Daily spatiotemporal precipitation simulation using latent and transformed Gaussian processes
A stochastic simulator for spatiotemporal daily precipitation amounts is proposed. It reduces to a simple generalized linear model at individual locations, accounting for the persistence of dry or wet spells and the skewness of the intensity distribution. To account for the spatial dependence of occurrence and intensity, the simulator uses a combination of latent and probability integral transformed Gaussian processes. It has the advantage of not only permitting spatial interpolation, but of attaching uncertainty to interpolated parameters via Kriging.
Authors: William Kleiber, Richard W. Katz, and Balaji Rajagopalan
Full text: www.agu.org
Published in: Water Resources Research, Vol. 48, W01523, pp. 17
Regional climate model assessment using statistical upscaling and downscaling techniques.
Using geostatistical tools to go from observations to grid squares (upscaling) and hierarchical modeling to go from grid squares to observations (downscaling) we study the performance of the Swedish Meteorological and Hydrological Institute regional climate model, run in weather forecasting mode. We find that the upscaling appears to need information about altitude, that the model sometimes drastically fails to agree with data, and that an urban heat island correction likely would improve both data and model output.
Node: University Of Michigan
Authors: Veronica J. Berrocal, Peter F. Craigmile, and Peter Guttorp.
Full text: download
Published in:Environmetrics, Vol. 23, No. 5, pp. 482-492
Interpolation of nonstationary high frequency spatial-temporal temperature data
We conduct a careful statistical analysis of high frequency spatial-temporal temperature data observed at a number of fixed monitoring facilities in the Southern Great Plains region of the U.S. The data were collected as part of the Department of Energy’s Atmospheric Radiation Measurement program. The temperature data exhibit nonstationary correlation and variance, which can be partially explained by varying amounts of incoming solar radiation, as well as spatial-temporal jumps due to cold fronts moving across the region. To model the data, we fit a nonstationary spatial-temporal model with jumps, and after the model is fit, we simulate temperatures at new locations conditional on the temperature records at the existing monitoring facilities. When the simulations are conducted repeatedly, the result is a suite of artificial interpolated temperature values that reflect the various uncertainties inherent in the spatial-temporal statistical model. In particular, the interpolated values are more uncertain at locations that are on the exterior of the region, during the daytime when temperatures tend to be more chaotic, and near the time of the cold front, as the trajectory of the cold front is not known exactly.