Modeling Salinity-Vegetation Interaction in Semi-Arid Areas: A Google Earth Engine Based Comparative Regression Analysis

Soil salinization is one of the serious environmental concerns around the world, especially in arid and semi-arid ecosystems, where it affects the agricultural productivity and ecological balance. In this study, the effect of salinization on vegetation was investigated in the context of the Kayseri Palas Tuzla Lake Basin, an area identified as a strictly protected sensitive zone, and it is an important wetland in the Central Anatolia Region of the country, with the integration of remote sensing with advanced statistical analysis. Sentinel-2 multi-spectral satellite images captured during the summer of 2024 were analyzed using Google Earth Engine (GEE) software. The dataset, derived entirely from within the hydrological boundary of the basin, was randomly split into 70% for training and 30% for testing and validation. In order to verify the temporal consistency of the proposed approach, an analysis with a long time span between 2016 and 2024 was performed as well. Normalized Difference Vegetation Index (NDVI) was chosen as the dependent variable, with Salinity Index (SI), Bare Soil Index (BSI), and Vegetation Soil Salinity Index (VSSI) variables introduced as the predictors in the context of Simple Linear Regression, Third-Degree Polynomial Regression, and Multiple Linear Regression (MLR) analysis. The straightforward linear regression analysis had very low explanatory power, but much better results were obtained with polynomial regression analysis. Using all environmental factors, the regression analysis was found very robust with high accuracy, yielding R² = 0.956 (RMSE = 0.034) on the training set, as well as R² = 0.950 (RMSE=0.035) on the test set. Moreover, statistical analysis of the coefficients indicated that the coefficient for the ‘salinization’ factor (SI) most strongly (and negatively) affected vegetation.