A Medium and Long-Term Runoff Forecast Method Based on Massive Meteorological Data and Machine Learning Algorithms

Accurate and reliable predictors selection model construction are the key to medium long-term runoff forecast. In this study, 130 climate indexes utilized as primary forecast factors. Partial Mutual Information (PMI), Recursive Feature Elimination (RFE) Classification Regression Tree (CART) respectively employed typical algorithms of Filter, Wrapper Embedded based on Selection (FS) obtain three final schemes. Random Forest (RF) Extreme Gradient Boosting (XGB) constructed representative models Bagging Ensemble Learning (EL) realize types lead time which contains monthly, seasonal annual sequences Three Gorges Reservoir in Yangtze River Basin. This study aims summarize compare applicability accuracy different FS methods EL The results show following: (1) RFE method shows best performance all time. (2) RF XGB suitable for but presents better skills both calibration validation. (3) With increase magnitudes, reliability improved. However, it is still difficult establish accurate forecasts only large-scale used. We conclude that theoretical framework Machine could be useful water managers who focus