Regresar

Ensemble-Tailored Pattern Analysis of High-Resolution Dynamically Downscaled Precipitation Fields: Example for Climate Sensitive Regions of South America

Abstract:

For climate adaptation and risk mitigation, decision makers in water management or agriculture increasingly demand for regionalized weather and climate information. To provide these, regional atmospheric models, such as the Weather Research and Forecasting (WRF) model, need to be optimized in their physical setup to the region of interest. The objective of this study is to evaluate four cumulus physics (CU), two microphysics (MP), two planetary boundary layer physics (PBL), and two radiation physics (RA) schemes in WRF according to their performance in dynamically downscaling the precipitation over two typical South American regions: one orographically complex area in Ecuador/Peru (horizontal resolution up to 9 and 3 km), and one area of rolling hills in Northeast Brazil (up to 9 km). For this, an extensive ensemble of 32 simulations over two continuous years was conducted. Including the reference uncertainty of three high-resolution global datasets (CHIRPS, MSWEP, ERA5-Land), we show that different parameterization setups can produce up to four times the monthly reference precipitation. This underscores the urgent need to conduct parameterization sensitivity studies before weather forecasts or input for impact modeling can be produced. Contrarily to usual studies, we focus on distributional, temporal and spatial precipitation patterns and evaluate these in an ensemble-tailored approach. These ensemble characteristics such as ensemble Structure-, Amplitude-, and Location-error, allow us to generalize the impacts of combining one parameterization scheme with others. We find that varying the CU and RA schemes stronger affects the WRF performance than varying the MP or PBL schemes. This effect is even present in the convection-resolving 3-km-domain over Ecuador/Peru where CU schemes are only used in the parent domain of the one-way nesting approach. The G3D CU physics ensemble best represents the CHIRPS probability distribution in the 9-km-domains. However, spatial and temporal patterns of CHIRPS are best captured by Tiedtke or BMJ CU schemes. Ecuadorian station data in the 3-km-domain is best simulated by the ensemble whose parent domains use the KF CU scheme. Accounting for all evaluation metrics, no general-purpose setup could be identified, but suited parameterizations can be narrowed down according to final application needs.