The use of regional models to simulate physical processes at higher-resolution scales has been the main theme of my research. Through a technique known as dynamic downscaling, results from general circulation model (GCM) integrations are downscaled with the help of a regional climate model (RCM) to resolutions beyond to those currently available in global models. The same technique can also be used to downscale reanalysis products, as well as coarse-resolution regional-scale simulations.
- Regional climate modeling
My modeling studies for North and South America have revealed that dynamic downscaling with RCM can add significant value to GCM seasonal precipitation hindcasts by simulating locally forced atmospheric process, which cannot be resolved by GCM’s coarser horizontal and vertical grids, as well as by providing a more realistic representation of land surface characteristics such as topography and land cover type distribution (De Sales and Xue, 2006, 2011, 2012). In these articles, the added value associated with dynamic downscaling, at seasonal and inter-annual scales, was assessed and verified through robust statistical methods, including discrete wavelet transform and precipitation energy decomposition analysis.
Wavelet decomposition of simulated precipitation skill scores over subtropical South America showed that dynamic downscaling with regional model is an effective method to improve the simulation of precipitation in the region, when compared to GCM results alone. Such improvement is not equally distributed throughout the precipitation spectrum; instead, it is concentrated at more intense and smaller scale events (Figure 1). In addition, wavelet decomposition showed that RCM downscaling can improve the simulation of inter-annual precipitation differences (De Sales and Xue 2011).
Figure 1: Decomposition of precipitation skill score over subtropical South America from a) GCM and b) RCM simulations into different spatial scales and thresholds. Darker shades indicate lower skill scores. De Sales and Xue, 2011.
I have also worked on dynamic downscaling of winter climate simulations over North America as part of the Multi-Regional climate model Ensemble Downscaling (MRED) project, which investigated, among other issues, whether a RCM can provide additional useful information for hydrologic applications of climate forecasts in the United States (De Sales and Xue, 2012). Figure 2 shows some of my results for the MRED project obtained with the ETA/SSiB-3 regional modeling system.
Figure 2: Observed and simulated winter monthly precipitation (mm/day) in the western USA from 1982 to 2004. Ribbon indicates 15-member ensemble spreads. Dynamic downscaling with regional model reduced significantly the excessive precipitation bias seen in the global model results. (De Sales and Xue, 2012).
Dynamic downscaling improved the simulation of monthly precipitation over the American West, when compared to GCM results. Large precipitation biases in the GCM results were reduced significantly through the dynamic downscaling (Figures 2). My research has revealed that the improvement was linked to enhanced topographic representation and correct simulation of surface energy partitioning by the RCM.
Another example of the dynamic downscaling’s potential is revealed in Figure 3, which shows the distribution of winter precipitation by intensity over the western United States calculated from observations, as well as from global and regional model simulations. While the GCM overestimates the frequency of precipitation events at nearly all thresholds (red line); the dynamic downscaling results with a RCM provides a better representation of precipitation distribution, especially for weaker to mid-intensity precipitation events (blue line). Analogous studies focused on the South American continent also show similar improvements (De Sales and Xue, 2011). These results corroborate the idea that high-resolution RCM simulations can add value to GCM simulations.
Figure 3: Distribution of precipitation by intensity over the western USA calculated from observation and model simulations (mm/day). GCM overestimates significantly the frequency of precipitation events (red line), whereas the RCM downscaling generates a distribution more comperable to the observetions (black line). De Sales and Xue, 2012.
The link between land surface and atmosphere through vegetation biophysical processes is an important driver of weather and climate. Among the effects of vegetation biophysical processes on climate, my attention has mainly centered on understanding how these processes impact the spatial distribution and the temporal evolution of precipitation. In my doctorate dissertation, I examined the role of vegetation biophysical processes on the South American summer climate through a series of seasonal climate simulations.
In 2010, we published an article showing that the explicit consideration of vegetation biophysical processes in GCM can significantly improve the simulation of surface energy and water balances, which in turn can generate more realistic precipitation simulations (Figure 4).Figure 4: Bias reduction for different regions due to the implementation of the vegetation biophysical processes in seasonal precipitation simulations in mm/day (Xue and De Sales, et al. 2010).
Currently, I am working on investigating the impact of burned areas on Africa’s seasonal climate. Every year extensive areas of the African continent vegetated land experience natural and anthropogenic fires, which are important disturbance agents to the regional climate. Burned area information for the study was aggregated from MODIS 500-meter monthly burned area product, which combines MODIS-TERRA and MODIS-AQUA land surface reflectance data. This information is being passed onto regional model for a series of year-long experiments.
Figure 5: Annual burned area (as percentage of the area of the grid cell), averaged over 2000-2010. Data is based on MODIS burned area product using the direct broadcast algorithm, aggregated from the native 500m resolution to 0.5° regional model grid.
Preliminary results show that the model is sensitive to the burned areas and that the land cover and soil degradation associated with burned areas can affect precipitation, in particular the West African monsoon system development.