Quick link to the deepLDB.org Landslide project website

Quick link to the soil moisture forecast website

Quick link to Our team on AGU TV

Quick link to Our team's code repos and benchmarks

 

Research Interests

Our research focuses on advancing fundamental understanding of the interactions between hydrology and other subsystems (e.g., ecosystem, energy and carbon cycles, solid earth and channels). Water scarcity and excess create varied conflicts and competitions in different parts of the world, and drastic changes in the water cycle put stress on natural and societal systems. Importantly, the changes in water states and flows are a significant driver for changes in other systems. We strive to provide sound physical science, produced by data, data-driven and process-based models, to support decision-making across multiple scales, from catchment to global scales. Meanwhile, our fundamental understanding of the hydrologic cycle, after decades of research, still remains much to be improved. We strive to identify commonalities and learn underlying principles.

Our primary methods include (1) state-of-the-art deep learning (DL); and (2) physically-based hydrologic models. Recently, we demonstrated a genre of physics-informed machine learning methods called "differentiable modeling". Differentiable models mix process-based equations (called priors) and neural networks (NNs) at the fundamental level so they can be trained together in one stage (called "end-to-end"). This way, the network components can be supervised indirectly by outputs of the combined system, and do not necessarily need training data for its direct outputs. Differentiability can be supported by automatic differentiation, adjoint, or any other method that can produce the gradients of loss with respect to large amounts of parameters efficiently. Such models can train one neural network using big data, improving the generalizability, robustness, and complexity of the learned relationships.  We have found massive scale, efficiency and performance advantages with differentiable model in rainfall-runoff , routing, ecosystem and wate quality modeling. Our newest version even surpasses state-of-the-art LSTM models in data-dense regions. See our benchmarks.  

We are opening sourcing our hydrologic deep learning and differentiable model code.

Besides machine learning, another tool we used in the past  is the Process-based Adaptive Watershed Simulator (PAWS), a comprehensive, computationally-efficient parallel hydrologic model designed for large-scale simulation. The model is coupled to the Community Land Model (CLM), and therefore is able to simulate Carbon/Nitrogen cycling, ecosystem dynamics and their interactions with the water cycle. PAWS is now open to all users. READ repo access will be granted to anyone who request it by sending an email to cshen@engr.psu.edu.

 

 

Recent news

Please read our paper on differentiable parameter learning (dPL) and differentiable modeling below! We believe differentiable modeling combines the best of neural networks and physical equations and represents the next stage of modeling.

 

 

Shen's book chapter Applications of Deep Learning in Hydrology is now online.

A major achievement on connecting physics and machine learning is to be published online. Check back here recently!

PhD student Dapeng Feng's GRL paper, on streamflow prediction in continuously data-sparse regions, is now online. Our code (find link above) trains such a prediction system for CAMELS data in about an hour.

PhD student Farshid Rahmani's first paper on stream temperature modeling is online.

Through our volunteer work, our group has just brought online a deep-learning-based tool to track soil moisture changes in Africa/Asia to help combat the locust swarms which threatens global food security.

We appreciate our funding organizations:

National Science Foundation presents FY 2021 budget request | NSF - National  Science FoundationSC Logos | U.S. DOE Office of Science (SC)

 

 

Past News

Older news ...