Analyzing the generalization capabilities of a hybrid hydrological model for extrapolation to extreme events

Data-driven techniques have shown the potential
to outperform process-based models in rainfall–runoff simu-
lation. Recently, hybrid models, which combine data-driven
methods with process-based approaches, have been proposed
to leverage the strengths of both methodologies, aiming to
enhance simulation accuracy while maintaining a certain in-
terpretability. Expanding the set of test cases to evaluate hy-
brid models under different conditions, we test their gener-
alization capabilities for extreme hydrological events, com-
paring their performance against long short-term memory
(LSTM) networks and process-based models. Our results in-
dicate that hybrid models show performance similar to that
of the LSTM network for most cases. However, hybrid mod-
els reported slightly lower errors in the most extreme cases
and were able to produce higher peak discharges.