Summary
This study develops Long Short-Term Memory (LSTM) neural network models for predicting streamflow in 500+ US catchments using globally available ERA5 meteorological forcing and catchment characteristics, rather than local datasets. While local data-trained models achieved higher overall performance (median daily NSE 0.71 vs. 0.54), the global approach performed comparably or better in Western and North-Western US catchments (median NSE 0.83 vs. 0.78), suggesting potential for scaling streamflow predictions to ungauged basins worldwide using publicly available datasets.
UK applicability
The methodology could be adapted for UK hydrology and water resource management, though performance would depend on ERA5 data quality over the UK and Ireland. The approach is particularly relevant for supporting predictions in poorly gauged UK catchments and informing water security and flood risk assessment under climate variability.
Key measures
Daily Nash-Sutcliffe Efficiency (NSE) for streamflow predictions; spatial variation in model performance across Western, North-Western, and other US regions
Outcomes reported
The study developed and evaluated LSTM models for streamflow prediction across over 500 catchments using global ERA5 meteorological data and compared performance against models trained with local, higher-resolution datasets. Performance metrics (Nash-Sutcliffe Efficiency) were reported for catchments across different regions of the United States.
Topic tags
Dig deeper with Pulse AI.
Pulse AI has read the whole catalogue. Ask about this record, its theme, or how the findings apply to UK farming and policy — every answer cites the underlying studies.