Changes in Surface Urban Heat Island Effect with the Development of New Towns

6 A new town is a planned and built within a relatively short period compared to existing cities. It is 7 suitable for climate and thermal research, particularly formulating urban planning strategies to analyse 8 problems such as urban heat islands (UHIs). Herein, a comprehensive approach was demonstrated for 9 determining changes in surface UHI (SUHI) distribution during 1989 – 2048 in two new towns with 10 different urban planning. A significant increase in built-up areas was observed from 1989 (< 5%) to 2018 11 (> 40%) in both new towns. However, the areas where SUHIs occurred before and after development 12 increased further (approximately 12.25%) in Bundang new town where the green area ratio was lower 13 and the building density was higher. However, in terms of SUHI intensification, the building structure of 14 Pangyo new town, which has a lower floor area ratio, was analysed to be more suitable. Moreover, 15 without effective mitigation, the built-up area in each new town is estimated to increase to approximately 16 60%, and the SUHI intensity in most areas to increase by 4 °C in 2048. Thus, these results combined with 17 architectural assessment models can improve the understanding of thermal environmental impacts of 18 urbanisation and help mitigate urban thermal hazards.

Building floor ratio means the sum of the gross horizontal area of each floor of a building as measured to 118 the exterior face of the exterior walls of the building. According to Oke et al. (2017) 14 , the facet surface 119 temperature in daytime in urban system is typically ranked as follows: T roof > T walls > T floor > T surrounding area . In 120 addition, in canyons formed in the city through high-rise buildings, overshadowing areas are formed to induce 121 surface coolness 60 . As a result, it was found that buildings newly built in Pangyo new town, which have a lower 122 height than Bundang new town and a certain level of building coverage ratio, are more suitable for increasing 123 surface temperature. (49.16%) to 19.78 km 2 (59.12%) between 2018 and 2048 in Bundang new town (Fig 2a). Moreover, it predicted 127 decreases in forest areas from 35.61% to 29.9% and the grass cover from 12.76% to 10.69%. As new town 128 development in the past primarily occurred through transformation of agricultural areas to built-up areas, it was 129 not predicted that a significant urban expansion would occur through deforestation. In addition, most of the 130 buildings in the housing complex of Bundang new town were completed in 1990, over 25 years ago. Therefore, 131 renovations are planned for most of these old apartment complexes to improve the poor residential environment 132 and meet the latest urban housing requirements. Hence, most urban expansion was predicted to occur through 133 renovation within the existing built-up areas and partial transformation of the forest surrounding the new town. 134 In the case of Pangyo new town, the proportion of urban expansion between 2018 and 2048 was 135 predicted to be higher than that of Bundang new town. According to the CA-MCM prediction, built-up areas 136 would increase by approximately 18.42% from 40.81% to 59.23%, the forest areas would decrease from 40.84% 137 to 32.25%, and the grass cover including golf courses would decrease from 15.34% to 7.92% (Fig. 3a). The 138 primary trend observed in the predicted urban expansion was that non-urban areas, such as forest and grass, 139 surrounding the main road were transformed into built-up areas. In contrast with Bundang new town, Pangyo 140 new town is public-transportation-oriented. During the past new town development, the areas surrounding the 141 main road that existed outside the city were underdeveloped. However, if urban expansion occurs in the future, 142 it would be evident primarily in areas with good road proximity. In addition, urban expansion due to the 6 within the city was also predicted. In terms of agricultural area and water, both new towns were predicted to 145 remain almost unchanged from 2018, with little fluctuation. 146 Predicted SUHI distribution for 2028, 2038, and 2048. CA-MCM predicted the increase in area and 147 intensity of the SUHI phenomenon in both new town and, unlike LULC prediction, a significant change was 148 predicted. In Bundang new town, the areas where the SUHI phenomenon occurs would increase by 149 approximately 5% between 2018 and 2048. For SUHI intensity distribution, the areas with SUHI ≤ 4 ℃ would 150 decrease from 17.12 km 2 (51.16%) to 11.44 km 2 (34.21%). Simultaneously, the areas with SUHI > 4 ℃ was 151 estimated to increase from 4.25 km 2 (12.73%) to 10.68 km 2 (34.71%), affecting the lower SUHI intensity areas. 152 It is predicted that SUHI intensity would expand and increase from the existing residential area, which may 153 reflect the renovation trend partially occurring between 2000 and 2018. Therefore, development of sustainable 154 renovation guidelines is required such as thermal insulation, replacement of the insulation material, and 155 improving the air tightness of the building envelope through renovation using insulation materials 19 . In addition, 156 the areas with SUHI > 6 °C are predicted to increase from 0. 56 km 2 (1.7%) to 2.77 km 2 (8.28%). It has been 157 observed that the higher the LST, the higher the frequency of heat waves at regional scales 20 . In the future, 158 additional thermal environmental policies and energy policies are required for areas where SUHI intensity is 159 expected to increase significantly (Fig. 3a). 160 In the case of Pangyo new town, the areas where the SUHI phenomenon occurred were predicted to 161 increase by 20%. The affected areas are similar to those predicted to change from forests existing around the 162 main road to built-up areas. For SUHI intensity distribution, the area with SUHI ≤ 4 ℃ would decrease from 163 7.75 km 2 (43.97%) to 5.08 km 2 (28.83%). Moreover, the areas with SUHI > 4 ℃ would increase from 2.53 km 2 164 (14.34%) to 8.7 km 2 (49.36%), and most areas were in the range 4 ℃-6 ℃ (49%) (Fig. 3c). Therefore, it can be 165 predicted that urban features, such as structural characteristics, materials, and building disposition type would 166 change according to the housing complex newly built through new town development. 167

