Recent studies have shown that deep learning models in hydrological applications significantly improved streamflow predictions at multiple stations compared to traditional machine learning approaches. However, recent studies do not integrate data from upstream to downstream links in generalization investigations. The spatial and temporal generalization ability of deep learning models in hydrology that can be gained by training a single model for multiple stations is evaluated in this study.
In this project, we developed a generalized model with a multi-site structure for hourly streamflow hindcasts on 125 USGS gauged watersheds in the state of Iowa. Considering watershed-scale features including drainage area, time of concentration, slope, and soil types, the proposed models have acceptable performance and slightly higher median NSE value than training individual models for each USGS station. This study demonstrates the potential of deep learning studies in hydrology where more domain knowledge and physical features can support further generalization.
Related Articles
- Xiang, Z., Demir, I., Mantilla, R., & Krajewski, W. F. (2021). A Regional Semi-Distributed Streamflow Model Using Deep Learning. EarthArXiv. (DOI: https://doi.org/10.31223/X5GW3V)
The map visualization of the longest sequenced gauged watershed in the state of Iowa that used in this study.
Flow duration curve of the observation and model 120-hr ahead forecast results in four selected gauges in the watershed with the longest upstream linkages.
