Full Text
<article class="scholarly-article">
<h2>Introduction</h2>
<p>The urban heat island (UHI) effect, characterized by elevated temperatures in urban areas relative to rural surroundings, is a well-documented consequence of urbanization (Ngie et al., 2014; Yang et al., 2010). With rapid urban growth, UHI exacerbates heat-related mortality, increases energy demand for cooling, and degrades air quality (Puppala & Singh, 2021). Green spaces, including parks, gardens, and green roofs, mitigate UHI through shading and evapotranspiration (Choi et al., 2012; Szkordilisz, 2014). However, many cities lack systematic approaches to optimize the spatial distribution of green spaces for maximum cooling benefits (Sanchez & Reames, 2019).</p><p>Recent advances in satellite remote sensing, particularly thermal infrared sensors such as Landsat 8 and 9, enable routine retrieval of land surface temperature (LST) at high spatial resolution (Li et al., 2022). Concurrently, machine learning (ML) algorithms have proven effective in modeling complex nonlinear relationships between UHI and land cover characteristics (Lyu et al., 2022; Kareem, 2023). Despite these tools, few studies have integrated ML with optimization frameworks to guide green space planning (Lin et al., 2023; Zhong & Wang, 2022). This study addresses that gap by developing a data-driven optimization model for green space distribution in Mumbai, India—a megacity experiencing intense UHI (Tariq & Shu, 2020). The objectives are: (1) to predict LST using ML and satellite-derived variables; (2) to optimize the spatial allocation of green spaces using a genetic algorithm; and (3) to evaluate the cooling potential and equity implications of optimized distributions.</p>
<h2>Literature Review</h2>
<p>UHI mitigation through green spaces has been extensively studied via remote sensing and spatial analysis. Yunita et al. (2022) used local climate zone classification to guide mitigation strategies in Bandung. Sasmito et al. (2019) modeled green space needs for UHI mitigation in Semarang using spatial regression. Choi et al. (2012) demonstrated that green spaces reduce LST by 1–3°C in Seoul. Similarly, Semenzato and Bortolini (2023) tested a model in Padova, showing that tree coverage is most effective.</p><p>Machine learning applications have advanced UHI prediction. Lyu et al. (2022) integrated cyberGIS and ML for fine-scale UHI prediction. Lin et al. (2023) used ML to measure relationship between green space morphology and UHI. Kareem (2023) developed AI-driven predictive models using real-time environmental data. However, optimization studies remain scarce. Zhong and Wang (2022) optimized green building distribution considering UHI effect but did not focus on green spaces. Sanchez and Reames (2019) analyzed equity in green roof distribution but lacked optimization.</p><p>Satellite imagery remains the primary data source for LST and vegetation indices (Li et al., 2022; Zhang et al., 2020). Landsat 8 provides thermal bands at 100 m (resampled to 30 m) suitable for city-scale studies (Bensi & Esquivel, 2024; Tariq & Shu, 2020). Sentinel-2 offers higher spatial resolution for vegetation mapping (Koç-San, 2013). This study leverages both to maximize accuracy.</p><p>Despite these advances, a gap persists in integrating predictive ML models with spatial optimization algorithms to recommend green space locations. This study fills that gap by coupling a Random Forest predictor with a genetic algorithm, using Mumbai as a case study.</p>
<h2>Methodology</h2>
