Integration of machine learning and remote sensing for drought index prediction: A framework for water resource crisis management

Abstract

A drought is a complex event characterized by low rainfall and has negative implications for agricultural and hydrological systems, as well as for community life. A common meteorological drought index used for drought monitoring and water resource management is the Standardized Precipitation Evapotranspiration Index (SPEI). Using SPEI can assist in predicting drought onset and estimating drought severity. The objective of this research is to assess the accuracy of machine learning models in estimating the SPEI-1 (one-month) index in semi-arid climates. To achieve this goal, the data will be analyzed using remote sensing parameters, a worldwide database, and meteorological station information. SPEI-1 was predicted in Tabriz, Iran, between 1990 and 2022 using multilayer perceptron (MLP) and random forest (RF) techniques combined with genetic algorithm (GA) methods. The parameters used are average air temperature, average relative humidity, monthly precipitation, wind speed, sunny hours, as well as the one-month standard precipitation index (SPI-1) (from ground data), daily precipitation products from satellites named PERSIANN (PRC-PR) (from remote sensing), and SPEIbase data (from global databases). The results suggest that the use of satellite remote sensing characteristics and global databases has significantly enhanced the precision and efficiency of prediction models. Based on the GA-RF model with an R² of 0.992 and an RMSE of 0.124, it exhibits the best performance among all models in Scenario 1. By combining remote sensing parameters, this study presents an innovative approach to predicting the SPEI index and demonstrates their capabilities in drought management and mitigation.