Daily Streamflow Forecasting in Mountainous Catchment Using XGBoost, LightGBM and CatBoost

Streamflow forecasting in mountainous catchments is and will continue to be one of the important hydrological tasks. In recent years machine learning models are increasingly used for such forecasts. A direct comparison use three gradient boosting (XGBoost, LightGBM CatBoost) forecast daily streamflow catchment our main contribution. As predictors we precipitation, runoff at upstream gauge station two-day preceding observations. All algorithms simple implement Python, fast robust. Compared deep (like LSTM), they allow easy interpretation significance predictors. tested achieved Nash-Sutcliffe model efficiency (NSE) range 0.85–0.89 RMSE 6.8–7.8 m3s?1. minimum 12 training data series required a result. The XGBoost did not turn out best forecast, although it most popular model. Using default parameters, results were obtained with CatBoost. By optimizing hyperparameters, by LightGBM. differences between much smaller than within themselves when suboptimal hyperparameters used.