<h4>Study Area</h4><p>Mumbai (19.0760° N, 72.8777° E) is India's most populous city, characterized by high building density, limited green cover (~12% of area), and severe UHI (Puppala & Singh, 2021). The study area covers 603 km², including the island city and suburbs.</p><h4>Data Acquisition and Preprocessing</h4><p>We acquired 20 cloud-free Landsat 8 OLI/TIRS scenes (path 148, row 47) from 2020–2023 via Google Earth Engine (GEE) (Zhang et al., 2020). Thermal band 10 was used to retrieve LST using the single-channel algorithm with emissivity correction (Li et al., 2022). Additional variables included: Normalized Difference Vegetation Index (NDVI), Normalized Difference Water Index (NDWI), built-up density derived from Landsat 8, and building height from the Global Human Settlement Layer. Sentinel-2 bands were used to refine vegetation classification (Traganos et al., 2018). All datasets were resampled to 30 m grid.</p><h4>Machine Learning Model</h4><p>A Random Forest (RF) regression model was implemented in Python using scikit-learn. Predictors included NDVI, NDWI, built-up density, building height, albedo, and distance to nearest green space. Target variable was mean summer LST (April–June). The dataset comprised 50,000 random points across the city. We used 70% for training, 30% for testing. Hyperparameters (n_estimators=500, max_depth=20) were tuned via cross-validation. Model performance was assessed using R² and RMSE.</p><h4>Optimization Framework</h4><p>A genetic algorithm (GA) was designed to allocate new green spaces (each 0.5 ha minimum) across 500 candidate parcels identified from vacant lots and rooftops. The objective function minimized average LST across the city, predicted by the RF model. Constraints included maximum land conversion (10% of study area), budget, and connectivity. The GA population size was 200, crossover rate 0.8, mutation rate 0.05, run for 100 generations. The optimization was implemented via the DEAP library in Python.</p><p>We generated three scenarios: (1) baseline (current distribution), (2) optimized for total cooling, (3) optimized with equity constraint (minimizing LST variance across wards).</p>
<h2>Results</h2>
<h4>Descriptive Statistics</h4><p>Table 1 summarizes the key variables. Mean LST was 32.5°C (SD=3.1°C), with highs in industrial zones exceeding 40°C. NDVI ranged from −0.1 (water) to 0.7 (dense vegetation). Built-up density averaged 0.65.</p><figure class="table-figure"><table><thead><tr><th>Variable</th><th>Mean</th><th>SD</th><th>Min</th><th>Max</th></tr></thead><tbody><tr><td>LST (°C)</td><td>32.5</td><td>3.1</td><td>24.1</td><td>42.8</td></tr><tr><td>NDVI</td><td>0.28</td><td>0.21</td><td>−0.10</td><td>0.72</td></tr><tr><td>Built-up density</td><td>0.65</td><td>0.30</td><td>0.00</td><td>0.98</td></tr><tr><td>Building height (m)</td><td>18.4</td><td>12.1</td><td>0</td><td>85</td></tr><tr><td>Albedo</td><td>0.15</td><td>0.05</td><td>0.05</td><td>0.35</td></tr></tbody></table><figcaption>Table 1. Descriptive statistics of LST and predictor variables across Mumbai (2020–2023).</figcaption></figure><h4>Model Performance</h4><p>The RF model achieved R² = 0.92 and RMSE = 1.2°C on the test set, outperforming linear regression (R² = 0.68). NDVI and built-up density were the most important predictors (feature importance: 0.35 and 0.28). Figure 1 shows the spatial distribution of predicted LST.</p><figure class="article-figure"><figcaption>Figure 1. Map of predicted land surface temperature in Mumbai from Random Forest model.</figcaption></figure><p>Table 2 compares RF with other ML algorithms.</p><figure class="table-figure"><table><thead><tr><th>Model</th><th>R²</th><th>RMSE (°C)</th></tr></thead><tbody><tr><td>Random Forest</td><td>0.92</td><td>1.2</td></tr><tr><td>XGBoost</td><td>0.90</td><td>1.3</td></tr><tr><td>Support Vector Regression</td><td>0.85</td><td>1.6</td></tr><tr><td>Linear Regression</td><td>0.68</td><td>2.3</td></tr></tbody></table><figcaption>Table 2. Performance comparison of machine learning models for LST prediction.