Hydrological Drought Forecasting Using a Deep Transformer Model

Hydrological drought forecasting is essential for effective water resource management planning. Innovations in computer science and artificial intelligence (AI) have been incorporated into Earth research domains to improve predictive performance planning disaster management. Forecasting of future hydrological can assist with mitigation strategies various stakeholders. This study uses the transformer deep learning model forecast drought, a benchmark comparison long short-term memory (LSTM) model. These models were applied Apalachicola River, Florida, two gauging stations located at Chattahoochee Blountstown. Daily stage-height data from period 1928–2022 collected these stations. The used predict stage five different time steps: 30, 60, 90, 120, 180 days. A series was created forecasted values using monthly fixed threshold 75th percentile (75Q). outperformed LSTM all timescales both locations when considering following averages: MSE=0.11, MAE=0.21, RSME=0.31, R2=0.92 station, MSE=0.06, MAE=0.19, RSME=0.23, R2=0.93 Blountstown station. exhibited greater accuracy generating same as observed after applying 75Q threshold, few exceptions. Considering evaluation criteria, accurately forecasts which could be helpful this area contested resources, likely has broad applicability elsewhere.