A random resampling approach to evaluate spatial representativeness of short temperature time-series

An issue that can arise when dealing with non-standard climate data sets is determining a minimum length for time-series before they can be included in an analysis. A method that uses a Monte-Carlo random resampling technique has been developed to deal with this question. The technique consists of randomly selecting a series of data pairs from a control (i.e. long-series) non-standard weather station and a standard (government) weather station. Regression coefficients are determined for the data set. The process is repeated sufficiently to generate a stable mean and variance for the regression coefficients. The number of randomly selected data pairs is then incremented and the process is re-run; this continues until the full data set is being retained. The point at which the generated regression coefficients are consistently similar to the regression coeffients generated using all the data is considered to be the end point - a time series with this number of observations closely approximates the regression results obtained with a longer time series, and thus may be included in an analysis. This problem is similar to that of determining how many missing data points a data set can have before it no longer gives a reliable period mean. Results using traditional random-resampling and a "time-series" resampling are compared.