Currently in the Grell-Freitas (GF) convective parametrization, a uniform average is taken over the results of four closures (stability, omega (Pa/s), moisture convergence, and ECMWF stability). Weighting the closure outputs based on model scale, forecast time and environmental conditions improves forecasts provided by deep convection parameterizations in General Circulation Models (GCMs). We compare high temporal and spatial resolution satellite precipitation data with outputs of the UFS weather model to guide the selection of closure averaging weights to improve model precipitation forecast performance. Our method uses the DBSCAN algorithm to cluster precipitation data into convective cells and associate these cells in modeled and observed datasets, avoiding issues where modeled and observed precipitation cells fail to overlap leading to spurious minimization of precipitation. Least squares regression can then be used to select closure weights that minimize RMSE between observed and modeled datasets. For the tropics, we produced optimized weights that improve model precipitation performance in regions of heavy precipitation. The results were implemented in an FV3GFS run.