期刊
WATER RESOURCES RESEARCH
卷 58, 期 10, 页码 -出版社
AMER GEOPHYSICAL UNION
DOI: 10.1029/2022WR032404
关键词
rainfall runoff; differentiable programming; machine learning; physical model; differentiable hydrology; LSTM
资金
- US National Science Foundation [1832294, 2221880, 1940190, 2018280]
- Office of Biological and Environmental Research of the U.S. Department of Energy [DE-SC0016605]
- Direct For Computer & Info Scie & Enginr
- Office of Advanced Cyberinfrastructure (OAC) [1940190] Funding Source: National Science Foundation
- Direct For Mathematical & Physical Scien
- Division Of Physics [2018280] Funding Source: National Science Foundation
- Directorate For Geosciences
- Division Of Earth Sciences [1832294] Funding Source: National Science Foundation
This paper introduces a differentiable and learnable process-based model (delta model) that approaches the performance level of purely data-driven deep learning models (such as LSTM) in predicting hydrologic variables. Experimental results show that the delta model performs similarly to LSTM in simulating variables like streamflow and can also output other untrained variables, such as soil and groundwater storage.
Predictions of hydrologic variables across the entire water cycle have significant value for water resources management as well as downstream applications such as ecosystem and water quality modeling. Recently, purely data-driven deep learning models like long short-term memory (LSTM) showed seemingly insurmountable performance in modeling rainfall runoff and other geoscientific variables, yet they cannot predict untrained physical variables and remain challenging to interpret. Here, we show that differentiable, learnable, process-based models (called delta models here) can approach the performance level of LSTM for the intensively observed variable (streamflow) with regionalized parameterization. We use a simple hydrologic model HBV as the backbone and use embedded neural networks, which can only be trained in a differentiable programming framework, to parameterize, enhance, or replace the process-based model's modules. Without using an ensemble or post-processor, delta models can obtain a median Nash-Sutcliffe efficiency of 0.732 for 671 basins across the USA for the Daymet forcing data set, compared to 0.748 from a state-of-the-art LSTM model with the same setup. For another forcing data set, the difference is even smaller: 0.715 versus 0.722. Meanwhile, the resulting learnable process-based models can output a full set of untrained variables, for example, soil and groundwater storage, snowpack, evapotranspiration, and baseflow, and can later be constrained by their observations. Both simulated evapotranspiration and fraction of discharge from baseflow agreed decently with alternative estimates. The general framework can work with models with various process complexity and opens up the path for learning physics from big data.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据