Author(s): Niloufar Beikahmadi; Calogero Mattina; Dario Treppiedi; Antonio Francipane; Leonardo V. Noto

Linked Author(s):

Keywords: Bias-Adjustment; Convolutional Deep Network; Distribution-Based Methods; IMERG

Abstract: Over the past decades, an increasing number of gridded precipitation datasets have been developed to overcome the limitations of in situ observations, such as their reduced availability and/or spatial coverage. While these products serve as crucial tools for enhancing the comprehension of global climatology and rainfall patterns, they are concurrently prone to systematic biases and exhibit coarse resolutions. These aspects often limit the possibility of using them directly in hydrological and hydraulic modeling or in forecasting applications. In this context, bias correction and downscaling techniques provide a useful tool to correct all these different sources of data. Focusing on Sicily, which lies in the center of the Mediterranean Sea, we applied different bias correction techniques to improve the accuracy and spatial resolution of the available reanalysis products (i. e., ERA5, MSWEP, and so on). We tested the reliability of different bias-correction and downscaling techniques, including traditional methods such as Delta Mapping (DM) and Quantile Mapping (QM), as well as newer approaches involving machine learning and deep learning-based techniques, which are becoming more and more widespread for precipitation bias correction and downscaling. Due to the availability of rainfall at hourly and sub-hourly temporal resolution, the recently released Convective Permitting Models (i. e., CMCC VHR-REA_IT and SPHERA) covering the Sicilian Island will also be considered. All these products, once corrected with observed data, will be combined into a more realistic high-resolution rainfall product that better represent the complexity of the precipitation phenomena in the region and can be used to force hydrological models.

DOI: https://doi.org/10.3850/iahr-hic2483430201-165

Year: 2024