Effect of observational sampling error on the detection of anthropogenic climate change

The effect of sampling error in surface air temperature observations is ass essed for detection and attribution of an anthropogenic signal. This error arises because grid-box values are based on varying densities of station an d marine data. An estimate of sampling error is included in the application of an optimal detection and attribution method based on June-August trends over SO yr. The detection and attribution method is applied using both the full spatial pattern of observed trends and spatial patterns from which th e global mean warming has been subtracted. Including the effect of sampling error is found to increase the uncertainty in estimates of the greenhouse gas-plus-sulfate aerosol signal from observ ations by less than 2%-6% for recent trend patterns (1949-98), and 3%-8% fo r signal estimates from observations in the first half of the twentieth cen tury. Random instrumental error shows even smaller effects. However, the ef fects of systematic instrumental errors, such as changes in measurement pra ctices or urbanization, cannot be estimated at present. The detection and a ttribution results for recent 50-yr summer trends are very similar between the case including and the case disregarding the global mean. However, resu lts based on observations from the first half of the twentieth century yiel d high signal amplitudes with global mean and low ones without, suggesting little pattern agreement for that warming with the anthropogenic climate ch ange signal.