Performance evaluation statistics applied to derived fields of NWP model forecasts

NWP model skill as obtained from the standard statistics applied to derived atmospheric fields such as thermal advection and moisture convergence is different from that obtained by the same statistics applied to basic model output fields such as temperature or wind components. An analysis with a combination of two simple wave functions shows that the errors in the forecast of the phase of the shortwave component are overwhelmingly more important, For an error of 2 degrees longitude in the phase forecast of the shortwave component (wavenumber similar to 20) the correlation coefficient for the derived fields is only 0.7 whereas it is nearly 0.9 for the basic variable fields. The prediction range of useful forecasts in terms of the derived variables decreases drastically in comparison to that obtained with the simple variables. These aspects are demonstrated with the Centro de Previsao de Tempo e Estudos Climaticos-Center for Ocean-Land-Atmosphere studies operational NWP model in two real synoptic cases that are representative of active weather situations in austral winter over the southern half of South America.