Towards learning human influences in a highly regulated basin using a hybrid DL-process based framework

• DOI: 10.5194/egusphere-egu24-4325
• Publication Date: 2024-03-08T13:34:41Z
• Authors: Liangkun Deng; Xiang Zhang; Louise Slater

Hybrid models have shown impressive performance for streamflow simulation, offering better accuracy than process-based hydrological (PBMs) and superior interpretability deep learning (DLMs). A recent paradigm modeling, integrating DLMs PBMs within a differentiable framework, presents considerable potential to match the of while simultaneously generating untrained variables that describe entire water cycle. However, this framework has mostly been verified in small unregulated headwater basins not explored large highly regulated basins. Human activities, such as reservoir operations transfer projects, greatly changed natural regimes. Given limited access operational management records, generally fail achieve satisfactory are challenging train directly. This study proposes coupled hybrid address these problems. is based on distributed PBM, Xin'anjiang (XAJ) model, adopts embedded neural networks learn physical parameters replace modules XAJ model reflecting human influences through structure. Streamflow observations alone used training targets, eliminating need records supervise process. The Hanjiang River basin (HRB), one largest subbasins Yangtze basin, disturbed by reservoirs national selected test effectiveness framework. results show can best parameter sets depicting improve simulation. It performs standalone achieves similar LSTM model. sheds new light assimilating simulation river with records.