A Bayesian Approach to Integrate and Predict Land Cover Changes in the Context of Land Surface Temperature Variations Using Multispectral Remote Sensing Data

Abstract

Land Surface Temperature (LST) is an important geophysical parameter in global energy balance studies and hydrologic modeling. Many studies have suggested that there is a strong correlation between LST and Land Cover (LC) changes due to its sensitivity to vegetation cover. Due to rapid urbanization and population growth, significant changes in LC are caused, which directly contribute to climate change through a variety of processes causing changes in LST. Thermal remote sensing is sensitive to an object's temperature and emittance depending on the wavelength. The objective of this experiment is to build a mathematical model to investigate the relationship between LST and the LC changes in Sri Lanka. The model built in the study could be used to analyze relationships between parameters depending on their frequency distributions, which is quite useful in the applied science domain. The model is based on Bayes' theorem, which determines the conditional probability of the events. Landsat 8 OLI/TIRS images throughout 2015 to 2020 were classified into five LC classes as water, soil, vegetation, impervious, and cloud using Support Vector Machine (SVM) classification. LST was retrieved for the same period using the standard equations for LST retrieval by Landsat 8 using the normalized difference vegetation index (NDVI). Multispectral bands were used for the LC classification, and thermal bands were used to obtain the LST. Vegetation cover has been significantly affected by the changes in LST, while other LC types show a less significant relationship with the LST. The novelty of this study lies in applying a Bayesian probabilistic framework to quantify the degree of association between LC types and LST using frequency distributions. Unlike traditional correlation-based approaches, this model enables a more robust conditional probability estimation, offering a more nuanced understanding of how each LC type interacts with LST variations. This probabilistic insight can be particularly useful for predictive land management and climate modeling.