Presenters-0643
From CSDMS
Revision as of 09:51, 5 January 2024 by Mamc9359 (talk | contribs) (Created page with "{{Presenters temp |CSDMS meeting event title=CSDMS 2024: Coastlines, Critical Zones and Cascading Hazards: Modeling Dynamic Interfaces from Deep Time to Human Time |CSDMS meeting event year=2024 |CSDMS meeting presentation type=Invited oral presentation |CSDMS meeting first name=Chaopeng |CSDMS meeting last name=Shen |CSDMS meeting institute=Pennsylvania State University |Country member=United States |CSDMS meeting state=Pennsylvania |CSDMS meeting email address=Cxs1024@...")
CSDMS 2024: Coastlines, Critical Zones and Cascading Hazards: Modeling Dynamic Interfaces from Deep Time to Human Time
Differentiable modeling to unify neural networks and process-based modeling for global geosciences under global change
Abstract
Process-based modeling offers interpretability and physical consistency in many domains of geosciences but struggles to leverage large datasets efficiently. Machine-learning methods, especially deep networks, have strong predictive skills yet are not easily interpretable and are unable to answer specific scientific questions. A recently proposed genre of physics-informed machine learning, called “differentiable” modeling (DM, https://t.co/qyuAzYPA6Y), trains neural networks (NNs) with process-based equations (priors) together in one stage (so-called “end-to-end”) to benefit from the best of both NNs and process-based paradigms. The NNs do not need target variables for training but can be indirectly supervised by observations matching the outputs of the combined model, and differentiability critically supports learning from big data. We propose that differentiable models are especially suitable as global- or continental-scale geoscientific models because they can harvest information from big earth observations to produce state-of-the-art predictions (https://mhpi.github.io/benchmarks/), enable physical interpretation naturally, extrapolate well (due to physical constraints) in space and time, enforce known physical laws and sensitivities, and leverage progress in modern AI computing architecture and infrastructure. Differentiable models can also synergize with existing process-based models in terms of providing to them parameters or identifying optimal processes, learning from the lessons of the community. Differentiable models can answer pressing societal questions on water resources availability, climate change impact assessment, water management, and disaster risk mitigation, among others. We demonstrate the power of differentiable modeling using computational examples in rainfall-runoff modeling, river routing, ecosystem and water quality modeling, and forcing fusion. We discuss how to address potential challenges such as implementing gradient tracking for implicit numerical schemes and addressing process tradeoffs. Furthermore, we show how differentiable modeling can enable us to ask fundamental questions in hydrologic sciences and get robust answers from big global data.
Please acknowledge the original contributors when you are using this material. If there are any copyright issues, please let us know (CSDMSweb@colorado.edu) and we will respond as soon as possible.
Of interest for:
