013.
[39]	Artificial Neural Network (ANN)	Forecast of the heating and cooling energy demands, energy consumptions and CO2 emissions of office buildings in Chile using a dataset of 77,000 data points.
[40]	Artificial Neural Network	Annual forecasts of CO2 emissions for 17 countries.
[41]	Comparison of several regression techniques, including Least Absolute Shrinkage and Selection Operator (LASSO).	Forecasting of long-route CO2 emission from shipping using 40 data points.
[42]	LASSO-Deep Belief Networks (DBN)-Bootstrap Model.	Long term streamflow forecasting using monthly data for the period 1956–2015.
[43]	Three different models are used, including Grey Model GM(1,N), ANN and LASSO.	Short-term forecasting annual CO2 emissions in Malaysia.
[44]	The LASSO model is compared to several shallow models.	China 2022–2027 forecasting of CO2 emissions using a dataset from 2011 to 2021.

* In the table, studies of distinct nature are distinguished through the incorporation of ticker lines for easy reference.

shows the methods followed by a selection of studies of similar nature for time series forecasting.

When prioritising the key requirements for MITICA’s methodology design, previous research was examined to assess the applicability and optimal performance of models. The primary criterion is the model’s efficacy in projecting GHG emissions over extended time periods, ensuring its appropriateness for long-term forecasting. A secondary consideration involves the model’s proficiency in managing small datasets, recognising the inherent limitations associated with restricted time series data. Additionally, the model’s capacity to integrate external proxies assumes critical importance for capturing external factors that influence emissions. Lastly, the requirement for flexibility underscores the necessity for the model to be adaptable and tailored to the specific contexts of different countries.

In light of these criteria, Seasonal Autoregressive Integrated Moving Average with Exogenous Factors (SARIMAX) distinguishes itself for its ability to incorporate external regressors and manage limited data, while Least Absolute Shrinkage and Selection Operator (LASSO), Gradient Boosting Regression (LASSO), Gradient Boosting Regression (GBR) and Random Forest Regression have also proven advantageous for their aptitude in capturing intricate patterns and offering flexibility across diverse datasets and geographical contexts. Furthermore, the comparative performance of these methods has provided good results in previous studies [27,32,38]. In contrast, deep learning techniques such as the Long Short-Term Memory (LSTM) models may encounter challenges with small datasets and interpretability [44], while ARIMA and Least Squares Support Vector Machine may face difficulties in accommodating external proxies and adapting to the distinct contexts of different countries [45].

Based on this assessment, MITICA incorporates projection modelling approaches better suited for small data sets, long-term forecasting, and considering exogenous drivers. Consequently, a hybrid approach, named Artificial iNtelligeNce And cLassIcal STatistics (ANNALIST), is developed for projecting the WOM scenario. This hybrid model integrates LASSO, SARIMAX, and Random Forest Regression, leveraging the primary advantages of the forecasting methods investigated in GHG forecasting literature. MITICA also offers alternative but suitable methods, empowering users to select the preferred option based on the characteristics of the dataset and enabling the maximisation of available data utilisation while mitigating potential limitations associated with various modelling approaches. The model initiates by decomposing the trend and noise, a commonly employed practice in GBR [38]. The Exponentially Weighted Moving-Average (EWMA) algorithm is applied to derive the trend, similar to the approach taken by authors in [46]. Subsequently, the noise is obtained through subtraction. The selection of the trend-noise tuple is based on demonstrating stationarity noise, assessed using the Augmented Dickey Fuller test, and maximal standard deviation.

Assuming a time series $Y_t=T_t+R_t$ where $Y_t$ represents the value at time t, $T_t$ denotes the trend, and $R_t$ the noise, the variance is formulated as:

$$VAR\left(Y_t\right)=VAR(T_t+R_t)$$

(1)

Under the assumption of independent noise and trend, it is posited that:

$$VAR\left(Y_t\right)=VAR\left(T_t\right)+VAR\left(R_t\right)$$

(2)

Further assuming that $VAR\left(Y_t\right)=C$ due to its constancy, and $VAR\left(R_t\right)=nVAR\left(T_t\right)$:

$$C=VAR\left(T_t\right)(1-n)\to VAR\left(T_t\right)=\frac{C}{(1+n)}$$

(3)

