How Well Can Machine Learning Models Perform without Hydrologists? Application of Rational Feature Selection to Improve Hydrological Forecasting

With more machine learning methods being involved in social and environmental research activities, we are addressing the role of available information for model training performance. We tested abilities several models short-term hydrological forecasting by inferring linkages with all predictors or only those pre-selected a hydrologist. The used this study were multivariate linear regression, M5 tree, multilayer perceptron (MLP) artificial neural network, long memory (LSTM) model. two river catchments contrasting runoff generation conditions to try infer ability different structures automatically select best predictor set from dataset compared models’ performance that operating on prescribed Additionally, how shuffling initial improved can conclude rainfall-driven catchments, performed generally better hydrologist, while mixed-snowmelt baseflow-driven automatic selection was preferable.