Estimating Urban Land Surface Temperature Using Spatial Machine Learning in Gorontalo City

Abstract

Urban expansion in tropical cities significantly alters surface thermal conditions, intensifying the urban heat island (UHI) phenomenon. This study aims to estimate and analyze the spatiotemporal dynamics of land surface temperature (LST) in Gorontalo City from 1995 to 2025 using a spatial machine learning (SML) approach based on the Random Forest (RF) algorithm. Multitemporal Landsat 5, 7, 8, and 9 images were processed in Google Earth Engine (GEE) to derive surface reflectance, Normalized Difference Vegetation Index (NDVI), emissivity, and brightness temperature, which were subsequently employed as predictor variables in the LST model. A total of 50 ground validation points were used to assess model performance. The RF model achieved high predictive accuracy with an R² of 0.833, RMSE of ±3.33 °C, and MAE of ±2.80 °C, outperforming conventional NDVI-based models. The long-term analysis revealed a consistent increase in LST across urbanized zones, particularly in the city center and northern districts, while areas with higher vegetation cover exhibited lower LST values. The negative correlation between NDVI and LST (R² = 0.3132) confirms the critical role of vegetation in mitigating urban thermal intensity. These findings highlight the applicability of the RF-based SML framework for accurate LST estimation and urban climate monitoring, providing a scientific basis for sustainable urban planning and green infrastructure development in tropical cities.