Submitted to: Rangeland Ecology and Management
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: December 27, 2009
Publication Date: May 1, 2010
Repository URL: http://handle.nal.usda.gov/10113/58364
Citation: Karl, J.W. 2010. Spatial predictions of cover attributes of rangeland ecosystems using regression kriging and remote sensing. Rangeland Ecology and Management. 63:335-349. Interpretive Summary: Reliable information on the condition of large rangeland landscapes is necessary for successful management. Remote sensing techniques where field observations are related to imagery via statistical techniques (e.g., regression) are commonly used to produce estimates of rangeland attributes (e.g., percent vegetation cover) over large areas. While this has worked well in some situations, it has met with mixed success in part because it does not take advantage of all information contained in the field data. In this paper I describe a technique used by geologists and soil scientists called regression kriging that combines generalized least-squares (GLS) regression with the geostatistical technique of kriging to improve the regression predictions. I demonstrate this technique for three rangeland attributes (percent cover of shrubs, bare ground, and cheatgrass (Bromus tectorum) in a southern Idaho landscape. Regression kriging produced more accurate results than GLS regression for all three variables, but the ability of kriging to improve prediction accuracy is dependent on the degree of spatial autocorrelation between the field observations. These results demonstrate that regression kriging could be a valuable technique for assessment of rangeland conditions over large landscapes.
Technical Abstract: Sound rangeland management requires accurate information on rangeland condition over large landscapes. A commonly-applied approach to making spatial predictions of attributes related to rangeland condition (e.g., shrub or bare ground cover) from remote sensing is via regression between field and remotely-sensed data. This has worked well in some situations but has limited utility when correlations between field and image data are low and does not take advantage of all information contained in the field data. I compared spatial predictions from generalized least-squares (GLS) regression to a geostatistical interpolator, regression kriging (RK), for three rangeland attributes (percent cover of shrubs, bare ground, and cheatgrass [Bromus tectorum L.]) in a southern Idaho study area. The RK technique combines GLS regression with spatial interpolation of the residuals to improve predictions of rangeland condition attributes over large landscapes. I employed a remote-sensing technique, object-based image analysis (OBIA), to segment Landsat 5 Thematic Mapper images into polygons (i.e., objects) because previous research has shown that OBIA yields higher image-to-field data correlations and can be used to select appropriate scales for analysis. Spatial dependence, the decrease in autocorrelation with increasing distance, was strongest for percent shrub cover (samples autocorrelated up to a distance [i.e., range] of 19,098 m), but present in all three variables (range of 12,646 m and 768m for bare ground and cheatgrass cover, respectively). As a result, RK produced more accurate results than GLS regression alone for all three attributes when predicted versus observed values of each attribute were measured by leave-one-out cross-validation. The results of RK could be used in assessments of rangeland conditions over large landscapes. The ability to create maps quantifying how prediction confidence changes with distance from field samples is a significant benefit of regression kriging and makes this approach suitable for landscape-level management planning.