| ABSTRACT | High resolution gridded daily precipitation datasets are valuable for agriculture, hydrology, ecology, climate research, etc. Observational datasets, however, do not necessarily reflect real picture of precipitation change over particular area. For that purpose, gridded datasets are more preferable. In order to predict precipitation at locations with no measurements, spatial interpolation methods are mostly used. Previous studies report lack of accurate models for producing gridded daily precipitation data. High resolution gridded precipitation datasets have not been available for Mediterranean basin so far. In this study we use random forest model for gridded daily precipitation to produce gridded dataset for 2017 in the Mediterranean area at 1 km spatial resolution. The model uses EUMETSAT images as covariates, allowing possibility to extend dataset over longer period of time, and not only for one year. Distances in space and time are also used as covariates in order to examine, not only spatial, but spatio-temporal dependence of precipitation. Observational precipitation datasets (GHCN, GSOD, ECA&A) are used for development and testing of the model. Training and test data are selected in such a way that both present a representative sample in space-time domain and respective to altitude. The model we propose might be used for production of high resolution daily precipitation grids in Mediterranean basin for longer period of time. |