Hierarchically Disentangled Recurrent Network for Factorizing System Dynamics of Multi-scale Systems: An application on Hydrological Systems

2024-07-29

AI Summary

To address the challenges of dynamic modeling for multi-scale hydrological systems and river flow forecasting, this paper proposes the Hierarchical Decoupled Recurrent Network (HDRN). HDRN innovatively decomposes system dynamics across temporal scales and enables cross-scale interaction and factor-wise decoupling via a unified inverse-forward modeling framework. The method supports time-series learning driven by physics-based simulation, observational data, or their fusion, and incorporates pretraining and global transfer strategies to enhance generalization under limited-data regimes. Evaluated across multiple basins in the NOAA National Water Center’s North Central River Forecast Center and the continental-scale CAMELS dataset, HDRN achieves statistically significant improvements over conventional physics-based models and Transformer-based baselines for 1–7-day streamflow forecasting—particularly excelling in low-runoff-ratio and cold-climate basins.

Technology Category

• Machine Learning: Time-Series/Data Streams
• Natural Language Processing: Language Grounding & Multi-modal NLP
• Cognitive Modeling & Cognitive Systems: Other Foundations of Cognitive Modeling & Systems

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• User Modeling, Personalization and Recommendation: Federated recommendation systems and personalization

Abstract

We present a framework for modeling multi-scale processes, and study its performance in the context of streamflow forecasting in hydrology. Specifically, we propose a novel hierarchical recurrent neural architecture that factorizes the system dynamics at multiple temporal scales and captures their interactions. This framework consists of an inverse and a forward model. The inverse model is used to empirically resolve the system's temporal modes from data (physical model simulations, observed data, or a combination of them from the past), and these states are then used in the forward model to predict streamflow. Experiments on several catchments from the National Weather Service North Central River Forecast Center show that FHNN outperforms standard baselines, including physics-based models and transformer-based approaches. The model demonstrates particular effectiveness in catchments with low runoff ratios and colder climates. We further validate FHNN on the CAMELS (Catchment Attributes and MEteorology for Large-sample Studies), which is a widely used continental-scale hydrology benchmark dataset, confirming consistent performance improvements for 1-7 day streamflow forecasts across diverse hydrological conditions. Additionally, we show that FHNN can maintain accuracy even with limited training data through effective pre-training strategies and training global models.

Problem

• Modeling multi-scale processes in hydrological systems
• Factorizing system dynamics at multiple temporal scales
• Improving streamflow forecasting accuracy across diverse conditions

Innovation

• Hierarchical RNN for multi-scale dynamics factorization
• Inverse-forward model for empirical state resolution
• Effective pre-training with limited data