Using an atmospheric observation to initialize a numerical weather forecast requires that the statistics of the observation error be known. Among the sources of error is the "error of representativeness," which arises because an observation of a variable at, say, a single point, is used to initialize that variable throughout a grid cell, and the value of the variable at that point includes a component due to turbulent motions, the properties of which vary over space and time. The PI will apply improved local estimates of observation error to the assimilation of atmospheric data from realistic observing systems under realistic atmospheric conditions. This will lead to full use being made of data from high quality observations, such as from Doppler radar, and it will permit an accurate evaluation of the utility of next-generation observing systems. Accurate estimates of observation error will also be applied to generating realistic ensembles of perturbed initial conditions for ensemble forecasts. Because these applications depend on accurate measurements of turbulent statistics, the PI will investigate the statistical description of turbulence from research aircraft and from radiosonde data.
Broader impacts of this research arise from its application to improving weather forecasts.
