Meteorological Drought Trend Analysis and Forecasting Using a Hybrid SG-CEEMDAN-ARIMA-LSTM Model Based on SPI from Rain Gauge Data

Abstract

Abstract. Meteorological drought presents considerable challenges to water supplies, agriculture, and socio-economic stability, especially in areas heavily reliant on precipitation. The Standardized Precipitation Index (SPI) is esteemed for its efficacy in drought monitoring, owing to its straightforwardness and applicability across many time scales. This study examines meteorological drought dynamics in the uMkhanyakude district using the Standardized Precipitation Index (SPI) at 6-, 9-, and 12-month timescales. Trend analysis was conducted using Mann–Kendall (MK), Modified Mann–Kendall (MMK), and Innovative Trend Analysis (ITA) methods. The study also proposes a hybrid model that integrates the Savitzky–Golay (SG) filter, Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN), Autoregressive Integrated Moving Average (ARIMA), and Long Short-Term Memory (LSTM) networks, referred to as SG-CEEMDAN-ARIMA-LSTM, for forecasting of the SPI time series. Analysis of SPI trends and variability revealed statistically significant declining trends at five monitoring stations, characterized by negative Z-scores and p-values, showing a marked downward trajectory across several SPI scales. On the other hand, the forecasting results demonstrate that the SG-CEEMDAN-ARIMA-LSTM methodology outperformed benchmark models across all temporal scales, achieving high prediction accuracy with R2 values of 0.9839 (SPI-6), 0.9892 (SPI-9), and 0.9990 (SPI-12). These findings highlight the effectiveness of decomposition techniques (SG, CEEMDAN) in enhancing model performance and confirm the suitability of the hybrid model for both short-term and long-term drought forecasting. This study merges robust trend analysis with advanced hybrid forecasting techniques, providing a reliable framework for early warning systems and sustainable water resource management in drought-prone regions.