A comparative study of the accuracy of model simulations of accumulated monthly rainfall is made. Two models have been compared: the PSU/NCAR MM5 and the RegCM3 models. For the purpose of the study, we simulated the accumulated rainfall for the month of June 2000 and also for June 2001. These two months provide a good test for the models because the former has lower precipitation than the latter. This is because June 2000 has no occurrence of tropical cyclones while June 2001 have two. The models are integrated for a nested computational domain. The mother domain, which includes the area bounded east-west by 100 to 140 E longitude and north-south by 2 S to 27 N latitude, has grid distance of 90 km. There are two nests within this domain – a coarse nest and a fine nest. The coarse nest has a grid distance of 30 km. The fine nest, which is inside the coarse nest, has a grid distance of 10 km. Since previous studies indicate that the simulated rainfall is highly dependent on the technique for parameterizing subgrid convective rainfall, we integrated the models by using different techniques for cumulus parameterization schemes (CPS). We use the Betts-Miller and Kain-Fritsch scheme for MM5 and we investigate all the available CPS in RegCM3. These are the Anthes-Kuo, Betts-Miller, Grell-Arakawa-Schubert and Grell-Fritsch-Chappell.
Our simulations confirmed the high sensitivity to the parameterization of convective rainfall. Several parameterizations are tested for accuracy in both the MM5 and the RegCM3 simulations. Two kinds of tests are made. First, they are tested in their ability to simulate the large-scale patterns of rainfall. For this purpose, the simulated rainfall distributions for the mother domain are compared with distributions based on satellite observations. Second, the simulations in Domain 3 (fine nest) of the best performing versions of both models in the first test are compared with rain gauge observations. This second test evaluates the accuracy of the models in simulating the small-scale patterns of the rainfall. The results of the first test show that the best performing MM5 version incorporates the Betts-Miller technique. On the other hand, the best performing RegCM3 version incorporates the Grell-Arakawa-Schubert and the Grell-Fritsch-Chappell. Both schemes produce good results in terms of similarity with the large-scale pattern of rainfall rate. These are further compared (second step) using the simulations for Domain 3. The comparison is made by computing the correlation coefficients between the observed and the simulated rainfall at rain gauge stations. The results of the comparison show that the MM5 with the Betts-Miller CPS produces higher correlation than those by RegCM3 model.