Assessing the Impact of Different Satellite Retrieval Methods on Forecast Available Potential Energy

Abstract

The isentropic form for available potential energy (APE) is used to analyze the impact of the inclusion of satellite temperature retrieval data on forecasts made with the NASA Goddard Laboratory for Atmospheres (GLA) fourth order model. Two analysis datasets are used for the forecasts, one containing the NEDIS TIROSN retrievals and the other GLA retrievals using the physical inversion method. A third analysis dataset did not contain satellite data and was used as a control. Two analysis datasets, with and without satellite data, were used for verification. Northern Hemisphere values for the total APE show an increase throughout the 72 h forecast period for all three sets, mostly due to an increase in the zonal component, in contrast to the verification sets, which showed a steady level of total APE. The three forecast sets start with different values of total APE but by 36 h, the differences begin to diminish. At 72 h the total APE values for the three sets are almost identical. The magnitude of the total APE in the Southern Hemisphere does not increase in time and remains within the range of the verification sets. The vertically integrated grid point distributions of the eddy APE which provide geographical representations of baroclinic zones show little difference between the 0 h forecast fields and the analysis fields. At the end of the forecast period, however, there are moderate differences in the southern Pacific and Indian oceans. The grid point distributions of eddy APE are quite different among the three forecasts as well as the verification sets. In the Northern Hemisphere there are very pronounced differences in grid point distribution over Asia, the North Pacific Ocean and North America. These differences increase in area and magnitude as the forecast period progresses. They show coherence in their eastward progression with time across the eastern Pacific and North America. Examination of the grid point distributions indicates that the forecasts are slow, for example, in developing a low in the northern Rockies and never correctly capture its position or intensity. The differences noted above are a reflection of incorrect timing, position and/or magnitude. Isentropic cross sections for 50°N show that the lime lag exists around the entire parallel. The forecast set using the GLAS rerievals does somewhat better than the other sets in predicting intensity and position of developing lows, but all three forecasts resemble each other rather than the verification sets. This would seem to indicate that model characteristics are overwhelming any differences in the data sets.