Based on Improved Long Short-Term Memory Network Feature Attention Extraction for China Drought Prediction

Agricultural drought significantly impedes crop growth and development. While Deep Learning (DL) has been extensively adopted in meteorological research, particularly using Long Short-Term Memory (LSTM) networks and Convolutional Neural Networks (CNNs) to exploit the memory persistence of soil moisture, existing methodologies often overlook data noise prior to model training and the correlations between predictor and target variables. Given the context of global warming, forecasting agricultural drought with sufficient lead time is critical for formulating proactive water management strategies. To address this, we propose a novel LSTM model incorporating Feature and Temporal Attention Extraction (FAELSTM) to forecast short-term agricultural drought using multivariate meteorological data. The feasibility and novelty of the model are validated by predicting the Soil Moisture Condition Index (SMCI) in China. Experimental results demonstrate that FAELSTM effectively assists decision-makers in devising timely and appropriate agricultural water strategies.