A spatial time series framework for simulating daily precipitation at regional scales

A spatiotemporal stochastic simulation approach for constructing maps of daily precipitation at regional scales in a hindcast mode is proposed in this paper. Parametric temporal trend models of precipitation are first established at the available rain gauges. Temporal trend model parameters are then regionalized in space accounting for their spatial auto- and cross-correlation, as well as for their relationships with auxiliary spatial information such as terrain elevation. The resulting residual values at the rain gauges are modeled as a realization of a stationary spatiotemporal process. Sequential simulation is then used to generate alternative synthetic realizations of daily precipitation fields, which reproduce: (i) the rain gauge measurements, and (ii) their histogram and a model for their spatiotemporal correlation over the entire study region and time period of interest. In addition, a post-processing transformation allows reproduction of the rainfall histogram at particular dates, while preserving the observed rain gauge data. A case study illustrates the applicability of the proposed methodology using daily precipitation measurements recorded at 77 rain gauges in the northern California coastal region from Nov 1, 1981 to Jan 31, 1982. Conditional stochastic simulation in space and time is performed for generating a 30-member ensemble of daily precipitation fields on a 300×360 grid of cell size 1 km2 for the above time period. It is shown that simulated precipitation fields reproduce the spatiotemporal characteristics of the rain gauge measurements, thus providing realistic inputs of precipitation forcing for hydrologic impact assessment studies.