Correcting for the Seasonal Cycle of Bias in the Global Ensemble Forecast System (GEFS) Using the GEFS reforecast v2

Statistical post processing of model forecasts has long been viewed as one of the most cost effective ways to improve model forecasts. Given a large enough sample size, techniques such as linear regression, non-homogeneous Gaussian regression, model output statistics, etc. can be applied to forecasts to reduce the overall model error. The largest issue one encounters when working with statistical post processing methods is that of sample size. A large sample size (30 years) is necessary to successfully sample the model climatology. Thus, there lies a strong need for operational forecast centers to provide reforecast datasets (historical forecasts where the current state of the model is fixed in time) derived from the current state of the model in order to develop statistical post processing techniques that can be used in real time.

Recently, the National Centers for Environmental Prediction (NCEP) released a reforecast dataset of the Global Ensemble Forecast System (GEFS), which issued historical forecasts from 01 December 1984 to the present, consisting of 10 ensemble members plus the control run. Here, we use this dataset to investigate the seasonal cycle of 2-meter temperature forecast bias globally. Once calculated, the seasonal cycle of bias can be subtracted from real time forecasts to provide a stable bias corrected forecast.

This presentation will focus on the seasonal cycle of 2-meter temperature bias in the GEFS reforecast v2, and demonstrate how correcting for the known model bias can improve the overall model 2-meter temperature forecast.