Simulation of 6-hourly rainfall and temperature by two resampling schemes

The joint simulation of time series of 6-hourly precipitation and temperature using nearest-neighbour resampling is studied for Maastricht, the Netherlands. Two resampling schemes are considered: (i) straightforward resampling of 6-hourly values, and (ii) resampling of daily values followed by disaggregation into 6-hourly values using the method of fragments. Second-order statistics of the simulated values are compared with those in the observed data. It appeared that straightforward resampling of 6-hourly values does not adequately preserve the slow decay of the autocorrelation functions of precipitation and temperature. As a result the standard deviations of the monthly precipitation totals and monthly average temperature are strongly underestimated. A negative bias also shows up in the quantiles of the multi-day seasonal maximum precipitation amounts. The autocorrelation coefficients and the standard deviations of the monthly values are much better reproduced if the daily values are generated first. A good correspondence between the historical and simulated distributions of the seasonal maximum precipitation amounts is also achieved with this alternative resampling scheme. (C) 2003 Elsevier Science B.V. All rights reserved.