Identifying the Optimal WRF-ARW Configuration for the Amazon Rainforest

Convection, which depends on the complex interaction between surface and atmosphere, can develop across a wide range of spatial and temporal scales. This complexity partly explains why large-scale models, especially in the tropics, often fail to accurately represent the intensity, timing, and location of convective precipitation compared to observations. One way to better understand the complex mechanisms responsible for developing and organizing convection is to use atmospheric models at cloud-resolving spatial resolutions. For this study, we used the Advanced Research Weather Research and Forecasting (WRF-ARW) model with resolutions of 9 km, 3 km, and 1 km to simulate the diurnal cycle of convection in the Amazon rainforests.The WRF-ARW model offers a wide range of physics and dynamic options. Thus, finding the optimal configuration to reproduce the observations closely enough is crucial. This step is essential for the subsequent use of the model to study of deep convection.We conducted a one-week-long simulation in December 2014, using either the European Center for Medium-Range Weather Forecasts (ECMWF) reanalysis (ERA-5 and ERA5-land) or the National Centers for Environmental Prediction (NCEP) Final Operational Global Analysis (FNL) dataset for the model initialization and boundary conditions. In addition to the reanalysis, we tested the Global Land Data Assimilation System (GLDAS) for soil moisture and temperature. We tested various configurations for microphysics (WSM6, WDM6, Thompson, and Morrison double moment), boundary layer (YSU, MYNN, MYJ), surface layer (MM5, MYNN, MYJ), and land surface (Unified Noah, Noah-MP).The model results were evaluated using the Taylor Skill Score (TSS) and diagram for key variables, including surface heat fluxes and precipitation rates that are integral to deep convection, with observations from the GoAmazon 2014/15 campaign. Among the configurations tested, the combination of ERA5 with ERA5-land, WDM6, YSU, MM5, and Unified Noah showed the best performance, using third-order Runge-Kutta time integration and monotic advection. In addition, enhancing the spatial resolution of the model leads to improved results. For instance, the abovementioned optimal configuration gives a TSS of 0.633 for surface heat flux and 0.632 for precipitation rate in the outer domain (d01), with corresponding values of 0.782 and 0.684 in the innermost (d03). Identifying this optimal configuration provides a valuable basis for further sensitivity experiments to understand the mechanisms contributing to deep convection in this region.