A novel hybrid XAJ-LSTM model for multi-step-ahead flood forecasting

Abstract The conceptual hydrologic model has been widely used for flood forecasting, while long short-term memory (LSTM) neural network demonstrated a powerful ability to tackle time-series predictions. This study proposed novel hybrid by combining the Xinanjiang (XAJ) and LSTM (XAJ-LSTM) achieve precise multi-step-ahead forecasts. takes forecasts of XAJ as input variables enhance physical mechanism hydrological modeling. Using models benchmark comparison purposes, was applied Lushui reservoir catchment in China. results that three could offer reasonable XAJ-LSTM not only effectively simulate long-term dependence between precipitation datasets, but also create more accurate than models. maintained similar forecast performance after feeding with simulated values during horizons . concludes integrates machine learning can raise accuracy improving interpretability data-driven internals.