Discussion 168
This study is the first attempt to simulate and compare the pattern of UHI occurrence according to new town 169 development using remote sensing and GIS technology. This discussion focuses on the principal two 7 The main contribution of our study is that the different patterns of changes in land use land cover and 173 SUHI phenomenon depending on urban planning were visually and quantitatively shown for the study sites In addition, through prediction analysis, the importance of building renovation and structural 207 characteristics in urban-level thermal environment changes was also suggested. According to Yahia et al.,208 (2018) 60 , closely speed high-rise buildings have a negative impact on ventilation and the average wind speed in 209 the dense high-rise buildings area is less than half of that of the low-rise buildings area. However, when 210 comparing the physiologically equivalent temperature, high-rise buildings is more comfortable, and the shade 211 seems to be more important factor than wind speed. It means that decreasing solar radiation through shade will 212 have a greater effect on decreasing sensible heat and thermal comfort near the surface than promoting the wind 213 speed. When renovating old buildings in the future, three-dimensional design considering the effects of shadow 214 and wind at the same time are required. 215 While the presented study provides useful method and information regarding the current and future status 216 of the UHI phenomenon, it is still faced some limitations. This study does not consider additional parameters 217 typically influencing the urban growth because of the specificity of the study area. As mentioned, new town is 218 the planned city where the physical and legal aspects of the site were reviewed through feasibility analysis 219 beforehand, the complication associated with urban expansion is relatively low for new town. However, the 220 factors for urbanisation are related to the complexity of the terrain, degree of socio-economic development, 221 urban regulations, etc 24 . Therefore, it is necessary to consider additional factors for urban expansion when 222 applying this methodology to a region other than new town in the future. In addition, a model that explains the 223 detailed behaviour of UHI using a combination of building renovation and structural characteristics is still 224 necessary. Future research studies should attempt to obtain structural and temporal data over the same period of 225 time and develop models able to explain the change of UHI based on structural characteristics changed by 226 building renovation. 227

Conclusions 228
Although the research methods and measures face certain conceptual and practical challenges, this study 229 suggested a proximate causal relationship between urban expansion and SUHI phenomenon change according to 230 urban planning. It is easy to apply for practitioners and the necessary data for application are available without 9 complex acquisition procedures or unopened access datasets. Therefore, the proposed novel method may be 232 applied to both existing and newly built cities to predict future UHI distribution according to urban planning. 233 Furthermore, the findings and methods constructed through this research can be useful to policy makers, urban 234 planners, researchers, and citizens to adopt sustainable thermal environment management practices including 235 adaptation and mitigation strategies for the city.
where i is the class number; n is the total number of points; n ii is the number of pixels belonging to the actual 274 data class i, which were classified as class i; C i is the total number of classified pixels belonging to class i; and 275 G i is the total number of actual data belonging to class i. Fifty sample points per class for each new town, except 276 water class, were selected automatically by QGIS 3.14. A minimum of 50 samples must be collected for each 277 land cover class in the error matrix to avoid the risk of a biased sample during accuracy assessment 29 . 278 LST estimation. LST estimation using ArcMap 10.5 includes transforming DNs to radiance (L λ ), measuring 279 radiance brightness temperatures (T B ), and adjusting emissivity to extract surface temperature from brightness 280 maps 30 . The LST values were obtained using thermal bands from Landsat TM (B6) and Landsat OLI/TIRS 281 (B10) because of the USGS recommendation to avoid using TIRS band 11 because of its higher calibration 282

