Modeling of land surface temperature (LST) in Ardabil plain using NDVI index and Bayesian neural network approach

The main purpose of this research is to evaluate the relationship between vegetation cover and land surface temperature (LST) and modeling of this relationship in Ardabil plain. To achieve this goal, Landsat images for the years 2010, 2012, 2014, 2016, 2018 and 2020 prepared from the USGS site and the Normalized Difference Vegetation Index (NDVI) and LST calculated from these images for Ardabil plain. The spatial changes of the indicators investigated during 2010 to 2020; then the relationship between them was obtained by Pearson correlation. Finally, using the Bayesian neural network, the average LST of Ardabil Plain was used as the dependent variable and the NDVI index was used as the independent variable. The results showed that the most vegetation cover was in the central areas of Ardabil plain, which belongs to agricultural use, and the NDVI index in these areas varied between 0.55 and 0.67. The maximum LST was 83 degrees Kelvin in the northern mountains and the minimum LST was 37 degrees Kelvin in the of Ardabil plain. The most changes in the NDVI index were between + 0.43 and − 0.47, with the most negative index occurring in the center of Ardabil plain, where agricultural lands have been converted into man-made lands. The most difference in the LST was in the central part of Ardabil plain, which increased up to 26 degrees Kelvin. Pearson's correlation coefficient between NDVI indices and LST was significant (− 0.52) in Ardabil plain, which indicates an inverse relationship between them. The Bayesian neural network model showed that the R coefficient was 0.75 for the test data, 0.74 for the validation and 0.75 for the whole model, which indicates the appropriate fit of the model in predicting the LST with NDVI is used, which means that the NDVI index can estimate up to 75% of the LST changes.