Advanced Spatio-Temporal Analysis of Surface Urban Heat Island Intensity in the Mumbai Metropolitan Region Using Surface Imperviousness, Land Use- Land Cover, and Local Climate Zones Via Google Earth Engine and Deep Learning (2003–2023)

Abstract Urbanisation significantly modifies land surfaces and amplifies local temperatures and heat stress. This study investigates Surface Urban Heat Island Intensity (SUHII) across the Mumbai Metropolitan Region (MMR) from 2003 to 2023 using Google Earth Engine (GEE) database and mapping. It integrates surface imperviousness from the Global Human Settlement Layer (GHSL), Land Use Land Cover (LULC), and the Local Climate Zone (LCZ) classification (2019) to explore spatial patterns in SUHII. Daily Land Surface Temperature (LST) data from MODIS was further analysed for details across 852 spatially sampled points, categorised by land cover types. Findings reveal a consistent urban–rural SUHII of ~2 °C, with mean LSTs of 35.63 °C in urban and 33.67 °C in rural zones. Urban cores exhibit greater seasonal variability, with LSTs peaking above 51 °C in some areas. Auto Regressive Integrated Moving Average - ARIMA (2,1,1) time-series modelling indicates persistent warming trends, with 2023 pre-monsoon LSTs projected to exceed 44 °C in central MMR with an average of existing 51 °C in urban MMR and about 49 °C in rural MMR. A weak negative Pearson correlation (r = -0.19) between impervious surface intensity and SUHII suggests that built-up extent alone does not explain thermal intensity. LCZ-based profiling shows that Compact High-Rise and Industrial zones have the highest LSTs, while vegetated zones maintain cooler profiles (<31 °C). Zones with >70 % impervious surfaces record disproportionately higher temperatures. Importantly, a review of existing studies shows that no published research has yet combined ARIMA forecasting with LOESS (Locally Estimated Scatterplot Smoothing) for MMR. This study uniquely combines remote sensing, statistical modelling (ARIMA, Pearson correlation), and urban climate zoning via cloud computing for the region. Deep Learning derived spatial datasets (GHSL, LCZ) enhance the spatial resolution of SUHII analysis. The results offer vital insights for climate-adaptive urban planning, emphasising zoning-based interventions, and landscape strategies to mitigate urban heat risks in expansive cities like Mumbai and its surroundings.