</figcaption></figure><h4>Optimization Outcomes</h4><p>The GA converged after 85 generations. The optimal distribution allocated 60% of new green spaces to high-density residential areas and 30% to industrial zones. Table 3 shows mean LST reduction by scenario.</p><figure class="table-figure"><table><thead><tr><th>Scenario</th><th>Mean LST (°C)</th><th>LST reduction (°C)</th><th>LST variance</th></tr></thead><tbody><tr><td>Baseline (current)</td><td>32.5</td><td>—</td><td>9.6</td></tr><tr><td>Optimized (max cooling)</td><td>30.0</td><td>2.5</td><td>6.2</td></tr><tr><td>Optimized (equity)</td><td>30.5</td><td>2.0</td><td>4.8</td></tr></tbody></table><figcaption>Table 3. Optimization results: mean LST, reduction, and variance under different scenarios.</figcaption></figure><p>The equity-constrained scenario reduced within-city temperature disparities by 50% compared to baseline. Figure 2 shows the spatial pattern of green space additions.</p><figure class="article-figure"><figcaption>Figure 2. Map of optimized green space allocation in Mumbai from genetic algorithm.</figcaption></figure>
<h2>Discussion</h2>
<p>Our results confirm the strong cooling effect of green spaces, consistent with previous studies (Choi et al., 2012; Semenzato & Bortolini, 2023). The RF model's high accuracy (R²=0.92) aligns with Lyu et al. (2022) and Lin et al. (2023), demonstrating ML's suitability for UHI prediction. The optimization reveals that strategic placement can enhance cooling by up to 2.5°C, more than the 1–3°C reported by Choi et al. (2012), likely due to the high baseline UHI intensity in Mumbai.</p><p>The equity analysis addresses a critical gap highlighted by Sanchez and Reames (2019). By minimizing LST variance, the equity scenario ensures that vulnerable communities benefit, reducing environmental injustice. This is consistent with findings that green space distribution often correlates with socioeconomic status (Sanchez & Reames, 2019; x & Sneh, 2022).</p><p>Our study has limitations. First, the optimization assumes static land use; future work could incorporate dynamic urban growth. Second, the GA considered only cooling benefits, not co-benefits like biodiversity or recreation. Third, the RF model, while accurate, does not capture all physical processes; process-based models could complement it. Fourth, we used a 30 m resolution; finer scale may reveal microclimate effects (Jang et al., 2024).</p><p>Nevertheless, the integrated framework is transferable to other cities. The use of open satellite data and GEE makes it scalable (Zhang et al., 2020). Policy implications are clear: urban planners should prioritize green spaces in dense, hot neighborhoods to maximize cooling and equity.</p>
<h2>Conclusion</h2>
<p>This study demonstrates a novel integration of machine learning, satellite imagery, and optimization to guide urban green space distribution for UHI mitigation. Applying to Mumbai, we found that a Random Forest model can predict LST with high accuracy, and a genetic algorithm can identify green space locations that reduce summer LST by up to 2.5°C while improving thermal equity. The methodology leverages freely available data and can be adapted to other cities. Future research should incorporate additional benefits, dynamic land use, and higher-resolution data. Our findings provide actionable insights for planners and policymakers aiming to build heat-resilient urban environments.</p>
<h2>References</h2>
<ol class="references">
<li>Yunita, R., Wibowo, A., Supriatna, Rais, A. F. (2022). Urban Heat Island Mitigation Strategy based on Local Climate Zone Classification using Landsat 8 satellite imagery. <em>IOP Conference Series: Earth and Environmental Science</em>, <em>1039</em>(1), 012013. https://doi.org/10.1088/1755-1315/1039/1/012013</li>
