Comparison of LSTM and SVM methods through wavelet decomposition in drought forecasting

Many researchers are working to prevent, monitor and identify drought, which is one of the most insidious and dangerous natural disasters that negatively affects life. For this purpose, various drought indices are developed and new methods are proposed. One of the most widely used of these indexes is the Standard Precipitation Index (SPI). Since it is not known when the drought will begin, taking preventive measures is a difficult and challenging task. In the last decade, machine learning techniques have been preferred to increase success in predicting droughts. In this study, SPI was used as the drought index and Support Vector Machine (SVM) and Long-Short Term Memory Network (LSTM) methods, which are increasingly reliable among the most preferred machine learning and deep learning methods in drought predictions, were used as the prediction method, and furthermore, to increase the prediction power of these methods, new powerful models have been proposed using Wavelet transform and Variational mode transform. Support Vector Machines with Wavelet decomposition (SVM-W), Long-Short Term Memory Networks with Wavelet decomposition (LSTM-W), and Long Short Term Memory Networks with Variational Mode Decomposition (LSTM-VMD) were used as prediction models for drought analysis and the performances of these models were compared.