LSTM outperforms Transformer for streamflow prediction in ungauged basins

A new study published on arXiv evaluates the effectiveness of Transformer and LSTM models for predicting streamflow in ungauged river basins. Researchers found that the LSTM architecture generally outperformed the Transformer model in reconstructing upstream flow, particularly when incorporating downstream hydrologic context. The findings suggest that recurrent memory mechanisms are better suited for this specific hydrological sequence inference task compared to encoder-only Transformers, and that including downstream data significantly enhances prediction accuracy for both model types. AI

IMPACT Suggests recurrent architectures may be better suited for hydrological sequence modeling than encoder-only Transformers.

RANK_REASON The cluster contains a research paper evaluating machine learning models for a specific scientific application. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• arXiv
• LSTM
• National Water Model
• Transformer

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →