The spatial structure of monthly temperature anomalies over Australia

Thirty years of quality controlled surface data from a network of 95 stations have been used to examine the limits which spatial sampling place on the analysis of Australian temperature variability. Use has been made of two methodological paradigms, with the goal of documenting the ability of a typical climate network combined with modern analysis techniques to support the monitoring of short-term climate variability over Australia. In the first, the representativeness of stations’ observations has been investigated through a parameter-based description of the scales and errors of monthly station data. This has revealed temperature anomalies to be highly coherent, with decorrelation scales longer than the typical distance between stations in our operationally feasible network, and for the associated observational error variance to be relatively modest, being typically 10 per cent to 20 per cent of the total variance. The parameter-based results have been complemented by a series of experiments in which objective analysis techniques were applied to the independent estimation of observed monthly anomalies. The methods examined were successive correction (Barnes), statistical interpolation, and Laplacian smoothing splines, chosen because of their relative popularity, and because of their differing approach to the analysis problem. For each method and both mean monthly maximum and minimum temperature the root mean square error in estimating observed values was near 0.6°C, with a ratio to the standard deviation of approximately 0.45. The most important determinant of this error was found to be local station density, with an approximate doubling of errors from the well observed southeast through to the sparsely observed northwest of the continent for all methods. The intercomparison of analysis methods has revealed statistical interpolation to be most accurate, followed by the second order (m=2) Laplacian smoothing spline and then successive correction. The differences in root mean square error between the techniques show wide statistical significance, but are otherwise modest, typically amounting to less than 0.1°C. Collectively, the results of this study imply that station temperature observations are representative of the variability over a substantial area of surrounding space, and networks such as that used here are sufficient to support the monitoring of Australian temperature variability with reasonable accuracy.