A novel AI approach for modeling land surface temperature of Freetown, Sierra Leone, based on land-cover changes

Land use/land cover (LULC) indices can be considered while developing land surface temperature (LST) models. The relationship between LST and LULC indices must be established to accurately estimate the impacts of LST changes. This study developed novel machine learning models for predicting LST using multispectral Landsat images data of Freetown city in Sierra-Leon. Artificial neural network (ANN) and gene expression programming (GEP) were employed to develop LST prediction models. Images of multispectral bands were obtained from Landsat 4-5 and 8 satellites to develop the proposed models. The extracted data of LULC indices, such as normal difference vegetation index (NDVI), normal difference built-up index (NDBI), urban index (UI), and normal difference water index (NDWI), were utilized as attributes to model LST. The results show that the root-mean-square error (RMSE) of the ANN and GEP models were 0.91(o)C and 1.08 C-o, respectively. The GEP model was used to yield a relationship between LULC indices and LST in the form of a mathematical equation, which can be conveniently used to test new data regarding the thematic area. The sensitivity analysis revealed that UI is the most influential parameter followed by NDBI, NDVI, and NDWI towards contributing LST.

Automated summary

This study developed machine learning models to predict land surface temperature (LST) and established the relationship between LST and land use/land cover (LULC) indices. The results showed that the urban index (UI) had the greatest influence on LST.