Ninth International Geostatistics Congress, Oslo, Norway
June 11 – 15, 2012


Plenary 9

Abstract No.:



Using Stochastic Partial Differential Equation Models for Spatial Reconstruction of Annual Precipitation


R. Ingebrigtsen, Norwegian University of Science and Technology (NO)
F. Lindgren, Norwegian University of Science and Technology (NO)
I. Steinsland, Norwegian University of Science and Technology (NO)


This work is motivated by the needs of spatial reconstruction of climate, especially precipitation, in hydro power planning. Traditionally, statistical models for reconstruction of precipitation only include topographical attributes, such as altitude, in the expectation term and not in the dependency structure. These models will therefore often miss the topographical impact on the precipitation level. In this work we will present a model that incorporate topography, not only as an altitude covariate, but also in the covariance structure.

Annual precipitation is a non-stationary process which depends locally on the topography and dominating wind direction. To allow for flexible enough modeling of the covariance structure, we will use a stochastic partial differential equation (SPDE) approach to represent the Gaussian field. The benefit of such an approach is twofold. First, we can simply embed the SPDE upon a topographic map, constructed to capture local topographic dependence, and easily include the effect of various covariates in a physical meaningful way. Secondly, we obtain a Markov representation of the Gaussian field which makes computations feasible (see Lindgren, Lindström and Rue (2011) for details).

The goal is to reconstruct an expectation map (with uncertainty) of annual precipitation over regions of Norway using data from the past years.




    Copyright 2012 International Geostatistics Congress

Produced by X-CD Technologies Inc.