This reveals that as n increases, reflecting a larger $VAR\left(R_t\right)$, $VAR\left(T_t\right)$ decreases. This drives the rationale for selecting noise with a higher standard deviation. SARIMAX is employed to capture intricate patterns, functioning optimally in the presence of stationary values. In such cases, SARIMAX predicts the noise seamlessly, and simultaneously, the trend remains simplified, thus enhancing predictive accuracy. In instances where noise does not meet these criteria, zero noise is considered, treating the entire dataset as the trend. Utilising regression techniques on a pool of potential variables for model development often results in overfitting, characterised by an excessive inclusion of variables in the final model and an overestimation of its performance [47]. To tackle this issue in trend prediction, the LASSO model is employed [48]. LASSO determines the best fit model specification by IPCC category, considering the proxies inputted by the user. As a default setting, in order to offer a priori information for various models by sector, ANNALIST includes a prioritisation weight for the proxies. This weight assigns greater importance to the parameters of sectoral drivers that have demonstrated better performance within their respective sectors, as outlined by [23].

For noise treatment and assembling, ANNALIST applies a first difference in the presence of non-stationarity by applying the Augmented Dickey Fuller test. Once stationarity is ensured, the following SARIMAX specification is applied: $(p,0,q)(0,0,0,0)$.

Following the addition of trend and noise predictions, a series of automated corrections employing machine learning is applied throughout the process. The initial step involves outlier filtering using the Isolation Forest model [49], a machine learning tool designed for outlier detection. Subsequently, a Random Forest Regressor model [38] is employed to train on historical data and update predictions. A comprehensive hyperparameter optimisation is then conducted using Grid Search CV [50], a machine learning technique seeking the optimal parameter combination to enhance model accuracy. This optimisation is reinforced with time series cross-validation, ensuring the model’s robustness and generalisability across different temporal datasets. Thus, a reliable model is obtained, maximising the strengths of SARIMAX while simplifying the trend in the most robust manner. Subsequently, the model undergoes various machine learning analyses to ensure the maximum likelihood of the outcome. The graphic workflow of ANNALIST is provided in Appendix B.

To enable the application of MITICA across various input datasets and time series, alternative approaches to ANNALIST for projecting the WOM scenario are integrated. This includes a GBR model without decomposition and the SARIMAX method. GBR has proven particularly useful for very short time series (from 2 observation years), making it valuable for countries with limited time series data. SARIMAX yielded results similar to ANNALIST, except for linear trend functions characterised by the consolidation of all noise variation predominantly at the upper end of the series. Therefore, it is incorporated for cross-comparison purposes. MITICA requires users to validate the forecasts produced for the WOM scenario by IPCC category, incorporating the possibility to modify the model specified by ANNALIST by changing the type of modelling approach from ANNALIST to GBR or SARIMAX.

Regarding the software aspect, MITICA has been deployed in a desktop application using Phyton as the main programming language. By adopting Python, MITICA ensures compatibility with various operating systems, including Windows, macOS, and Linux, making it accessible to a broader user base. This cross-platform compatibility enhances the usability and accessibility of MITICA, allowing users from different backgrounds to leverage MITICA’s features for GHG forecasting and mitigation analysis. Furthermore, MITICA’s deployment as a desktop application offers several practical benefits. For instance, users can run MITICA locally on their computers, ensuring data privacy and security. Moreover, being a standalone application, MITICA does not require an internet connection to function, providing users with uninterrupted access to its features and functionalities, regardless of their location or internet connectivity.

2.3. Mitigation Impact of PAMs and Definition of WM and WAM Scenarios

The PAMs accounting approach of MITICA extends the methodological framework described in [24] to encompass all IPCC sectors and main mitigation alternatives, aligning with the principles and requirements described in Section 2.1. The methodological framework outlined in [24] has already undergone testing and its estimates have been included in the National Energy and Climate Plan of Greece [51], proving its applicability in the context of the study. The basic estimation approach is depicted as:

$$M{E}_{t_i-t_f}=R·M_{t_i-t_f}[RE{F}_t-ME{F}_t]$$

(4)

where $M{E}_{t_i-t_f}$ represents the mitigation effect of the PAM for the entire projected period, $M_{t_i-t_f}$ is the magnitude of the PAM representing the affected activity levels, R represents the reduction factor in magnitude from PAM implementation, $RE{F}_t$ stands for the reference emission factor in the absence of the PAM at time t, and $ME{F}_t$ is the mitigation emission factor post implementation of the PAM at time t. Based on this generalisation, PAM methodologies are specified case-by-case, inked to the reference national GHG inventory through the REF, and associated with the WOM scenario through $M_{t_i-t_f}$.

