For a period of one year, both 0000 and 1200 UTC runs of the operational Eta model have been archived at the National Severe Storms Laboratory. The Eta analysis is used to verify forecasts from previous runs. Hence, forecast errors can be computed for any of the archived fields (e.g., 24 h forecasts of 500 hPa geopotential). Using the collection of forecast error fields, the spatial distribution of 95% confidence intervals are constructed to show whether or not the mean bias errors are statistically significant. Mean bias errors are defined for the four seasons and also for the year. Some notable differences between the spatial bias error structure of the 0000 UTC forecast cycle and the 1200 UTC forecast cycle are observed.

To determine if similar analysis patterns yield similar error patterns, we employ cluster analysis. Using clustering techniques, we define independent forecast error patterns (e.g., 24 h 500 hPa forecast) and initial analysis patterns (e.g., verifying 500 hPa analysis). If most members in a forecast error pattern cluster have corresponding counterparts in the verifying initial analysis pattern cluster, then similar errors result from similar initial conditions. However, if the clusters are disjoint, then little can be said about how errors are likely to evolve given the initial conditions. We will show results of such an analysis.