Statistical Downscaling of Remote Sensing Precipitation Estimates Using MODIS Cloud Properties Data over Northeastern Greece

Remote Sensing in Earth Systems Sciences https://doi.org/10.1007/s41976-024-00107-1 RESEARCH Statistical Downscaling of Remote Sensing Precipitation Estimates Using MODIS Cloud Properties Data over Northeastern Greece Stavros Stathopoulos1 · Alexandra Gemitzi1 · Konstantinos Kourtidis1 Received: 13 September 2023 / Revised: 29 April 2024 / Accepted: 29 May 2024 © The Author(s) 2024 Abstract The aim of this study is to spatially downscale the daily precipitation data from the Global Precipitation Measurement (GPM) mission, using the Integrated Multi-satellite Retrievals for GPM (IMERG), utilizing cloud properties from the Moderate Resolution Imaging Spectroradiometer (MODIS) instrument. Cloud optical thickness (COT), cloud effective radius (CER), and cloud water path (CWP) are used to statistically downscale IMERG precipitation estimates from 0.1 to 0.01° spatial resolution, using the Multivariate Linear Regression (MLR) and residual correction methods. The downscaled precipitation estimates were subsequently validated using in situ rain gauge measurements. The residual corrected IMERG downscaled precipitation estimates were found to be more accurate than the downscaled predicted precipitation without the implementation of the residual correction algorithm (up to 37%), with a respective decrease of the Root Mean Square Error (RMSE) (up to 75%), Normalized Root Mean Square Error (NRMSE) (up to 79%), and the Percent Bias (PB) (up to 98%). In addition, the final downscaled product after the MLR method implementation with residual correction was better correlated with the rain gauge observations than the initial IMERG product (up to 20%). Thus, the implementation of the MLR method in conjunction with the residual correction algorithm is an efficient tool for downscaling remote sensing products with a coarse spatial resolution. Keywords GPM IMERG · MODIS · Residual correction · Statistical downscaling 1 Introduction Accurate precipitation measurements comprise a valuable resource for the scientific community, allowing their implementation into different hydrological models for various purposes, e.g., prediction of floods and droughts [1–4]. However, precipitation measurements from ground stations do not provide sufficient coverage, with many basins having little or no precipitation measurements, constraining thus the performance of hydrological models and their related applications. To overcome this limitation, many researchers have used interpolated gauge-based precipitation products in their hydrological studies. Lim Kam Sian et al. [5] studied the seasonal precipitation modes over Africa using daily precipitation data (1° × 1° spatial resolution), from the Global * Stavros Stathopoulos 1 Department of Environmental Engineering, Democritus University of Thrace, Xanthi 67100, Greece Precipitation Climatology Centre (GPCC) dataset, while Moazzam et al. [6] investigated the impact of climate change on snow cover over Pakistan, utilizing monthly rainfall and temperature data (0.5° × 0.5° spatial resolution), from the gridded Climatic Research Unit (CRU) dataset. Even though these products are based on gauge data, they have some uncertainty due to the scarcity of the gauge network, the interpolation method that was produced with, and the orographic variability [7–9]. Mallakpour et al. [8], on the other hand, used six different daily gridded precipitation datasets to study the rainfall characteristics over the United States. Two of them were only gauged-based, namely, the Climate Prediction Center (CPC) and the Daily Surface Weather and Climatological Summaries (DAYMET) datasets, one was produced by interpolating gauge and remote sensing data, the Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks-Climate Data Record (PERSIANN-CDR) dataset, and the rest were produced by interpolating gauge, remote sensing, and reanalysis data, namely, the Multi-Source Weighted-Ensemble Precipitation (MSWEP), the Climate Hazards Group Infrared Vol.:(0123456789) Remote Sensing in Earth Systems Sciences Precipitation with Stations (CHIRPS), and the ModernEra Retrospective analysis for Research and Applications (MERRA) datasets. According to Timmermans et al. [9], the uncertainty in these datasets derives from the source data and the assimilation method used, even though they present high spatial and temporal resolution. On the other hand, remote sensing precipitation estimates with their high spatial and temporal coverage provide an alternative, when measurements from gauges are sparse or not available. The Tropical Rainfall Measuring Mission (TRMM) (1997–2014) was the first satellite mission focused on estimating rainfall from space [10]. Its successor, the Global Precipitation Measurement (GPM) mission, using the Integrated Multi-satellite Retrievals for GPM (IMERG), from numerous Passive Microwave (PMW) and Infrared (IR) Imagers/Sounders and precipitation radars and gauges, has drawn the attention of the researchers owing to its improved temporal resolution (approximately every 30 min), latency (minimum latency 4 h), and reliability [11, 12]. The drawback of IMERG GPM precipitation estimates, though, is their coarse spatial resolution (0.1°). For this reason, many researchers have applied various statistical downscaling techniques to improve the spatial resolution and enhance the information content [13, 14]. Statistical downscaling techniques are based on the correlations between a predictand and one or more predictors. Their performance depends on the correlated variables and their statistical relation. Many different variables have been used that present a statistical correlation with precipitation, i.e., Normalized Differential Vegetation Index (NDVI) and Land Surface Temperature (LST), resulting in the spatial downscaling of the initial precipitation estimates [15, 16], with most applications focusing on downscaling monthly or annual precipitation estimates. Regarding downscaling of daily precipitation, cloud properties from the Moderate Resolution Imaging Spectroradiometer (MODIS), onboard the Aqua satellite, were also used as predictors, namely, cloud optical thickness (COT), cloud effective radius (CER), and cloud water path (CWP). COT is a measure of how much light is attenuated as it travels through the atmosphere due to the presence of cloud droplets describing the cloud thickness, while CER denotes the size of the cloud droplets. In addition, CWP is the total amount of water in an air column inside the cloud per unit area. Many previous studies have underlined their inter-connection with precipitation [17, 18]. For example, Zhao et al. [19] found that polluted clouds with high COT can cause an increase in CER, triggering collision/coalescence processes and the initiation of precipitation, affecting CWP. Different algorithms and approaches have been utilized for the statistical downscaling of the initial precipitation estimates, inclu (...truncated)