**3. Methods**

#### *3.1. Landsat 7 ETM+ Data and Processing*

The LST maps, from which the SUHI ones are obtained, and the corresponding albedo maps, were retrieved using observations from the Enhanced Thematic Mapper Plus (ETM+) sensor on board the Landsat 7 satellite. ETM+ has six reflective bands in the visible (VIS), near infrared (NIR) and short-wavelength infrared (SWIR), a band in the thermal infrared region (TIR) and a panchromatic band, as reported in Table 1. The six reflective channels have a spatial resolution of 30 m, the thermal channel of 60 m, and the panchromatic one of 15 m. TIR band data are also delivered at 30 m by the U.S. Geological Survey [29], by means of a cubic convolution resampling method.

**Table 1.** Landsat 7 ETM+ channels and spectral range. Visible (VIS); near infrared (NIR); short-wavelength infrared (SWIR); thermal infrared region (TIR).


In this work, nine Landsat 7 scenes over Terni before and after the intervention on Corso del Popolo were considered, as reported in Table 2, downloaded from USGS at 30 m resolution. The native 60 m pixel size of ETM + TIR channel, the highest resolution of current satellite-based TIR channels, allows monitoring SUHI changes at the district level with sufficient details. Summer images in clear-sky condition were selected, since it is the period with the more intense SUHI effects, generally with the maximum intensity during the month of July in Italy, as detailed in [20]. The satellite passages are at around 11:50, Central European Summer Time (CEST).



The images were calibrated according to [30], in order to convert digital number values to at-sensor spectral radiances. The surface reflectances of the ETM+ reflective bands were computed through the scheme reported in [31], where only satellite scenes are needed, without the support of ground measurements.

#### *3.2. LST and Albedo Retrieval from Landsat 7 Data*

ETM + channel in the TIR band (Ch\_6) was employed to retrieve LST by inverting the following radiative transfer equation [32]:

$$L\_{\rm sens,\lambda} = \left[\varepsilon\_{\lambda} B\_{\lambda}(T\_s) + (1 - \varepsilon\_{\lambda}) L\_{\lambda}^{\downarrow}\right] \tau\_{\lambda} + L\_{\lambda}^{\uparrow} \tag{1}$$

where *Lsens*,*<sup>λ</sup>* is the radiance (W·m<sup>−</sup>2·sr<sup>−</sup>1·μm<sup>−</sup>1) received by the sensor, *ελ* is the emissivity of the surface, *Bλ* (*Ts*) is the spectral radiance (W·m<sup>−</sup>2·sr<sup>−</sup>1·μm<sup>−</sup>1) of a black body at temperature *Ts* (K) expressed by the Planck's law, *Ts* represents the LST, *L*↓*λ* and *L*↑*λ* are, respectively, the downwelling and upwelling atmospheric radiances (W·m<sup>−</sup>2·sr<sup>−</sup>1·μm<sup>−</sup>1), and *τλ* is the transmissivity of the atmosphere. If the surface emissivity is known, LST is retrieved from Equation (1) by the inversion of the Planck's law [33]. The terms *τλ*, *L*↓*λ* and *L*↑*λ* were calculated through the tool available at [34]: it makes use of the atmospheric profiles of the National Centers for Environmental Prediction as input for the code of radiative transfer MODTRAN [35]. The threshold method described in [36,37] and applied to the normalized difference vegetation index (NDVI) was used to estimate the land surface emissivity.

Overall, both thermal and reflective data, selected in cloud-free conditions, were atmospherically corrected.

The total broadband albedo is obtained from the reflective ETM + channels using the following relation [38]:

$$\text{Ra} = 0.356 \times \text{Ch}\_{\text{L}}1 + 0.130 \times \text{Ch}\_{\text{L}}3 + 0.373 \times \text{Ch}\_{\text{L}}4 + 0.085 \times \text{Ch}\_{\text{L}}5 + 0.072 \times \text{Ch}\_{\text{L}}7 - 0.0018 \quad \text{(2)}$$

Since the albedo is defined as a bi-hemispherical reflectance, whilst the Landsat spectral reflectance is the ratio of the hemispherical incoming radiation to the conical reflected radiation [39], assuming that the surfaces are Lambertian, the broadband albedo can be retrieved from the spectral reflectances of Landsat, regarded as narrowband albedos [38–40].

