Spatiotemporal Stochastic Simulation of Monthly Rainfall Patterns in the United Kingdom (1980–87)
Journal
Journal of climate
Date Issued
August 15, 2007
DOI
10.1175/JCLI4233.1
Abstract
With few exceptions, spatial estimation of rainfall typically relies on information in the spatial domain
only. In this paper, a method that utilizes information in time and space and provides an assessment of
estimate uncertainty is used to create a gridded monthly rainfall dataset for the United Kingdom over the
period 1980–87. Observed rainfall profiles within the region were regarded as the sum of a deterministic
temporal trend and a stochastic residual component. The parameters of the temporal trend components
established at the rain gauges were interpolated in space, accounting for their auto- and cross correlation,
and for relationships with ancillary spatial variables. Stochastic Gaussian simulation was then employed to
generate alternative realizations of the spatiotemporal residual component, which were added to the estimated
trend component to yield realizations of rainfall (after distributional corrections). In total, 40 realizations
of rainfall were generated for each month of the 8-yr period. The methodology resulted in reasonably
accurate estimates of rainfall but underestimated in northwest and north Scotland and northwest
England. The cause for the underestimation was identified as a weak relationship between local rainfall and
the spatial area average rainfall, used to estimate the temporal trend model in these regions, and suggestions
were made for improvement. The strengths of this method are the utilization of information from the time
and space domain, and the assessment of spatial uncertainty in the estimated rainfall values
only. In this paper, a method that utilizes information in time and space and provides an assessment of
estimate uncertainty is used to create a gridded monthly rainfall dataset for the United Kingdom over the
period 1980–87. Observed rainfall profiles within the region were regarded as the sum of a deterministic
temporal trend and a stochastic residual component. The parameters of the temporal trend components
established at the rain gauges were interpolated in space, accounting for their auto- and cross correlation,
and for relationships with ancillary spatial variables. Stochastic Gaussian simulation was then employed to
generate alternative realizations of the spatiotemporal residual component, which were added to the estimated
trend component to yield realizations of rainfall (after distributional corrections). In total, 40 realizations
of rainfall were generated for each month of the 8-yr period. The methodology resulted in reasonably
accurate estimates of rainfall but underestimated in northwest and north Scotland and northwest
England. The cause for the underestimation was identified as a weak relationship between local rainfall and
the spatial area average rainfall, used to estimate the temporal trend model in these regions, and suggestions
were made for improvement. The strengths of this method are the utilization of information from the time
and space domain, and the assessment of spatial uncertainty in the estimated rainfall values