Learning from dense meteorological networks: a regional LSTM model for streamflow simulation

Long Short-Term Memory (LSTM) networks have become the current state-of-the-art for streamflow modelling due to their ability to capture non-linear relationships and long-term temporal dependencies. In regional LSTM applications, meteorological forcings are typically aggregated at the catchment scale, which removes spatial variability and may limit the representation of key hydrological processes, particularly in large and heterogeneous basins. This study departs from this conventional practice by proposing a regional LSTM model that directly exploits the full spatial structure of raw meteorological observations without any prior spatial aggregation. The model integrates dynamic and static information from more than 700 meteorological stations across the northern Ebro River Basin (Spain) to simulate streamflow at 31 catchments. Missing data are handled using a masked mean approach, enabling direct use of observations. Compared to a benchmark regional LSTM based on catchment-averaged forcings, the proposed approach improves overall performance, and yields substantial reductions in error for medium- and high-flow conditions. These results indicate that preserving spatial meteorological variability enhances the representation of hydrological processes associated with runoff generation and peak-flow response. To further investigate the learned relationships, explainable artificial intelligence (XAI) techniques are applied. The analyses reveal physically consistent spatial and temporal attribution patterns, showing that the model dynamically adjusts the relevance of meteorological stations depending on hydrological conditions. This study demonstrates that spatial aggregation of meteorological inputs can limit their ability to capture hydrological dynamics in complex basins. The proposed framework provides a flexible and data-driven approach for exploiting dense observational networks in regional streamflow modelling.

Authors

Institutions

Publication Details

Published
2026-11-18
DOI
https://doi.org/10.5194/egusphere-2026-5583
Primary Topic
Hydrological Forecasting Using AI
Type
preprint

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Learning from dense meteorological networks: a regional LSTM model for streamflow simulation

Daniel Klotz, Carmelo Juez, Inmaculada González-Planet
1 links
Hydrological Forecasting Using AI
preprint

Learning from dense meteorological networks: a regional LSTM model for streamflow simulation

Daniel Klotz, Carmelo Juez, Inmaculada González-Planet
preprint en

Abstract

Long Short-Term Memory (LSTM) networks have become the current state-of-the-art for streamflow modelling due to their ability to capture non-linear relationships and long-term temporal dependencies. In regional LSTM applications, meteorological forcings are typically aggregated at the catchment scale, which removes spatial variability and may limit the representation of key hydrological processes, particularly in large and heterogeneous basins. This study departs from this conventional practice by proposing a regional LSTM model that directly exploits the full spatial structure of raw meteorological observations without any prior spatial aggregation. The model integrates dynamic and static information from more than 700 meteorological stations across the northern Ebro River Basin (Spain) to simulate streamflow at 31 catchments. Missing data are handled using a masked mean approach, enabling direct use of observations. Compared to a benchmark regional LSTM based on catchment-averaged forcings, the proposed approach improves overall performance, and yields substantial reductions in error for medium- and high-flow conditions. These results indicate that preserving spatial meteorological variability enhances the representation of hydrological processes associated with runoff generation and peak-flow response. To further investigate the learned relationships, explainable artificial intelligence (XAI) techniques are applied. The analyses reveal physically consistent spatial and temporal attribution patterns, showing that the model dynamically adjusts the relevance of meteorological stations depending on hydrological conditions. This study demonstrates that spatial aggregation of meteorological inputs can limit their ability to capture hydrological dynamics in complex basins. The proposed framework provides a flexible and data-driven approach for exploiting dense observational networks in regional streamflow modelling.

Consejo Superior de Investigaciones Científicas (ES), Instituto Pirenaico de Ecología (ES), Interdisciplinary Transformation University Austria (AT)
HORIZON EUROPE Framework Programme, HORIZON EUROPE European Research Council
Climate action, Clean water and sanitation
Hydrological Forecasting Using AI
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon