Sensitivity of US Drought Prediction Skill to Land Initial States on Subseasonal to Seasonal Timescales

The NCEP CFSv2 ensemble reforecasts initialized with two land analyses for the period of 1979-2010 have been conducted to assess the effect of the uncertainty in land initial state on drought prediction over CONUS. The two observation-based land initial states are adapted from the NCEP CFS Reanalysis (CFSR) and the NASA GLDAS-2 analysis and the corresponding reforecasts are referred to as the CFSR and GLDAS reforecasts respectively. Atmosphere, ocean, and ice initial states are identical for both reforecasts, which are from the instantaneous restart files of the ECMWF Ocean Reanalysis System 4 with a five-member ensemble and the atmosphere and sea ice restart files of CFSR. Ensemble reforecasts of 12-month duration are initialized at the beginning of January, April, July, and October. For each initial month, a 20-member ensemble per land analysis is constructed by pairing each of the ocean initial states with four atmosphere/ice/land initial states at 00Z of the first four days, achieving a total of 40 ensemble members if the two reforecasts are combined. Most of the reforecasts have been completed except that the July-initialized runs are now in progress.

We have examined the anomaly correlation skill of monthly mean precipitation over CONUS in the CFSR and GLDAS reforecasts initialized in April and October. The spatial patterns of correlation skill for the predicted precipitation are similar to each other for October-November (April-May) at 0 lead month and December-February (June-August) at 2 lead month. Quantitatively, the differences of skill score pass the statistical significance in some areas but the spatial distribution varies with lead-time and/or target season. In particular, the skill difference in the April-initialized reforecasts is intensified over broader regions during summer, possibly because the influence of ocean memory on precipitation forecast is weakest then and the land-atmosphere interaction becomes more significant. It is also interesting that the GLDAS prediction skill seems to be always better than that of the CFSR reforecasts near the northwestern coast of US. With the two sets of reforecasts, further evaluation of the prediction skill of soil moisture, surface temperature and selected drought indices from days to seasons will be discussed with respect to both ensemble mean and spread.