Which strategy to improve the performances of an LSTM-based model for extreme stream temperature values?
Abstract
Deep-learning models have demonstrated strong performances in reproducing stream temperature dynamics, which is promising for the reconstruction of missing stream temperature records at ungauged locations. However, model accuracy over the range of high, summer stream temperature values has been usually overlooked, raising the question of the suitability of using deep-learning methods during this crucial season. In this study, we investigated strategies to improve the performances of a stream-temperature model based on LSTM (Long Short-Term Memory) cells over the extreme, highest 10 % observed values at 21 stations located in the Garonne river catchment. We quantified the gain in model performance thanks to regional multi-catchment training with static attributes, exploiting hydrologically relevant variables, and further penalizing the errors at extreme temperature values using custom loss functions. Our key results are: (1) Regional multi-catchment training is the best strategy to improve the performances of LSTM models not only over the extreme, top 10 % values but also over the whole range of observations. (2) The gain in performances was mainly brought by the use of static, catchment and reach attributes. (3) Customizing the loss function to emphasize the model errors on extreme temperature values did not lead to significant gains in test performances. This study further confirms the suitability of regionally trained LSTM models that exploit static attributes for the reproduction of extreme stream temperature values, offering significant advantages for water management at data-sparse regions.
Related
- Polar front jet regulates tropical sea surface temperature impacts on South China extreme precipitation
- Agrivoltaic systems can improve agricultural yield by mitigating extreme plant temperatures
- Spatializing daytime street-level thermal exposure frequency profiles in urban areas during extreme heat events using the operative temperature index
- From ‘yuck’ to ‘ahhh’: a focused review on the bacteriophage-based strategy for malodour mitigation to improve the aspirational quotient of toilets
- Emergent constraints improve near-term temperature projections over China: evaluations against observations based on the preindustrial baseline
BibTeX
@article{saadi2026which,
title = {{Which strategy to improve the performances of an LSTM-based model for extreme stream temperature values?}},
author = {Mohamed Saadi and Louis Guichard and Gabrielle Cognot and Laurent Labbouz and Hélène Roux},
journal = {Hydrology and Earth System Sciences},
year = {2026},
month = oct,
doi = {10.5194/hess-30-3623-2026},
url = {https://hal.science/hal-05795711v1},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Which strategy to improve the performances of an LSTM-based model for extreme stream temperature values?
AU - Mohamed Saadi
AU - Louis Guichard
AU - Gabrielle Cognot
AU - Laurent Labbouz
AU - Hélène Roux
PY - 2026
DA - 2026/10/12
JO - Hydrology and Earth System Sciences
DO - 10.5194/hess-30-3623-2026
UR - https://hal.science/hal-05795711v1
AB - Deep-learning models have demonstrated strong performances in reproducing stream temperature dynamics, which is promising for the reconstruction of missing stream temperature records at ungauged locations. However, model accuracy over the range of high, summer stream temperature values has been usually overlooked, raising the question of the suitability of using deep-learning methods during this crucial season. In this study, we investigated strategies to improve the performances of a stream-temperature model based on LSTM (Long Short-Term Memory) cells over the extreme, highest 10 % observed values at 21 stations located in the Garonne river catchment. We quantified the gain in model performance thanks to regional multi-catchment training with static attributes, exploiting hydrologically relevant variables, and further penalizing the errors at extreme temperature values using custom loss functions. Our key results are: (1) Regional multi-catchment training is the best strategy to improve the performances of LSTM models not only over the extreme, top 10 % values but also over the whole range of observations. (2) The gain in performances was mainly brought by the use of static, catchment and reach attributes. (3) Customizing the loss function to emphasize the model errors on extreme temperature values did not lead to significant gains in test performances. This study further confirms the suitability of regionally trained LSTM models that exploit static attributes for the reproduction of extreme stream temperature values, offering significant advantages for water management at data-sparse regions.
SN - 1027-5606
SN - 1607-7938
ER -