Investigating the performance of precipitation products and outputs of cmip6 gcms in reproducing precipitation extremes observed across Uganda

Abstract

Precipitation products are increasingly being utilized as substitutes for observed precipitation data. This study evaluated the performance of five well-known precipitation products including CFSR, CHIRPS, JRA-55, ERA-5, and PGF and outputs from the CMIP6 General Circulation Models (GCMs) in replicating observed precipitation extremes across Uganda. Annual precipitation extreme events observed at 20 meteorological stations were compared with those from the precipitation products and GCMs outputs over the period 1979-2022. Utilized 6 precipitation extreme indices i.e. total precipitation (TPREC1), annual maxima series (AMS), number of dry days (NDD1), maximum dry spell (MWS1), number of wet days (NWD1), and maximum wet spell (MWS1) using a threshold of 1mm/day. To compare observed data with precipitation products, frequency analysis and bias ratios were applied. Correlation analysis and Taylor diagrams were utilized for comparing observed data with GCMs outputs. The quantile mapping bias correction method, implemented using the qmap package in R Studio, was applied to downscale GCM outputs with both observed data and precipitation products. This approach facilitated the determination of future climate change projections for the SSP2- 8.5 and SSP5-8.5 scenarios for the 2050s and 2080s. PGF systematically overestimated observed AMS at all the selected stations. Contrastingly, JRA-55, ERA-5, and CHIRPS datasets consistently underestimated observed AMS quantiles across all stations and return periods. The regression line for AMS when plotted against log-transformed return period exhibited a steeper slope for CFSR than that of the observed data. The null hypothesis (H0) of no correlation between observed and GCMs-based historical AMS, was rejected in varying proportions of the 20 meteorological stations. For instance, the H0 was rejected (p < 0.05) in 10% of the 20 precipitation stations across Uganda when observed AMS was compared with that of NOR_ESM2. For each of the three GCMs including CESSM_WACCM, INM_CM5, MRI_ESM2, the H0 was rejected (p < 0.05) in 5% of the 20 stations considered. The extent to which GCMs capture precipitation extremes varies due to their coarse spatial resolution. Bias correction of GCM outputs using quantile mapping method yielded different climate change signals in the cases when observations and precipitation products were considered. For instance, climate change impact on the total of precipitation above 1 mm/day intensity were over the ranges 40-170%, 20-61%, 15-41%, 5-36%, 8-67%, and 20-55% when bias correction of GCMs outputs were done using CFSR, PGF, CHIRPS, ERA-5, JRA-55, and observed precipitation, respectively. The choice of precipitation product significantly impacts climate projections. Thus, there is a need for careful selection of which precipitation product should be used for bias correction of GCMs’ outputs in absence of observed precipitation when analysis is being done for planning water resource management and climate change impact assessments.