Integrated remote sensing data and machine learning for drought prediction in Eastern Saudi Arabia

Drought has become more common problem and more severe that widespread over the world in recent decades, increasing their negative effects on the environment. The most vulnerable communities can be warned and prepared for the negative effects of droughts. This study aims to assess and predict the drought vulnerability using Landsat satellite images from 2014 to 2022 over El- Dammam City. Using remote sensing, drought phenomenon was simulated and predicted through Perpendicular drought index (PDI) and Normalized Difference Drought Index (NDDI). In the present study, some indices were selected to sasses vegetation as Normalized Difference Vegetation Index (NDVI), water as Normalized Difference Water Index (NDWI) and salinity as salinity index (SI) and Normalized Difference Salinity Index (NDSI) with the aid of Machine Learning (ML) techniques, Support Vector Machine (SVM) and Partial Least Square (PLS). Results showed that NDDI and PDI were negatively correlated with NDVI and NDWI; respectively where NDDI and PDI were positively correlated with NDSI and SI; respectively. Principal Component Analysis (PCA), PC-1, PC-2 and PC-3 are contributed about 98% of all data. Finally, PLS and SVM are two good multivariate analysis for prediction of drought with R2 0.95 and 0.98 respectively. The predicted model showed the severity of drought far away from the sea along the barren area in the southern part. Finally, it should be noted that NDVI, NDSI, SI, and NDWI were better indicators for tracking and assessing changes in drought. The results can be utilized to quickly evaluate data produced from remote sensing and water-related indices and make wise decisions. This study contributes to our understanding of droughts and sheds information on how ML technology could be used to monitor droughts.