Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Hierarchical Conditional Multi-Task Learning for Streamflow Modeling

Shaoming Xu, Arvind Renganathan, Ankush Khandelwal, Rahul Ghosh, Xiang Li, Licheng Liu, Kshitij Tayal, Peter Harrington, Xiaowei Jia, Zhenong Jin, Jonh Nieber, Vipin Kumar

TL;DR

H Hierarchical Conditional Multi-Task Learning (HCMTL) is proposed, a hierarchical approach that jointly models soil water and snowpack processes based on their causal connections to streamflow and incorporates the Conditional Mini-Batch strategy to improve long time series modeling.

Abstract

Streamflow, vital for water resource management, is governed by complex hydrological systems involving intermediate processes driven by meteorological forces. While deep learning models have achieved state-of-the-art results of streamflow prediction, their end-to-end single-task learning approach often fails to capture the causal relationships within these systems. To address this, we propose Hierarchical Conditional Multi-Task Learning (HCMTL), a hierarchical approach that jointly models soil water and snowpack processes based on their causal connections to streamflow. HCMTL utilizes task embeddings to connect network modules, enhancing flexibility and expressiveness while capturing unobserved processes beyond soil water and snowpack. It also incorporates the Conditional Mini-Batch strategy to improve long time series modeling. We compare HCMTL with five baselines on a global dataset. HCMTL's superior performance across hundreds of drainage basins over extended periods shows that integrating domain-specific causal knowledge into deep learning enhances both prediction accuracy and interpretability. This is essential for advancing our understanding of complex hydrological systems and supporting efficient water resource management to mitigate natural disasters like droughts and floods.

Hierarchical Conditional Multi-Task Learning for Streamflow Modeling

TL;DR

H Hierarchical Conditional Multi-Task Learning (HCMTL) is proposed, a hierarchical approach that jointly models soil water and snowpack processes based on their causal connections to streamflow and incorporates the Conditional Mini-Batch strategy to improve long time series modeling.

Abstract

Streamflow, vital for water resource management, is governed by complex hydrological systems involving intermediate processes driven by meteorological forces. While deep learning models have achieved state-of-the-art results of streamflow prediction, their end-to-end single-task learning approach often fails to capture the causal relationships within these systems. To address this, we propose Hierarchical Conditional Multi-Task Learning (HCMTL), a hierarchical approach that jointly models soil water and snowpack processes based on their causal connections to streamflow. HCMTL utilizes task embeddings to connect network modules, enhancing flexibility and expressiveness while capturing unobserved processes beyond soil water and snowpack. It also incorporates the Conditional Mini-Batch strategy to improve long time series modeling. We compare HCMTL with five baselines on a global dataset. HCMTL's superior performance across hundreds of drainage basins over extended periods shows that integrating domain-specific causal knowledge into deep learning enhances both prediction accuracy and interpretability. This is essential for advancing our understanding of complex hydrological systems and supporting efficient water resource management to mitigate natural disasters like droughts and floods.

Paper Structure

This paper contains 42 sections, 9 equations, 12 figures, 8 tables.

Table of Contents

  1. Introduction
  2. Hierarchical Conditional Multi-Task Learning (HCMTL)
  3. Hierarchical Multi-Task Network (HMTnet)
  4. Embedding-Based Connections
  5. Conditional Mini-Batch (CMB) Learning
  6. Background
  7. Hydrological cycle
  8. LSTM for modeling Dynamical Systems
  9. Single-Task Learning (STL)
  10. Multi-Task Learning (MTL)
  11. Simultaneous Multi-Task Learning (SMTL)
  12. Hierachical Multi-Task Learning(HMTL)
  13. Mini-batch learning strategies
  14. Experiment Settings
  15. Data
  16. ...and 27 more sections

Figures (12)

  • Figure 1: Hierarchical Conditional Multi-Task Learning for streamflow (SF) modeling during inference. Each segment contains t time steps, with the predicted Snowpack (SNO) and Soilwater (SW) from the current segment used to initialize the next.
  • Figure 2: An abstraction of the hydrological cycle.
  • Figure 3: Model Setup and Baselines.
  • Figure 4: HCMTL achieves the best performance in 123 out of 319 basins in the United States.
  • Figure 5: All models show higher RMSEs as segment length decreases, as shorter segments capture less temporal information and result in a greater loss of interactions between segments.
  • ...and 7 more figures