Matching of Observational Accuracy and Sampling Resolution in Meteorological Data Acquisition Experiments

Abstract

On the basis of a 10-year record of rawin observations in the Caribbean, experiments were carried out to determine the appropriate matching between the accuracy and density of meteorological observations. The results indicate that at 850 mb, if the distance between neighboring stations is of the order of 800 km or more, it makes little difference whether the random errors of observations are 2 kt or twice this amount. This is especially true if the observations are used to construct a regular grid of interpolated values. At 200 mb, an even larger range of observational errors is permissible if the distance between stations is large. The experiments further demonstrate that the distance between the raw observations must be smaller than that between the reconstructed grid points if the field gradients are to be computed with reasonable accuracy. As the scale of interest becomes smaller, accuracy requirements become more stringent. For instance, if we can measure the zonal wind components at 850 mb in January with an accuracy of 3 kt and if we have two observations 100 km apart, there is only a 0.2 probability that the error in the measured difference between the two observations is equal to or less than one-fifth of the true difference. To increase this probability to 0.4, the accuracy of the measurement must be within 1 kt. Similarly, if the rms error of temperature observations, 100 km apart, is 1C, the probability is 0.15 that the error in the measured difference is equal to or less than one-fifth of the true. On the other hand, if the rms observation error is only 0.1C, the probability is 0.6 that the error is equal to or less than one-fifth of the true difference. The study is intended to provide planners of meteorological data acquisition experiments with some insight into the options at their disposal. Thus, for field experiments designed to delineate broad-scale atmospheric features, it should help identify the point of diminishing returns for instrumental accuracy. Conversely, if only a given instrumental accuracy is obtainable, it should provide an estimate of the maximum useful sampling resolution.