Fusion of gridded satellite and earth-observed daily precipitation data in the United States using tree-based ensemble learning algorithms

Fusing gridded satellite products and ground-based measurements is regularly made for producing precipitation datasets that: (a) cover large geographical regions with high density; and (b) are more accurate than pure gridded satellite products. Therefore, the respective procedures are commonly referred to as “correction” of satellite products. The same procedures often rely on the application of machine and statistical learning regression algorithms in spatial settings. Regression problems can be solved with high accuracy and low computational cost by applying tree-based ensemble learning algorithms. Despite this, information on which tree-based ensemble learning algorithm to select for the contiguous United States (US) and at the daily time scale was until recently missing from the literature of satellite precipitation product correction. Here, we present the first comparison of tree-based ensemble learning algorithms in this particular context. The algorithms compared are random forests, gradient boosting machines (gbm) and extreme gradient boosting (XGBoost). For the comparison, we extracted and utilized information from the PERSIANN (Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks) and the IMERG (Integrated Multi-satellitE Retrievals for GPM) gridded datasets. We also extracted and utilized earth-observed precipitation data from the Global Historical Climatology Network daily (GHCNd) database. XGBoost was found to perform better than random forests and gbm, and IMERG was found to be more useful than PERSIANN in the context investigated.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).
The 28th IUGG General Assembly (IUGG2023) (Berlin 2023)