US National Oceanic and Atmospheric Administration
Climate Test Bed Joint Seminar Series (2009-2010)
NASA, Goddard Visitor Center, Greenbelt, Maryland, 14 April 2010                                                                            [Print Version]

Drought Monitoring and Prediction:

The Recent Efforts at the Climate Prediction Center

Kingtse Mo

Climate Prediction Center, NOAA/NWS/NCEP

Camp Springs, MD, 20746

The Standardized Precipitation Index (SPI) has been used to classify the meteorological drought. The advantage is that the SPI can be derived from precipitation (P) time series alone without the need of a land-surface model. For the drought monitoring purpose, we update 3, 6, 9, 12 and 24 month SPI daily to reflect the wet and dry situations over the United States from short term (3 months) to longer term duration of 24 month.

We are developing a method to forecast 3-month SPI (SPI3) and 6-month SPI (SPI6) based on the CFS seasonal ensemble forecasts over the United States. Before predicting SPI, the P forecasts from a coarse resolution global model CFS have to be downscaled to a regional grid. In this case, we downscale from the CFS 250 km grid to a fine resolution 50 km regional grid. Four different methods of statistical downscaling and error correction are tested. The four methods are: linear interpolation, a bias correction and spatial downscaling based on the probability distribution functions (BCSD) developed by the University of Washington group, a linear regression method by John Schaake and the Bayesian method used by the Princeton University group. We tested cases with initial conditions in November, February, May and August.

The downscaled CFS P forecasts out to 6 months were appended to the precipitation analyses to form an extended P data set. The SPI was calculated from this extended time series to forecast the meteorological drought. The skill is regionally and seasonally dependent. Figure 1 shows the 6 month SPI over the U.S. is skillful out to 3 months. For the first 3 month lead time, there is no statistical significant difference among different methods of downscaling. After 3 month lead time, the Bayesian methods have small advantages because it takes into account of the hindcast skill and the spread among the CFS forecasts. However, the skill is too low to have any practical use.

 

 

 

 

 

 

Figure 1

The EMC Regional Spectral Model (RSM) group has been performing the dynamic downscaling of the CFS outputs by nesting the RSM in the CFS forecasts. The RSM has the same physics and dynamical core as the CFS and has the resolution of 50 km. While the RSM improves the P climatology, it does not improve interannual variability. For the SPI prediction, the model climatology is replaced by the observed P climatology. Therefore, there is no significant gain in skill in comparison with the Bayesian corrected T62 CFS ensemble P forecasts. To improve rainfall forecasts, the model physics is important. One can not expect to gain significant skill of P forecasts by increasing the model resolution alone.
 

Contact  Kingtse Mo