2026-03-17 ミネソタ大学
A new flood forecasting model developed at the University of Minnesota combines elements of the traditional physics-based models with newer machine-learning techniques.
＜関連情報＞
- https://cse.umn.edu/college/news/ai-model-could-revolutionize-flood-forecasting
- https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2024WR039064
知識誘導型機械学習による洪水予測の実用化 Knowledge-Guided Machine Learning for Operational Flood Forecasting
Zac McEachran, Rahul Ghosh, Arvind Renganathan, Somya Sharma, Kelly Lindsay, Michael Steinbach, John Nieber, Christopher Duffy, Vipin Kumar
Water Resources Research Published: 13 November 2025
DOI:https://doi.org/10.1029/2024WR039064
Abstract
We present a knowledge-guided machine learning framework for operational hydrologic forecasting at the catchment scale. Our approach, a Factorized Hierarchical Neural Network (FHNN), has two main components: inverse and forward models. The inverse model uses observed precipitation, temperature, and streamflow data to generate a representation of the current underlying catchment state. The forward model predicts streamflow using the learned catchment state. The FHNN architecture is designed to model multi-scale processes and capture their interactions, a critical ability for flood modeling. FHNN also improves forecasts based on real-time data through an inference-based data integration approach using inverse modeling. FHNN’s data integration approach improves forecasts in response to observed data more efficiently than data assimilation methods that require computationally intensive optimization. We compare the FHNN to a leading deep learning alternative (autoregressive LSTM) on the large-sample CAMELS-US data set, and operational flood forecast data from the US National Weather Service (NWS). Official NWS flood forecasts are generated by expert human forecasters using a physics-based model, in a human-in-the-loop process. Thus, we assess the flood forecast ability of FHNN by directly comparing its performance against these NWS expert-derived forecasts. The human forecaster creates a more accurate forecast within the first 12–18 hr of a forecast’s issuance, but FHNN has significantly better predictions thereafter. This research lays the groundwork for leveraging the predictive performance of AI-based models with the expertise in forecasting agencies to produce better river forecasts.
Plain Language Summary
Recent advances in Machine Learning (ML) have made strides in improving the prediction of river levels, and often outperform physics-based models that make a forecast based on the physics of how rivers respond to precipitation. However, expert human forecasters with field intelligence know how to modify and improve physics-based models to achieve high accuracy. Our research responds to key needs in the operational forecasting community: (a) by making direct comparisons between ML models and official forecasts that are based on both physics-based models and forecaster expertise, and (b) by combining the strengths of pure physics-based and ML approaches. We have developed a new “Knowledge-Guided” ML (KGML) model with its algorithm informed directly by hydrologic science, called the Factorized Hierarchical Neural Network, and demonstrate that it performs as well or better than NWS flood forecasts 2–7 days after a forecast is issued, and better than a leading ML alternative that does not incorporate physical science knowledge in its architecture. An expert human forecaster using a physics-based model is still more skilled than the state-of-the-art ML methods within the first day. If forecasters could use our KGML approach operationally, the skill of river forecasts has the potential to improve substantially.
