EFFECT OF DATA COVERAGE ON THE ACCURACY OF 500-MB. FORECASTS

Abstract

Networks of data are simulated by interpolating height and wind from a hypothetically ?correct? analysis at a uniform array of points. The interpolated values are added to random numbers whose statistics are typical of non-systematic errors of observation, and are then regarded as genuine data. Such artificially constructed ?data? for several networks of different densities are analyzed independently. The differences between these analyses and the hypothetically ?correct? analysis are taken to be representative of the initial analysis error under conditions of varying station density. Numerical forecasts computed from the different analyses are compared with the forecast made from the ?correct? initial data. Several such comparisons indicate that initial analysis errors do not grow to an important degree so long as the spacing between synoptic reports confines error fields to a scale smaller than that of the synoptic disturbances. However, with data spacing comparable with that over existing regions of poor data coverage, initial errors are amplified two-fold in a 48-hr. forecast interval. Further experiments, in which the data are analyzed objectively to eliminate inconsistencies of analysis, are carried out.