Theoretical and Practical Considerations for Network Design

Abstract

Two main approaches are available for the rational design of networks, the statistical and the experimental. The statistical approach presupposes the ability to compute accurate correlation functions of the meteorological fields to be observed and analyzed. Strictly speaking, these correlations depend not only on distances between various points but also on the locations and relative positions of these points. However, as a matter of practical convenience, the assumption is usually made that the correlations are homogeneous and isotropic.

By making use of the correlation functions and the statistics of errors of observations, it is possible to interpolate values of meteorological elements such that the rms error of interpolation is a minimum. This method, known as the method of optimum interpolation, has been used to determine characteristic distances between stations which are associated with any required interpolation accuracy. The method is capable of taking into account the effect of random errors and provides an insight into the trade-off relationships between station density and accuracy of observation.

The experimental approach attempts to simulate the numerical analysis/forecast cycle with data input of varying densities and compare the results against reference analyses and forecasts made from very dense initial data. This approach presupposes a capability to generate a reference atmosphere of high meteorological verisimilitude to serve as a data source for the experiments.

The simulated data should comprise various possible combinations of conventional and non-conventional observations such as radiometric measurements from satellites and wind measurements by floating balloons. The simulation experiments should take into consideration the effects of instrumental errors, the possible trade-off between observational density and frequency, and model characteristics including the resolution of the computational grid. For completeness, at least one situation typical of each season should be considered.