Multiply Nested Regional Climate Simulation for Southern South America: Sensitivity to Model Resolution

Abstract

Abstract Results are reported from two 5-month-long simulations for southern South America using the fifth-generation Pennsylvania State University–NCAR Mesoscale Model (MM5). The periods of simulation correspond to May–September 1997 and 1998, which were anomalously wet and dry winters for central Chile, respectively. The model setup includes triply nested, two-way-interacting domains centered over the eastern South Pacific and the western coast of southern South America, with horizontal grid intervals of 135, 45, and 15 km. Boundary conditions are provided from NCEP–NCAR reanalyzed fields. The analysis focuses on two subregions of central Chile (30°–41°S). Region 1 (32°–35°S), which is where the observed interannual precipitation differences are largest, is topographically very complex, with a mean height of the Andes Cordillera around 4500 m. Region 2 (35°–39°S) has relatively smooth terrain, as the mean height of the Andes drops to 3000 m. Station precipitation and temperature data are used for model validation. The model exhibits a negative temperature bias (from 2° to 5°C), as well as a positive precipitation bias (40%–80%). This precipitation bias can be partially explained by a positive moisture bias over the ocean in the model. In addition, these biases are highly correlated to the representation of terrain and station elevation in the model. The highest-resolution domain has the smallest precipitation bias for low-elevation stations, but a large positive bias at high altitudes (up to 300%). It also has a better representation of the spatial distribution of the precipitation, especially in region 1, where topography has a larger impact on the precipitation. Overall, the model domain with highest resolution best reproduces the observed precipitation and temperature, as well as the interannual differences. However, this study also shows that large improvements in the simulations of the surface variables are obtained when downscaling from 135 to 45 km, but much smaller improvements are found when downscaling from 45 to 15 km. These simulations represent the first effort in simulating seasonal precipitation in this topographically complex region of the Southern Hemisphere.