Advancements and Challenges in Gridded Precipitation Datasets: Bias Correction and Downscaling Techniques for Enhanced Modeling in Sicily

Abstract

Spatial-temporal resolution advantages of Satellite Precipitation Products are limited by biases compared to ground observations. Various bias adjustment methods exist, from scaling methods to sophisticated techniques like Quantile Mapping (QM). However, many assume stationarity, leading to inaccuracies in variable climates. Alternative methods, instead, including Equidistant CDF Matching (ECDFM), CDF transferring (CDFt), Quantile Delta Mapping (QDM), aim to address non-stationarity. Also, recent advancements in deep learning offer new methods, such as Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) networks, promising good performances in capturing spatial-temporal dependencies of climate phenomena. Our study proposes a two-stage bias adjustment framework integrating multiple methodologies and advanced deep learning techniques to mitigate systematic bias in Global Precipitation Measurement (GPM) Integrated Multi-satellitE Retrievals for GPM (IMERG) satellite estimatesfor the Sicily, Italy. Also, to tackle one of the most important issues that affects semiarid regions as Sicily, i.e., precipitation datasets with zero-inflation, it exists different methods varying from Singularity Stochastic Removal (SSR) to complex zero-truncated models.