Model Climatological Analysis of Precipitation from NCEP GEFS Reforecast

ESRL/PSD GEFS Reforecast V2 is an extensive dataset of historical weather forecasts generated with NCEP's 2012 operational Global Ensemble Forecasting System (GEFS) version for the past 28 years. It is developed mainly for the purpose of a number of applications, including statistical post-processing, diagnosis of the forecast ability of uncommon phenomena and initialization of regional model reforecasts. In particular, this long reforecast dataset gives us an unprecedented opportunity to develop a dataset of model precipitation climatology for GEFS, which could be used for forecast calibration and verification studies and anomaly forecast guidance. Due to non-Gaussian feature and heavy-tailed distributions of precipitation amount, the L-moment method with Gama distribution as a fitting function was employed to derive the NCEP Climatology-Calibrated Precipitation Analysis (CCPA) daily climatology in our previous study, and also is utilized here to calculate model daily climatology based on 26 years (1985-2010) of the GEFS Reforecast as well.

Furthermore, it is necessary to evaluate model climatology from the Reforecast dataset. In this study, CCPA climatology for a period of 8-years (2002-2010) is used to evaluate Reforecast climatology with the coincident period for the Contiguous United States (CONUS) domain and each River Forecast Center (RFC). These datasets are compared at 1-deg grid resolution and various time scales ranging from daily, monthly to seasonal time scales. Detailed comparisons are provided to decide a selection of sampling methods in the calculations of daily climatology, to assess the quality of Reforecast and to understand the error characteristics associated with the Reforecast precipitation.

Preliminary results show good overall agreement between the CCPA and the Reforecast over CONUS in the shortest lead time. However, agreement has a wide variety from one RFC to another and from short to long forecast lead time, as we find that the strength of the correlation and model bias varies significantly from one RFC to another. Our preliminary conclusions suggest that the GEFS Reforecast could provide a useful precipitation climatology dataset for our future applications, such as forecast calibration and verification, and anomaly forecast generation.