A COMPARISON OF GEOSTATISTICAL PROCEDURES FOR SPATIAL-ANALYSIS OF PRECIPITATION IN MOUNTAINOUS TERRAIN

Spatially distributed measurements or estimates of precipitation over a region are required for modelling of hydrologic processes and soil moisture for agricultural and natural resource management. Simple interpolation methods fail to consider the effects of topography on precipitation and may be in considerable error in mountainous regions. The performance of three geostatistical methods for making mean annual precipitation estimates on a regular grid of points in mountainous terrain was evaluated. The methods were: (1) kriging; (2) kriging elevation-detrended data; and (3) cokriging with elevation as an auxiliary variable. The study area was the Willamette River basin, a 2.9 million hectare region spanning the area between the Coast Range and the Cascade Range in western Oregon. Compared with kriging, detrended kriging and cokriging both exhibited better precision (as indicated by estimation coefficients of variation of 16 and 17% vs. 21%; and average absolute errors of 19 and 20 cm vs. 26 cm) and accuracy (as indicated by average errors of -1.4 and -2.0 cm vs. -5.2 cm) in the estimation of mean annual precipitation. Contour diagrams for kriging and detrended kriging exhibited smooth zonation following general elevation trends, while cokriging showed a patchier pattern more closely corresponding to local topographic features. Detrended kriging and cokriging offer improved spatially distributed precipitation estimates in mountainous terrain on the scale of a few million hectares. Application of these methods for a larger region, the Columbia River drainage in the USA (57 million hectares), was unsuccessful due to the lack of a consistent precipitation-elevation relationship at this scale. Precipitation estimation incorporating the effects of topography at larger scales will require either piecewise estimation using the methods described here or development of a physically based orographic model.