Evaluating Transformer and LSTM Frameworks for Prediction in Ungauged Basins

2026-06-01

AI Summary

This study addresses the high uncertainty in hydrological prediction in ungauged basins due to scarce observational data. It evaluates the ability of encoder-only Transformers and LSTMs to infer upstream runoff under limited hydrological information, using historical simulations from NOAA’s National Water Model. Comparative experiments are conducted under two configurations: using only upstream data and incorporating both upstream and downstream data. The findings reveal that the inductive bias inherent in recurrent architectures—specifically LSTM’s memory mechanism—is better suited for upstream runoff reconstruction, as LSTMs consistently outperform Transformers. Moreover, when downstream information is introduced as a strong auxiliary constraint, the median normalized Nash–Sutcliffe efficiency (NNSE) across all models improves by over 60%, underscoring the critical role of downstream hydrological data in enhancing predictive performance.

Abstract

Watershed networks exhibit convergent topologies in which multiple tributaries merge into downstream channels,integrating diverse upstream hydrological processes. In ungauged basins, the absence of direct observations increases uncertainty and limits the ability to anticipate extreme events. This study evaluates whether an encoder-only Transformer provides an advantage over an LSTM for upstream streamflow inference under limited hydrologic information, using retrospective simulations from the NOAA National Water Model (NWM). Across both upstream-only and combined configurations, the LSTM showed stronger overall performance than the Transformer model across the two configurations. Incorporating downstream information further boosted performance for all models, increasing median NNSE by more than 60%. Rather than treating this as a leaderboard-style comparison, we interpret the experiments as a test of architectural inductive bias for hydrologic sequence inference. The results indicate that recurrent memory remains better aligned with this upstream reconstruction task than an encoder-only Transformer, while downstream hydrologic context provides a strong auxiliary constraint that substantially improves prediction skill across architectures

Problem

• ungauged basins
• streamflow prediction
• hydrologic inference
• extreme events
• predictive uncertainty

Innovation

• inductive bias
• ungauged basins
• hydrologic sequence inference
• downstream context
• recurrent memory