Assessing extreme precipitation from a regional climate model in different spatial–temporal scales: A hydrological perspective in South America

Abstract

AbstractGlobal and regional climate models (GCM and RCM respectively) are the current mathematical tools used to project alterations on precipitation regimes given different greenhouse gases emissions scenarios. However, these models have specific resolutions, physical equations and numerical approaches that provide a diverse set of performances across different regions and spatial–temporal scales. In South America, most hydrological impact studies have used the Eta RCM to yield precipitation projections without a proper uncertainty analysis. It is important to acknowledge its uncertainties prior to any hydrological assessment to adequately support climate change investigations and related water decision making. Therefore, we aim to investigate how Eta extreme precipitation biases vary in different spatial–temporal scales from a hydrological perspective. Thus, we evaluate the extreme precipitation generated by the Eta RCM driven by four different GCMs. It is investigated Eta biases across different temporal (3 hr–5 days) and spatial scales (0.2–1.0°) and how those errors affect river streamflow simulations. It is used local intensity–duration–frequency (IDF) curves and gridded precipitation datasets (MSWEP and ERA5‐Land) as references for Eta assessment. In general, Eta underestimates subdaily extreme precipitation across South America, regardless of the driven GCM. The median bias of a 10‐year return period daily precipitation is −36 mm (1st and 3rd quantiles −58 and −17 mm) compared to MSWEP and −26 mm (1st and 3rd quantiles −45 and −11 mm) compared to ERA5‐Land. However, the relative errors reduce with temporal and spatial aggregation. For example, the average bias of extreme precipitation decreases 8.4 and 5.4 percentage points from 1‐ to 5‐day duration compared to MSWEP and ERA5‐Land, respectively. The negative biases observed for precipitation (20%) are propagated to the flood discharges (40%), and these errors reduce with the drainage area. In general, there are greater biases in extreme discharges for small basins, but these errors considerably reduce for basins larger than 30,000 km2 compared to MGB‐MSWEP simulations. Compared to MGB‐ERA5‐Land simulations, MGB‐Eta presents relatively similar errors for basins of different sizes, probably due to the high negative bias for not only extreme but average precipitation as well.