Interpretable Baseflow Segmentation and Prediction Based on Numerical Experiments and Deep Learning
qiying yu, Caihong Hu · 2024
Baseflow is a crucial water source in the inland river basins of high-altitude cold regions, playing a significant role in maintaining runoff stability. Analyzing the impact of climate change and underlying surface conditions on base flow, based on scientifically separating baseflow, is helpful for maintaining river ecological health and rational water resource allocation, especially in severely water-scarce high-altitude cold regions. The challenge lies in selecting the most suitable base flow separation method in data-scarce high-altitude cold regions, qualitatively analyzing the effects of climate factors and underlying surface changes on baseflow values and seasonal distribution characteristics, and providing interpretable scientific predictions for baseflow changes. Therefore, this study aims to contribute further to cold region hydrology by addressing the gap in understanding how meteorological factors and underlying surface changes under the backdrop of climate change affect base flow more reasonably and comprehensively. The study introduces the Grey Wolf Optimizer Digital Filter Method (GWO-DFM) for rapid baseflow separation, utilizes the Long Short-Term Memory (LSTM) neural network model for scientific base flow prediction, and explores the interpretability of the LSTM model in base flow forecasting. The proposed method was successfully implemented using a 63-year time series (1958-2020) of flow data from the Tairan River basin in the high-altitude cold region, along with 21 years of ERA5 meteorological data and MODIS data (2000-2020). The results indicate that: (1) GWO-DFM can rapidly identify optimal filtering parameters, and compared with three other methods (Eckhardt filter, Boughton-Chapman, and Chapman-Maxwell), the average base flow separation using GWO-DFM as the best method for the high-altitude cold region significantly increased after the second baseflow rate mutation. (2) Baseflow sources are mainly influenced by precipitation infiltration, glacier frozen soil layers, and seasonal ponding. (3) Solar radiation, temperature, precipitation, and NDVI are the primary factors influencing base flow changes, with Nash-Sutcliffe efficiency coefficients exceeding 0.78 in both the LSTM model training and prediction periods. (4) Changes in base flow are most influenced by solar radiation, temperature, and NDVI.