Representativity is not a fixed attribute of observations, since it depends on user requirements. Most station networks have been established for meso- and macroscale users, e.g. forecasters. Agrometeorological application is generally at local scale. Use of the nearest single network station as a data source encounters two main problems, namely usefulness of the station observations and extrapolation of these data to the application location. It is assumed that realtime remote sensing support is not available, and of course mathematical interpolation methods, such as kriging, are irrelevant.
Station observation procedures for non-macroscale users are discussed. It is shown that, for general applicability of local-scale extrapolation methods, the actual station observations should preferably be normalized into hypothetical standard parameters, such as potential wind speed or grass reference evapotranspiration. The practical question is, how realtime production of such standardized half-products can be organized economically. Some proposed requirements are technical, such as sampling procedures which allow output of trend-corrected variances, or systematical azimuth-dependent processing. Other requirements are circumstantial, such as actual knowledge of instrument exposure, or availability of good information on local albedo, soil properties and surrounding roughness. Existing extrapolation methods are very briefly reviewed, separately for parameters with basic input from the atmosphere, such as wind, and parameters with a basic surface source, such as humidity. For each, an estimate is given of the geographical range of acceptable representativity at agricultural scales in moderately heterogeneous terrain. It is also indicated, in which way the possibility of extrapolation is influenced by the presence or absence of certain realtime observational input or circumstantial information.
Special Session -- Weather Data Requirements for Integrated Pest Management