Monday, 8 January 2018
Exhibit Hall 3 (ACC) (Austin, Texas)
Climate prediction provides weather information (e.g. wind, precipitation, temperature) as well as warning of extreme events (such as flood and drought) to minimize the damage. Global Climate Models (GCMs) provide general climate features at global scale, but the coarse resolution does not provide precise details needed for local application. Dynamic downscaling is one solution for providing climate features at finer spatial resolution. It utilizes Regional Climate Models(RCMs) forced by coarse GCMs for better predictions at local scales. There has been much research to evaluate regional hydro-meteorological features using this method and the results have informed model constraints, setup and limitations for using dynamic downscaling. Yet, there is still much debate regarding the trade-off between accuracy and computation cost. The intense computation effort impedes the wide application of using RCMs, for downscaling models to high resolution (<10km) with multiple realizations. There have been many studies diagnosing the prediction accuracy and computational tradeoff from dynamic downscaling, however, the results are typically evaluated at a single spatial or temporal resolutions and do not explicitly quantify the increase in information from downscaling. Downscaling not only provides information at the fine resolution, which is usually a representative of model skill, but also may have increased skill at coarser resolutions due to temporal/spatial averaging that reduces model errors. On the contrary, running model at fine resolution is unnecessary if there’s little difference from interpolating result from a coarse model. These two questions are critical to indicate if temporal/spatial averaging or spatial interpolation provides a means to increase the skill/efficiency of the predictions and justify computational cost. To test this hypothesis, we use downscaled runs from a recent NASA sponsored downscaling project that utilized NASA Unified Weather Research and Forecast (NU-WRF) model constrained by MERRA-2 reanalysis over the contiguous US (CONUS). This work will evaluate the downscaled precipitation from the NU-WRF model at a spatial resolution of 4km, 12km, 24km against the Livneh gridded observational dataset. The results will be presented using Taylor diagrams and derived skill measures based on the analysis at different spatio-temporal resolutions. This analysis will provide new insights into the utility of dynamic downscaling.
- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner