Evaluating precipitation simulations from a land-atmosphere coupling model

Abstract

Nowadays climate models are widely used for weather forecasting. With the increased computational capacity, the simulated resolution of climate models has been enhanced from tens of kilometers in Global Climate Models (GCMs) to less than 10 km in Regional Climate Models (RCMs). The increased resolution has the potential for improved model skill due to capturing more heterogeneity within the system. However, this potential increase in skill is not guaranteed and thus there is a need for further model evaluation studies to provide a better understanding of model configuration and corresponding performance, especially for resolutions in the grey zone (1-10 km grid resolution). Among all variables in climate model simulation, precipitation is one of the most important and also the most difficult variable to predict due to the complexity of land-atmosphere (L-A) interactions. To better understand how well precipitation is simulated and the role of L-A interaction in climate models simulations, three research studies are carried out using the NASA Unified Weather Research and Forecast (NU-WRF) model. First, the impact of model resolution (4 km, 12 km and 24 km) and the use of cumulus parameterization are evaluated in terms of the prediction of precipitation frequency, bias and the spatial patterns. Results shows the cumulus parameterization has more impact than increasing model resolution on precipitation forecast. At the 4 km resolution, the model without cumulus parametrization better captures the spatial patterns in the precipitation forecast. These findings are then used to select the optimal model configuration which is used in the remaining two studies. The second study identifies and evaluates the different drivers of precipitation during the 2018 warm season drought (May-Sep) in the Central Plain. Contributions of different precipitation drivers with and without the impact of Mesoscale Convective Systems (MCSs) and Low Level Jets (LLJs) are analyzed in order to understand the roles of local and non-local precipitation during drought evolution. The lack of MCSs is found to be the dominant driver during the drought, and the LLJ is found to have indirect weaker impact to precipitation with a lag time of up to two days. The last study then quantifies and analyzes the role of local soil moisture feedback during drought through an additional model run with switched initial soil moisture between a relatively wet and dry region to provide insights and understanding of L-A interactions. Results show a 2% change in lifted cloud fraction caused by a soil moisture change less than 0.05 m3 m-3 and LLJ can enhance L-A interactions to promote precipitation formation. Overall, this study provides a comprehensive evaluation of the skill in precipitation forecast of the NU-WRF model and provides insights into the implication of MCS, LLJ, the L-A interaction during local and non-local influenced atmospheric environment and identified model deficiencies. Addressing these deficiencies could improve future precipitation forecasts in the Central Great Plains.