Predictive uncertainty estimation in satellite precipitation data correction using machine learning

Predictive uncertainty estimates for precipitation data acquired through merging satellite and ground-based observations are usually not provided. Here, we present the first benchmark experiments on the use of machine learning algorithms for fulfilling the task of delivering such estimates. These experiments compared six machine learning algorithms (i.e., quantile regression, quantile regression forests, generalized random forests, gradient boosting machines, light gradient boosting machines and quantile regression neural networks) and relied on 15-year-long monthly data that originate from across the contiguous United States. The comparison referred to the ability of the machine learning algorithms in delivering predictive quantiles at various levels. The results allow the ordering from the best to the worst of the machine learning algorithms for the problem of interest.

Acknowledgements: The research project was supported by the Hellenic Foundation for Research and Innovation (H.F.R.I.) under the “3rd Call for H.F.R.I. Research Projects to support Post-Doctoral Researchers” (Project Number: 7368).