#### *3.3. SUHI Computation and LST Model*

The SUHI intensity is the parameter used to assess the heating effects in urban areas: in this case, it is the difference in physical surface temperature between urban (LSTurban) and rural (LSTrural) pixels within a given image:

$$\text{SUPII} = \text{LSTM}\_{\text{turban}} - \text{LSTM}\_{\text{ural}} \tag{3}$$

It was chosen to compute the LSTrural as the mean of LST associated with the rural pixels in the Terni area not affected by land cover changes over the years of the study.

In order to understand deeply the LST variations, an analytical model function of construction material behaviour, air temperature and heat transfer is considered. In fact, the LST variation is not simply driven by the albedo values, but further physical parameters have to be taken into account. Therefore, starting from the energy balance of a built-up surface when exposed to solar radiation, the LST is estimated as a function of different parameters beyond the albedo [10]:

$$\varepsilon \left( 1 - \alpha \right) I = \sigma \varepsilon \left( T\_s^4 - T\_{sky}^4 \right) + h\_c \left( T\_s - T\_a \right) \tag{4}$$

where *α* is the albedo, *I* the incident solar radiation on the surface (W/m2), *ε* the emissivity of the surface, *σ* the Stefan–Boltzmann constant (5.67 × 10−<sup>8</sup> W/m<sup>2</sup>·<sup>K</sup>4), *T*s is the LST (K), *<sup>T</sup>*sky the radiating temperature of the sky (K), *h*c the convection (and radiation) coefficient (W/m<sup>2</sup>·K), and *Ta* the air temperature (K).

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

The changes of the area under investigation (Figure 2) is monitored in terms of SUHI intensity maps as shown in Figure 4 (before the intervention) and Figure 5 (after the intervention).

**Figure 4.** Pre-intervention surface urban heat island (SUHI) maps (◦C): 27 June 2005; 24 June 2004; 16 August 2000, and; 31 July 2000. Landsat pixels are 30 × 30 m from the USGS. Lat/lon are in UTM WGS84 coordinates, 33T zone.

The 60 m pixel size of Landsat 7 TIR channel, resampled at 30 m, allows assessing SUHI pattern variations at district level (the selected area is 300 m × 300 m). The level of detail of data derived from satellite is coarser than that of Figure 3, but looking at Figure 3 itself, it emerges that there are areas with the same characteristics (lawn finish, or painted metal roofs) covering a surface comparable or larger than the satellite pixel size. Even though sharp thermal variations are averaged with the surroundings inside the pixel area, the trend of the SUHI pattern can be delineated, revealing its behaviour inside the district.

**Figure 5.** Post-intervention SUHI maps [◦C]: 28 August 2016; 27 July 2016; 25 July 2015; 7 June 2015, and; 6 July 2014.

It is clear from Figure 5 how the new construction area (pixels inside the circle) produces a mitigation of the SUHI with respect to the heat island of the built-up area located in the upper part of the reported images (the latter area is present both pre- and post-interventions, as shown in Figure 2). The SUHI reduction is evident also during July, when a strong heating effect is found in the upper built-up area, and a difference of about 10 ◦C is observed between the two zones. In Figure 4, where the paved car park was not ye<sup>t</sup> replaced by the new area, the heat island is similar or greater than the upper area, and the tree cooling effect is detectable only in the right part of the image, near the Nera River.

The SUHI diminution of the zone under analysis after June 2014 (around 5–6 ◦C) could be linked to various factors. As widely confirmed from literature [13–15,41], urban greenery mitigates SUHI: in the new district, from summer 2000 to summer 2016, the mean NDVI computed from satellite data increased from 0.19 to 0.23. Furthermore, a three-level underground car park was constructed and a significant part of the roadway transferred underground, thereby avoiding the temperature increase linked to the heat generated by the circulating vehicles on the corresponding area. This benefit is confirmed in a simulated study [42] reporting how the underground space development improves the urban microclimate. Finally, the construction of relatively high buildings with respect to the previous flat car park lot produced a shading effect and a probable improvement of the local air circulation. The impact of building design on the shadow effect was assessed in [43], showing how shorter urban buildings cause higher surface and near-surface temperatures during the daytime. Hu and Liao [44] sugges<sup>t</sup> how a suitable space distance of buildings contributes to improve ventilation between them, and the new building position in Corso del Popolo seems to match the favourable conditions in [44].