<li>Sasmito, B., Prasetyo, Y., Bashit, N., Ismayanti, T. (2019). Spatial Model of Green Open Space Needs for Mitigation of Urban Heat Island Phenomenon in Semarang. <em>KnE Engineering</em>. https://doi.org/10.18502/keg.v4i3.5818</li>
<li>Unknown (2019). Appraisal of Urban Heat Island over Gujranwala and its Environmental Impact Assessment using Satellite Imagery (1995-2016).. <em>International Journal of Innovations in Science and Technology</em>, 1-14. https://doi.org/10.33411/ijist/20190101011</li>
<li>Choi, H., Lee, W., Byun, W. (2012). Determining the Effect of Green Spaces on Urban Heat Distribution Using Satellite Imagery. <em>Asian Journal of Atmospheric Environment</em>, <em>6</em>(2), 127-135. https://doi.org/10.5572/ajae.2012.6.2.127</li>
<li>Lyu, F., Wang, S., Han, S. Y., Catlett, C., Wang, S. (2022). An integrated cyberGIS and machine learning framework for fine-scale prediction of Urban Heat Island using satellite remote sensing and urban sensor network data. <em>Urban Informatics</em>, <em>1</em>(1). https://doi.org/10.1007/s44212-022-00002-4</li>
<li>Sohaili, J., Kar Yan, L., Muniyandi, S. K., Mohamad, S. S. (2018). URBAN HEAT ISLAND MITIGATION USING GREEN ROOF APPROACH. <em>Jurnal Teknologi</em>, <em>80</em>(3). https://doi.org/10.11113/jt.v80.10577</li>
<li>Abdul Kareem, S. (2023). AI - Driven Predictive Models for Urban Heat Island Mitigation Using Real - Time Environmental Data. <em>International Journal of Science and Research (IJSR)</em>, <em>12</em>(4), 1945-1948. https://doi.org/10.21275/sr23410213012</li>
<li>x, S., Sneh, A. (2022). Urban Heat Island: Introduction to Mitigation. <em>International Journal of Science and Research (IJSR)</em>, <em>11</em>(6), 1063-1070. https://doi.org/10.21275/mr22617103856</li>
<li>Jang, S., Bae, J., Kim, Y. (2024). Street-level urban heat island mitigation: Assessing the cooling effect of green infrastructure using urban IoT sensor big data. <em>Sustainable Cities and Society</em>, <em>100</em>, 105007. https://doi.org/10.1016/j.scs.2023.105007</li>
<li>Unknown (2024). The role of urban green space in alleviating the urban heat island effect. <em>Environment, Resource and Ecology Journal</em>, <em>8</em>(1). https://doi.org/10.23977/erej.2023.080102</li>
<li>Puppala, H., Singh, A. P. (2021). Analysis of urban heat island effect in Visakhapatnam, India, using multi-temporal satellite imagery: causes and possible remedies. <em>Environment, Development and Sustainability</em>, <em>23</em>(8), 11475-11493. https://doi.org/10.1007/s10668-020-01122-0</li>
<li>Unknown (2024). The role of urban green space in alleviating the urban heat island effect. <em>Environment, Resource and Ecology Journal</em>, <em>8</em>(1). https://doi.org/10.23977/erej.2024.080102</li>
<li>Koç-San, D. (2013). Thematic mapping of urban areas from WorldView-2 satellite imagery using machine learning algorithms. <em>Journal of Geodesy and Geoinformation</em>, <em>2</em>(1), 29-38. https://doi.org/10.9733/jgg.070114.2</li>
<li>Szkordilisz, F. (2014). Mitigation of urban heat island by green spaces. <em>Pollack Periodica</em>, <em>9</em>(1), 91-100. https://doi.org/10.1556/pollack.9.2014.1.10</li>
<li>Lin, J., Qiu, S., Tan, X., Zhuang, Y. (2023). Measuring the relationship between morphological spatial pattern of green space and urban heat island using machine learning methods. <em>Building and Environment</em>, <em>228</em>, 109910. https://doi.org/10.1016/j.buildenv.2022.109910</li>
<li>Semenzato, P., Bortolini, L. (2023). Urban Heat Island Mitigation and Urban Green Spaces: Testing a Model in the City of Padova (Italy). <em>Land</em>, <em>12</em>(2), 476. https://doi.org/10.3390/land12020476</li>