Building from this conceptual framework, an extensive list of PAMs (Appendix D) is available within MITICA, providing default factors and specific methodologies covering all emission sources and sinks defined by the IPCC Guidelines [15,16]. In the final tool, users are required to define the magnitude of the desired PAM and adjust, if necessary, any of the methodological parameters. Once the list of PAMs is defined, MITICA aggregates the individual impact assessment of PAMs to produce the WM and WAM scenarios, thereby allowing to define scenarios based on national circumstances and stakeholder agreements. Figure 2 shows MITICA´s rationale to account for the impact of PAMs from the estimation of the WOM scenario.

3. Results and Discussion

The projection of the WOM scenario constitutes a critical step in developing mitigation scenarios, as it creates the benchmark against which the impact of PAMs is evaluated. The robustness f the results provided by MITICA for the WOM are analysed by applying different projection methods on historical datasets to ascertain the modelling approach providing best results compared to real data. Yearly datasets spanning a typical historical data length (approximately 30 years since UNFCCC data collection starts in 1990) plus the maximum prediction range (year 2050) are selected for analysis. Three datasets from the World Bank database are used: total energy use [52], goods export data [53] and use of alternative and nuclear energy [54]. The time period from 1960 to 2022 is categorised into historical (1960–1990) and projection (1991–2022) periods. The forecasting process considers GDP [55] and population [56] as proxies. These datasets serve as plausible representative examples of the data that MITICA will process, and the results can be compared to observed values. The selection of countries for the analysis is driven by data availability, and includes Australia, Finland, France, Greece, Italy, Luxembourg, Netherlands, Norway, Portugal, Spain, Sweden, Turkey, the United Kingdom, and the United States. Figure 3 shows an extract of the results obtained for four countries.

The brown line depicts the actual observed data, juxtaposed with alternative forecasting methodologies within the MITICA framework, namely ANNALIST, GBR, and SARIMAX. Additionally, two methods are included for comparative analysis: a linear regression forecast and a linear extrapolation utilising the annual growth rate of the latest time series value (AG Forecast). AG Forecast serves as a metric for evaluating the performance of these methods, observing that the annual growth deviates further from the actual values. The examination of the figures reveals that, in general, ANNALIST and SARIMAX exhibit results closest to the real data, effectively capturing observed trends. The deviation of each model’s results from actual data is quantified in

Table 3. Deviation between the actual mean and the predicted mean of different models in the tested datasets.

	Linear Regression	Annual Growth	GBR	SARIMAX	ANNALIST
Goods export data	29%	127%	22%	28%	19%
Total energy use	19%	52%	25%	18%	16%
Use of alternative and nuclear energy	52%	-	56%	57%	44%
Mean	33%	-	34%	34%	26%
Average computing time (s)	0.038	0.013	0.05	52	0.97

, serving as a metric for assessing the performance of each modelling approach.

The average deviation, considering results across all countries, further indicates that ANNALIST, with its modelling approach, yields outcomes closest to the actual data. This underscores the proficiency of MITICA in generating scenario projections. The validation of PAMs has been analysed broadly in several studies [57,58,59], offering supporting evidence for the reliability of MITICA’s outcomes in developing WM and WAM scenarios based on WOM results.

In addition to the earlier evaluations, the beta release of MITICA underwent testing with a variety of input datasets. These datasets encompassed the Tajikistan national GHG inventory sourced from [60], confidential information provided by Uruguay regarding its inventory, and a set of simulated databases from the IPCC software [61]. The testing primarily focused on assessing the functionality of the software and identifying any potential bugs.

4. Conclusions

Previous research has highlighted gaps and challenges in generating mitigation scenarios and ensuring consistent reporting under the ETF across various facets of GHG emissions. This encompasses areas such as national GHG emission inventories, the impact assessment of PAMs’ impact, GHG emission projections, and NDC design and tracking, particularly for developing parties within the Paris Agreement.