#### *Albedo Maps and Analytical Analysis*

The previous SUHI patterns may be also analysed in conjunction with the albedo pattern variations pre- and post-intervention. The albedo retrieved by Landsat data proved to be a useful tool to detect surfaces with different albedo behaviour in a built environment [40,45], without expensive in situ measurements unevenly distributed in space, and therefore unable to provide a global pattern. A high reflective surface (typically light in colour) absorbs less solar radiation than a conventional low reflective surface (dark in colour). Therefore, it is expected that high albedo surfaces lead to lower LST and that low albedo ones lead to higher LST. It is important to underline that vegetation pixels are characterized by low albedo values, but the effect on local overheating is very different with respect to the low albedo of impervious surfaces. In fact, the solar energy stored by vegetation is mainly used for evapotranspiration and life processes, producing a cooling effect of the green areas. Figure 6 shows an example of two albedo maps pre- and post-intervention.

**Figure 6.** Albedo maps; (**a**) 27 June 2005; (**b**) 27 July 2016.

The purpose of this analysis is the evaluation of the albedo pattern in relation with the heating/cooling of the urban area. Figure 5 points out that the albedo variation in the changed area is not remarkable (pixel values are in the range of 0.18–0.24 for both images in the circled zone, with a slight increase post-intervention). The same pattern is roughly found in all the nine images. The satellite remote sensing using Landsat 7 reflective band observations provides a map of albedo values with a 30 m pixel size, averaging intrinsic reflectance heterogeneities at a finer scale. For instance, the presence of vegetation spots (low albedo) inside a built-up texture having high albedo produces a decreased value in the satellite pixel, even though both surfaces have a cooling effect. The shade generated by the new tall buildings may produce a retrieved albedo lower than its correct value, even if the Landsat 7 passages at around 12:00 reduce this effect, since midday sun position causes a reduction of the buildings shadow extension. Overall, taking into account the impact of the area with vegetation on albedo measurements, the new building system and impervious surfaces present an undiminished albedo pattern [46] with respect to the pre-intervention scene.

Before 2006 the paved parking had also perimetral trees that should have been a potential mitigation effect on summer heating of the area. Since the SUHI trend of Figure 3and Figure 4 shows a clear enhancement post-intervention, the synergy of different urban design solutions, as greenery, parking/roadway covering and building characteristics (dimension and materials) proves to be a more impacting feature. Therefore, to understand deeply the SUHI variations, an overall assessment by means of Equation (4), depending on construction material behaviour, air temperature and heat transfer, is useful.

Equation (4) suggests that the most practical parameter to be changed on a large scale is albedo [10]. At the same time, the emissivity of a surface affects LST. For SUHI mitigation, it is necessary to avoid low-emissivity surfaces since they maintain a higher surface temperature when exposed to the sun than a high-emissivity material. Low-emissivity materials include many unpainted/polished metal surfaces, whereas most not metallic urban surfaces exhibit high emissivity. High albedo roofs and walls can be obtained with white surface coatings, while high albedo pavement includes concrete and conventional asphalt with white aggregate. Considering the thermal storage, the convection coefficient describes the convective heat transfer between the surface and the air flowing over, depending on thermal properties of the medium, characteristics of its flow, surface geometry, thermal boundary conditions and the variability of the surface temperature. Furthermore, Equation (4) is valid in steady-state conditions: since satellite passages are at around midday, the thermal storage phenomenon is nearly terminated, considering that the surface warming by the sun started some hours before in summertime.

An example of the impact of each single parameter variation on the LST is reported in Table 3. The aim of this analysis is the evaluation of the most significant key factors inducing SUHI variation for impervious surfaces. The values for the electromagnetic parameters *α* and *ε* are chosen in a typical range for construction materials [47], as well as the *hc* range follows the characteristic behaviour in a built environment. The parameters *Ta* and *<sup>T</sup>*sky are chosen in a range following the atmospheric conditions of continental Central Italy summer around midday [48] ( *Ta* in the range 25–35 ◦C and *<sup>T</sup>*sky in the range 278–288 K, the latter computed considering the relative humidity of air ranging from 30% to 70%). The incident solar radiation on the surface *I* is settled at 900 W/m2, an average value measured in the Central Italy in July 2016 at around 12:00 CEST.


**Table 3.** Land surface temperature (LST) variation from Equation (4) (*I* = 900 W/m2).

It is evident that the LST in correspondence to impervious surfaces turns out to be strongly sensitive not only to the surface albedo, but also to the local convection coefficient, and to the surrounding air temperature. The increase of *α* shows a clear benefit on urban heating mitigation, as well as the *hc* increase and *Ta* decrease. The emissivity variation impact, in the typical range of built-up materials (significant unpainted or polished metallic surfaces are not present in the study area), is negligible.

Overall, the analysis confirms how the understanding of the SUHI variation requires the evaluation of both city planning factors (green surface, roadway and building design) and physical parameter behaviour (albedo, heat transfer and air circulation) [49].
