Abstract
Objective analysis procedures such as statistical interpolation require reliable estimates of forecast-error statistics in order to optimize the analysis weights. Reasonably good estimates of the forecast-error statistics can be obtained from radiosonde networks by the zero lag innovation covariance technique. However, over the data-sparse regions of the tropics, Southern Hemisphere, and oceans, these techniques cannot he applied and much more ad hoe procedures must be used.
This study uses a simple Kalman filter system to actually generate forecast-error statistics for a hierarchy of wind-height observation networks-from uniform, time-invariant networks to inhomogeneous, time-dependent networks. The forecast-error statistics are characterized by their variance and measures of their spatial scale and anisotropy. Several methods of generating forecast-error statistics in data-sparse regions are compared with the optimal results.
