A gridded multisite weather generator and synchronization to observed weather data

Procedures are described for constructing a daily multisite weather generator at a collection of arbitrary (e.g., gridded) locations and for synchronizing the gridded generator to observed weather series at a set of reference stations. The gridded generator is constructed by interpolating conventional single-station weather generator parameters using locally weighted regressions and producing coherent simulations of daily weather from them using spatial correlation functions. When implemented, the synchronization algorithm results in simulated spatial weather fields at the grid points that are consistent with daily weather observations at nearby locations for particular years. The synchronization is achieved by exploiting the latent multivariate Gaussian structure of the spatially distributed weather generator and making use of well-known statistical results that define conditional multivariate Gaussian distributions given known values for a subset of variables from the larger joint distribution. The primary focus is on precipitation, but the nonprecipitation variables in the weather generator are also amenable to gridding and to synchronization with nearby observed weather series. The motivating idea is to allow calibration of spatially distributed hydrological models consistent with the climate of the spatial weather generator, potentially allowing more realistic hydrological simulation, but the procedure may also be useful for interpolation of missing daily weather data.