The global forecast is computed after a multiple linear regression between the ensemble system and a training period of observations. Different statistical weights have been assigned to models in agreement with the results of the regression which has been performed separately for every grid point.
All the models and observations have been regridded at an intermediate horizontal resolution of 0.2 degrees (approx. 21 km). The MAP-SOP period has been divided in a training period of about 60 days used to calibrate the weights for each single model forecast and a verification period of about 15 days to evaluate the performance of the system.
The models used to build the superensemble system are from the MAP-SOP dataset. Precipitation data used for the regression and the verification of the system come from MAP-SOP station data archive and have been combined with SSMI-TRIMM satellite observations.