Pre-processing satellite rainfall products improves hydrological simulations with machine learning

ABSTRACT

A new pre-processing methodology for gridded Satellite Precipitation Products (SPPs) is developed to improve the performance of Machine Learning (ML) algorithms for runoff prediction. The developed approach was applied to capture the rainfall patterns, and to select relevant input data. This approach was tested using the FeedForward Neural Network (FFNN) and the Extreme Learning Machine (ELM) given their flexibility and ability in hydrological modelling. The methodology was tested in a semiarid transboundary watershed located in North Africa (Algeria, Tunisia) with the Integrated MultisatellitE Retrievals for Global Precipitation Measurement (GPMIMERG) and the Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks (PERSIANN) products. The results demonstrate the effectiveness of the proposed approach using all employed SPPs. In terms of Nash-Sutcliffe efficiency, the suggested pre-processing technique improved the prediction ability of FFNN by 13%, and of ELM by 15%, which highlights how pre-processing techniques significantly enhance ML models with SPP data.

KEYWORDS:

• machine learning
• IMERG
• PERSIANN
• pre-processing
• rainfall
• runoff
• satellite precipitation products

Editor K. Soulis Associate Editor M. Ionita

Editor K. Soulis Associate Editor M. Ionita

Acknowledgements

The authors acknowledge the ANRH of Algeria and the DGRE of Tunisia for providing the important database of monthly rainfall for this research. This work was developed with the support of the Directorate-General for Scientific Research and Technological Development DGRSTD (Algeria).

Disclosure statement

No potential conflict of interest was reported by the authors.

Authors’ contributions

TB, MG and KF designed and developed the models and methods. Data collection and analysis were performed by TB and HA. HB guided and supervised the whole process. TB, MG and RH drafted the manuscript; MS, SAK and YT revised the manuscript; and all authors read and approved the final manuscript.

Code availability

A pseudo-code of the proposed pre-processing methodology is presented in the Supplementary material.

Data availability statement

Daily rainfall and runoff data are obtainable from the ANRH (Algeria) and the DGRE (Tunisia) on request.

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/02626667.2024.2378108

Additional information

Funding

This work was supported by the PRFU-MESRS Project (Code# A17N01UN230120220001) named “Flood forecasting using remote sensing and using machine learning techniques.”