Machine learning-based modeling of land surface temperature at an open dumpsite in Khulna, Bangladesh

Open dumping of untreated waste is a very common practice in developing countries like Bangladesh. The biodegradation of this waste generates heat and gases that are often released without control, creating operational challenges in landfills. However, prior studies have not clearly quantified the resulting temperature increase. To address this gap, this study predicts the Land Surface Temperature (LST) of the Rajbandh landfill in Khulna using Generalized Additive Models (GAM), Multiple Linear Regression (MLR), and Artificial Neural Networks (ANN). The amount of waste received at the open dump was collected from the municipality’s logbook. LandGEM has been used for modeling methane emissions. NDBI, NDVI, specific humidity, wind speed and LST were obtained from Google Earth Engine. The GAM results show that all predictors of LST are statistically significant (p < 0.001). A strong cubic relationship was found between humidity and LST (R2 = 0.92), while NDVI and wind speed exhibited strong linear correlations (R2 = 0.73 and 0.71, respectively). Variable importance analysis highlighted that methane and humidity emerged as the most influential variables, contributing 25.40% and 24.35% to LST variation, whereas NDBI had a minor impact. The MLR also showed statistical significance (p < 0.001) with a moderate predictive power with an R2 of 0.60. These results confirm that combining remote sensing and waste management data enables accurate, low-cost LST prediction for proactive landfill monitoring in resource-constrained regions.