Combining deep learning methods and multi-resolution analysis for drought forecasting modeling

This paper proposes a novel approach for drought forecasting based on combining Long Short-Term Memory (LSTM) and Multi-Resolution Analysis Wavelet Transform (MRA-WT), called MRA-WT-LSTM. In fact, Deep Learning (DL) methods provided an outstanding performance in several forecasting fields, especially drought. LSTM has proved a high ability in dealing with time-series drought indices compared to the others existing methods. Therefore, in this study LSTM was used to provide long-term drought forecasts. However, the non-stationarity of the drought indices is a major challenge that needs to be taken into consideration. In this regard, MRA-WT was applied to analyze the non-stationary time-series drought indices. Experiments were carried out in the Sarab region, Iran in the period between 1988–2016 in order to predict the standardized precipitation Evaporation index (SPEI). Drought is the most natural hazards in Sarab region that affect the socioeconomic development. The input data include the station data (i.e. rainfall, temperature (mean, minimum, and maximum), humidity, pressure, and Evaporation) and Normalized Difference Vegetation Index (NDVI) derived from Landsat images. Through the experiment, the proposed methodology is evaluated compared to three ML methods (i.e. Artificial Neural Network (ANN), Support Vector Regression (SVR), Random Forest Regression (RFR)) and LSTM. The different models were compared in terms of statistical metrics such as the coefficient of determination (\({R}^{2}\)), Root-Mean-Square Error (RMSE) and Mean Absolute Percentage Error (MAPE). Results revealed the effectiveness of MRA-WT-LSTM for drought forecasting with \({R}^{2}\) up to 0,93 and RMSE reached 0,02 for the different time lags.