A daily weather generator based on a two-stage resampling algorithm

A two-stage time-series resampling algorithm is presented that is capable of generating daily values of weather variables outside their historical ranges. In this algorithm the simulated daily values are composed of an expected value and a sampled historical residual. The residuals broaden the range of the simulated daily values. Both the estimation of the expected value and the sampling of the residuals are based on the nearest-neighbours concept. In particular the influence of the neighbourhood sizes in both nearest-neighbour searches was studied. The algorithm was tested with data generated by two theoretical time-series models. Using observed precipitation and temperature data, a 12,000-year series of precipitation and temperature for the Ourthe catchment (Belgium) was simulated and used as input for a rainfall–runoff model to produce a long synthetic sequence of daily discharge. The two-stage algorithm correctly reproduces the mean, standard deviation and lag 1 autocorrelation of daily precipitation. The simulated distributions of 4-day and 10-day precipitation maxima in winter also show good correspondence with those observed, while the largest daily amounts substantially exceed those in the original data. However, the widened range of daily precipitation amounts has no discernible effect on the simulated discharge maxima in winter.