Thermal hotspot detection in landfills using high-resolution land surface temperature data: A case study of active and closed sites.

Landfills play a crucial role in municipal solid waste management and exhibit extreme surface thermal activity due to various physical and chemical processes occurring within these sites. Although previous studies emphasize the need for monitoring post-closure landfills due to ongoing subsurface activity, there is a lack of research utilizing high-resolution land surface temperature (LST) data for this purpose. Addressing this gap, the present study investigates the thermal behaviour of two landfill sites in Istanbul, Türkiye - an active site (Kömürcüoda) and a closed site (Odayeri) - using machine learning (ML)-based downscaling techniques. Landsat 8 and Landsat 9 LST data were enhanced to 10-m spatial resolution by incorporating three spectral indices (normalized difference vegetation index, normalized difference built-up index, normalized difference water index) from Sentinel-2 imagery. A monthly observation period was established for the year 2023-2024. To optimize the downscaling process, a wide range of regression algorithms - Ensemble, Gaussian process (Gp), kernel, linear, neural network (Net), support vector machine and Decision Tree - were evaluated within an automated ML framework. Results showed that Net performed best for the Kömürcüoda Landfill, whereas Gp was most successful for the Odayeri Landfill. The downscaled LST data exhibited strong agreement with the original datasets, with root mean square error values ranging from 0.98°C to 2.01°C for Kömürcüoda and from 0.74°C to 2.38°C for Odayeri. Hotspot analysis revealed persistent high-temperature zones in areas where waste was actively stored or had been stored in the past. Notably, despite being closed, the Odayeri Landfill exhibited ongoing thermal activity, suggesting that landfill surface temperatures can remain elevated for extended periods.