E1.7 E3SM Evaluation using GPM/NEXRAD

Abstract

Simulation of cloud and precipitation has been challenging in global models. The newly-developed earth system model by DOE, i.e., the Energy Exascale Earth System Model (E3SM), targets at addressing challenging in simulating water cycle. In this study, we provide a comprehensive evaluation of cloud and precipitation properties in warm seasons over the continental United States simulated by E3SM, using both NEXRAD WSR-88D observations (good accuracy at low and middle altitudes) and Ku-band radar onboard the Global Precipitation Measurement (GPM) core observatory (good accuracy at high altitudes). The combination of both datasets reveals vertical and horizontal storm structure. Comparison of WSR-88D and the Ku-band radar GPM overpasses shows agree well in overall magnitude, spatial distribution and vertical structure of radar echo over the U.S. With the two 3D radar reflectivity fields serving as the ground-truth, the E3SM Ku band radar reflectivity simulation using CFMIP Observation Simulator Package (COSP) is evaluated for the period of April-September 2014-2016. By coarsening the fine resolution observations (5 km for GPM and 2 km for NEXRAD) to the E3SM grid (1^o), good match is found between E3SM and GPM/NEXRAD observations in both magnitude and spatial distribution below 4 km altitude, but there is severe model underestimation in the upper levels. The simulated echo top height is about 2 km lower compared than the maximum echo heights shown by GPM measurements. To determine what physical processes could contribute to the significantly underestimated upper-level cloud and echo top height, we conducted sensitivity tests such as changing the Bergeron process efficiency, which is set too low in E3SM, and evaporation efficiency.  In addition, we also examined the sub-grid variability estimated by E3SM within a 1^o grid box and compared to the observations of the actual subgrid variation of radar reflectivity. The bimodal distribution simulated by E3SM differs from lognormal distribution observed, indicating the bi-Gaussian distribution in the sub-column generator and CLUBB parameterization might not be appropriate.