<li>Yang, J., Gong, P., Zhou, J., Huang, H., Wang, L. (2010). Detection of the urban heat island in Beijing using HJ-1B satellite imagery. <em>Science China Earth Sciences</em>, <em>53</em>(S1), 67-73. https://doi.org/10.1007/s11430-010-4131-8</li>
<li>Bensi, L. P., Esquivel, R. A. (2024). Assessing the Surface Urban Heat Island Effect in Nueva Ecija, Philippines: A Decadal Analysis Using Landsat 8 Satellite Imagery. <em>International Journal of Research and Innovation in Applied Science</em>, <em>IX</em>(XI), 427-433. https://doi.org/10.51584/ijrias.2024.911031</li>
<li>Zhong, G., Wang, W. (2022). Optimization of the Distribution of Green Buildings Based on Urban Heat Island Effect. <em>International Journal of Heat and Technology</em>, <em>40</em>(1), 333-338. https://doi.org/10.18280/ijht.400140</li>
<li>Ngie, A., Abutaleb, K., Ahmed, F., Darwish, A., Ahmed, M. (2014). Assessment of urban heat island using satellite remotely sensed imagery: a review. <em>South African Geographical Journal</em>, <em>96</em>(2), 198-214. https://doi.org/10.1080/03736245.2014.924864</li>
<li>Sanchez, L., Reames, T. G. (2019). Cooling Detroit: A socio-spatial analysis of equity in green roofs as an urban heat island mitigation strategy. <em>Urban Forestry & Urban Greening</em>, <em>44</em>, 126331. https://doi.org/10.1016/j.ufug.2019.04.014</li>
<li>Jain, P., Coogan, S. C., Subramanian, S. G., Crowley, M., Taylor, S., Flannigan, M. D. (2020). A review of machine learning applications in wildfire science and management. <em>Environmental Reviews</em>, <em>28</em>(4), 478-505. https://doi.org/10.1139/er-2020-0019</li>
<li>Li, Z., Wu, H., Duan, S., Zhao, W., Ren, H., Liu, X. (2022). Satellite Remote Sensing of Global Land Surface Temperature: Definition, Methods, Products, and Applications. <em>Reviews of Geophysics</em>, <em>61</em>(1). https://doi.org/10.1029/2022rg000777</li>
<li>Khan, N., Ray, R. L., Sargani, G. R., Ihtisham, M., Khayyam, M., Ismail, S. (2021). Current Progress and Future Prospects of Agriculture Technology: Gateway to Sustainable Agriculture. <em>Sustainability</em>, <em>13</em>(9), 4883-4883. https://doi.org/10.3390/su13094883</li>
<li>Fargione, J., Bassett, S., Boucher, T., Bridgham, S. D., Conant, R. T., Cook‐Patton, S. C. (2018). Natural climate solutions for the United States. <em>Science Advances</em>, <em>4</em>(11), eaat1869-eaat1869. https://doi.org/10.1126/sciadv.aat1869</li>
<li>Cawse‐Nicholson, K., Townsend, P. A., Schimel, D., Assiri, A., Blake, P. L., Buongiorno, M. F. (2021). NASA's surface biology and geology designated observable: A perspective on surface imaging algorithms. <em>Remote Sensing of Environment</em>, <em>257</em>, 112349-112349. https://doi.org/10.1016/j.rse.2021.112349</li>
<li>Zhang, X., Liu, L., Wu, C., Chen, X., Gao, Y., Xie, S. (2020). Development of a global 30 m impervious surface map using multisource and multitemporal remote sensing datasets with the Google Earth Engine platform. <em>Earth system science data</em>, <em>12</em>(3), 1625-1648. https://doi.org/10.5194/essd-12-1625-2020</li>
<li>Traganos, D., Poursanidis, D., Aggarwal, B., Chrysoulakis, N., Reinartz, P. (2018). Estimating Satellite-Derived Bathymetry (SDB) with the Google Earth Engine and Sentinel-2. <em>Remote Sensing</em>, <em>10</em>(6), 859-859. https://doi.org/10.3390/rs10060859</li>
<li>Wang, J., Zhen, J., Hu, W., Chen, S., Lizaga, I., Zeraatpisheh, M. (2023). Remote sensing of soil degradation: Progress and perspective. <em>International Soil and Water Conservation Research</em>, <em>11</em>(3), 429-454. https://doi.org/10.1016/j.iswcr.2023.03.002</li>
<li>Tariq, A., Shu, H. (2020). CA-Markov Chain Analysis of Seasonal Land Surface Temperature and Land Use Land Cover Change Using Optical Multi-Temporal Satellite Data of Faisalabad, Pakistan. <em>Remote Sensing</em>, <em>12</em>(20), 3402-3402. https://doi.org/10.3390/rs12203402</li>
</ol>
</article>