Research Highlights

<p>Hybrid machine learning models transform rainfall-runoff predictions in Ethiopia</p><p></p>

Published online 22 August 2023

Neuro-fuzzy ensemble and LSTM-BRT models show promising results in rainfall-runoff modelling

DonCamillo/ CC BY-SA 4.0 Enlarge image
The performances of four machine-learning models in accurately modelling the rainfall-runoff process, which is crucial for water-resource management and disaster prevention, were compared.

Of the four machine-learning models, the long short-term memory model best modelled the rainfall-runoff process.

Additionally, four ensemble techniques were examined to improve the accuracy of single models. Of these, the neuro-fuzzy ensemble performed the best in both calibration and validation phases. Furthermore, hybrid models combining boosted regression tree (BRT) with machine-learning models showed promising results in improving prediction accuracy.

This study highlights the potential of using ensemble techniques and hybrid BRT models for predicting rainfall-runoff processes. Future research should explore other deep-learning models and optimization algorithms to further improve modelling accuracy.