Improvement of Forecasting Drought by Means of Various Machine Learning Algorithms and Wavelet Transformation

Abstract

Defined as a decrease in average rainfall amounts, drought is one of the most insidious natural disasters. When it starts, people may not be aware of it, that's why droughts are difficult to monitor. Scientists have long been working to predict and monitor droughts. For this purpose, they have developed many methods such as drought indices one of which Standardized Precipitation Index (SPI) is. In this study, SPI to detect droughts and machine learning algorithms, support vector machines (SVM), artificial neural networks (ANN), random forest (RF), decision tree (DT), frequently used in the literature to predict droughts and 3 different statistical methods: correlation coefficient (r), Root Mean-square Error (RMSE), Nash-Sutcliffe Efficiency (NSE) Coefficient to investigate model performance values were used. Wavelet analysis was also applied to improve model performances. Konya closed basin located in the middle of Türkiye in terms of location and is among the leading regions of Turkey in terms of grain is one of the regions most affected by droughts in Türkiye. One of the most important water resources of the region is the Apa dam. It provides water to many fields which fertile land in its vicinity and is affected by droughts. Therefore, this region was determined as the study area. Meteorological data, total monthly precipitation, that could represent the region were obtained between 1955 and 2020 from general directorate of state water works and general directorate of meteorology. The results show that among the models analyzed with machine learning algorithms, the best results were obtained from M04 model whose input structure was created from SPI, different times steps, data delayed up to 5 months and total monthly precipitation data for time t-1. Among machine learning algorithms, SVM has achieved the most successful results in not only without wavelet transform (WT) but also with WT. Effective results were obtained from M04 in which SVM with WT was used (NSE = 0.9942, RMSE = 0.0764, R = 0.9971).