An evaluation of existing models and approaches revealed several insights: (i) developed countries commonly employ a combination of sectoral models for each IPCC sector coupled with a top-down macroeconomic framework; (ii) alternatives used by developing countries exhibit strengths in assessing individual PAMs and developing sectoral models but show limitations in integrating all IPCC sectors and assessing PAMs within mitigation scenarios; (iii) diverse time-series innovative forecasting methods, applied in prior studies, offer applicability to address the study’s challenges.

Building upon this foundation and introducing an innovative approach utilising machine learning regression techniques for GHG forecasting, MITICA successfully addresses identified goals. It establishes an integrated methodological framework for mitigation scenario production, ensuring consistency between national GHG emission inventories, PAMs, and projections. This allows for the transparent generation of mitigation scenarios, thereby facilitating NDC design and tracking, as well reporting under the ETF.

The MITICA approach has been subjected to comprehensive testing in three distinct phases. Initially, projections were computed utilising data spanning from 1960 to 1990 as a baseline, forecasting variables such as total energy use, goods exports, and the utilisation of alternative and nuclear energy for the period spanning 1991 to 2022. The exogenous variables GDP and population were incorporated into the analysis, computed for Australia, Finland, France, Greece, Italy, Luxembourg, Netherlands, Norway, Portugal, Spain, Sweden, Turkey, the United Kingdom, and the United States. A comparison of the MITICA projections against observed values from 1991 to 2022 reveals a cumulative deviation throughout the projected time series below 20 percent for total energy use and goods exports (16 and 19 percent, respectively). This alignment effectively captures the observed trends within the respective time series data, thereby affirming the robustness of the approach’s results. Secondly, MITICA’s methodology for evaluating PAMs has been tested in several published studies, providing additional evidence supporting the reliability of MITICA’s outcomes. Finally, the beta release of MITICA underwent comprehensive testing using a diverse range of input datasets. This testing phase was crucial for assessing the functionality of the software and refining any operational bugs.

Several limitations are discerned in the development of MITICA. Aligned with its objectives, MITICA formulates projections based on the assumption that GHG emissions are solely influenced by a set of proxies, holding other factors constant. This assumption dictates that changes in technology mix and consumer behaviour solely occur due to the implementation of PAMs. While essential for the study’s objectives, this assumption presents a significant limitation by overlooking the inherent evolution of emission profiles even in the absence of public intervention. Technology mixes and resource consumption are subject to modification based on different consumer preferences, which are not adequately captured by the current model. Although MITICA incorporates various PAMs allowing interventions at different levels (industries, consumers, or sectors), it fails to account for the natural evolution of emission sources and sinks without public intervention. Future enhancements could address this limitation by incorporating varying levels of change in further endogenous parameters into the WOM design to capture the evolution of emissions in the absence of public intervention.

The conducted robustness analysis indicates that the reliability of the forecast diminishes significantly in very long time frames. To enhance the reliability of mitigation scenarios beyond 2035, alternative methodological approaches such as back casting methods could be considered in conjunction with MITICA’s existing approach. Moreover, the incorporation of back casting approaches into MITICA’s framework could allow for a more holistic, long-term analysis without the necessity to define and calculate all PAMs individually. By adopting a forward-looking perspective, MITICA could provide a more comprehensive assessment of GHG emission scenarios.

Moreover, MITICA has not undergone empirical testing against alternative modelling frameworks. The use of sectoral methods instead of integrated sectors and the challenges in achieving consistency impede the identification of equivalent approaches to MITICA. Future research could address this limitation by conducting comprehensive empirical comparisons with alternative modelling frameworks, or sectoral modelling frameworks.

Further work in MITICA is identified in several key areas. Firstly, there is a need for extensive testing and fine-tuning, a process informed by rigorous testing outcomes. This iterative phase is crucial to enhance the reliability and precision of MITICA.

Additionally, MITICA should consider the incorporation of costs associated with PAMs and the generated scenarios. This enhancement would enable the model not only to estimate GHG emission reductions but also to provide insights into the potential costs associated with these mitigation measures. Such an inclusive approach would offer a more nuanced understanding of the economic implications of implementing various PAMs.

A further dimension for exploration involves assessing the impact of PAMs and GHG emission reductions on key economic variables such as GDP and population. Integrating the MITICA framework into a general equilibrium model would facilitate a comprehensive analysis, shedding light on the interconnected dynamics between environmental and economic factors.