Forecasting streamflow in Colorado by deep learning and NASA satellite remote sensing

In this study, Wavelet Long Short-Term Memory (WLSTM) and Wavelet Convolutional Neural Network Long Short-Term Memory (WCNNLSTM) are two Artificial Intelligence (AI) methods that are applied for the subseasonal-to-seasonal (S2S) streamflow forecasting in seven rivers in Colorado. Daily and monthly meteorological variables from 1992 to 2022 were collected from Colorado’s Mesonet (CoAgMet) as input data. Daily and monthly discharge data were acquired from the United States Geological Survey (USGS) as output data. In addition, precipitation data were gathered from the National Aeronautics and Space Administration (NASA) global precipitation network, including the Tropical Rainfall Measuring Mission (TRMM) and the Integrated Multi-satellitE Retrievals for Global Precipitation Measurement (GPM) (IMERG). WCNNLSTM-IMERG had the best accuracy in forecasting streamflow in Colorado due to its finer spatial resolution. The findings of this study will be beneficial for modelers and dataset developers in evaluating the performance of NASA products in practice for S2S streamflow forecasting.