uncertainty. 283
Every object on the Earth emits thermal electromagnetic radiation when its temperature is above absolute 284 zero (K), and the signal received by the thermal sensors can be transformed to radiance (L λ ) using equation (2): 285 where L λ is the spectral radiance in W/(m 2× sr×μm); M L is the radiance multiplicative scaling factor for the band; 286 A L is the radiance additive scaling factor for the band; and Q cal is the level 1 pixel value in DN, whose values are 287 obtained from the metadata of the Landsat images. After the DN value was converted to radiance, the radiance 288 values were converted to T B using equation (3): 289 where T B is the At-satellite brightness temperature and K 1 and K 2 represent the band-specific thermal conversion 290 constants from the metadata. To obtain the temperature in Celsius, the radiant temperature is revised 30  The obtained values of T B were referenced as a black body, whose properties are different from that of 297 real objects on the Earth's surface and would also be different from real LST 33 . The LST values across a city can 298 have a wide range, and it depends on LULC states constructed within the city. Furthermore, LSE, which is 299 essential for estimating the LST, has strong land use/land cover dependence 34,35 . 300 The LSE value is calculated conditionally using equation (5), and the condition is represented by the 301 formula for each emissivity value 36,37 : 302 where v and are the vegetation and soil emissivity, respectively and C is the surface roughness (C = 0 for 303 homogeneous and flat surfaces), with a constant value of 0.005 38 . When the normal difference vegetation index 304 (NDVI) is less than NDVI S = 0.2, it is classified as bare soil and its emissivity value is acquired from the 12 soil and vegetation surfaces, and equation (5) is used for extracting their emissivity values. In the equation, ε λv is 307 the emissivity value of vegetation (= 0.9863 μm) and ε λs is emissivity value of soil (= 0.9668 μm) in this 308 range 40 . When the NDVI value is larger than NDVI v = 0.5, it is considered as a vegetation surface and an 309 emissivity value of 0.99 is assigned to it 30 . Visible red and near-infrared (NIR) bands were used for calculating 310 NDVI using equation (6). In addition, NDVI values were used to evaluate the proportion of the vegetation (P v ) 311 related to emissivity (ε) using equation (7) 41,42 . A method for calculating P v using the NDVI values for 312 vegetation soil, which can be applied in global conditions, was suggested in a previous study 36 . 313 where S(t) is the system state at time t, S (t+1) is the system state at time t+1, and P ij is the transition probability 325 matrix in a state, which is calculated using equation (9). 326 (0 ≤ P ij ≤ 1) (9) P is the Markov probability matrix, P ij is the probability of converting from current state i to another state j in 327 prediction time, and P N is the state probability of any time. Low transition pixels have a low probability value 328 near (0), and high-transition pixels have a high probability value near (1)  where T s is the LST (℃) distribution of new town, and T mean and δ are the mean and standard deviation of LST 369 in non-urban areas of new town. By subtracting the average temperature of non-urban areas from the 370 temperature of the entire city, it may be verified that the actual SUHI effect was due to urban expansion, rather 371 than the temporary LST value. In addition, the water bodies were excluded while calculating the SUHI intensity 372 because it can irregularly influence the surface temperature (Lee et al, 2020). (2) The SUHI intensity variation 373 was classified into six appropriate ranges: (ⅰ) value ≤ 0 ℃, (ⅱ) 0 ℃ < value ≤ 2 ℃, (ⅲ) 2 ℃ < value ≤ 4 ℃, (ⅳ) 374 4 ℃ < value ≤ 6 ℃, (ⅴ) 6 ℃ < value ≤ 8 ℃, (ⅵ) 8 ℃ < value. Thus, the difference in distribution and intensity 375 of the SUHI phenomenon can be compared according to the change in LULC for each new town at each time 376 period. In addition, classes are divided into value ranges, to facilitate future SUHI intensity distribution 377 prediction using CA-Markov analysis. The indices, which were positively and negatively correlated with LST, 378 were used to develop transition suitability maps for predicting the SUHI distribution. The normalised difference 379 built-up index (NDBI) was used as the index that highly correlated with LST 56 . NDBI is the most widely were also used to standardise the factor maps to 0-1, where 0 represents a low SUHI potential and 1 represents a 388 high SUHI potential.                  not imply the expression of any opinion whatsoever on the part of Research Square concerning the legal status of any country, territory, city or area or of its authorities, or concerning the delimitation of its frontiers or boundaries. This map has been provided by the authors.