Skip to main content
ARS Home » Plains Area » Las Cruces, New Mexico » Range Management Research » Research » Publications at this Location » Publication #251481
Title: Spatial dependence of predictions from image segmentation: A variogram-based method to determine appropriate scales for producing land-management information

Author
item Karl, Jason
item MAURER, BRIAN - Michigan State University

Submitted to: Ecological Informatics
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 2/25/2010
Publication Date: 7/1/2010
Citation: Karl, J.W., Maurer, B.A. 2010. Spatial dependence of predictions from image segmentation: A variogram-based method to determine appropriate scales for producing land-management information. Ecological Informatics. 5:194-202.

Interpretive Summary: Selecting scales for analysis of remotely-sensed images is a significant challenge in ecological studies. Choosing an appropriate scale can accentuate patterns of interest, but inappropriate scales can obscure patterns and alter relationships between imagery and field observations. Segmentation of images into objects has been shown to be an effective methods for scaling remotely-sensed data into units having ecological meaning, but the selection of image object sets to represent landscape patterns is largely subjective. We used observations of percent bare ground cover from field sites in southern Idaho to look at the changes in spatial autocorrelation of regression predictions and residuals for 10 different levels of image segmentation to determine if the degree of spatial autocorrelation could be used to help select appropriate scales of image analysis. We found that the segmentation level whose regression predictions had spatial dependence that most closely matched the spatial dependence of the field samples also had the strongest predicted-to-observed correlations. With the incorporation of a geostatistical interpolator to predict the value of regression residuals at unsampled locations, however, we achieved consistently strong correlations across many segmentation levels. These results suggest that while an optimal analysis scale can be defined for single variables, the use of statistical techniques to account for unexplained spatial autocorrelation can yield consistently high results within a scaling domain, reducing the need to search for a single "optimal" scale.

Technical Abstract: A significant challenge in ecological studies has been defining scales of observation that correspond to the relevant ecological scales for organisms or processes of interest. Remote sensing has become commonplace in ecological studies and management, but the default resolution of imagery often used in studies is an arbitrary scale of observation. Segmentation of images into objects has been proposed as an alternative method for scaling remotely-sensed data into units having ecological meaning. However, to date, the selection of image object sets to represent landscape patterns has been largely subjective. Changes in observation scale affect the variance and spatial dependence of measured variables, and may be useful in determining which levels of image segmentation are most appropriate for a given purpose. We used observations of percent bare ground cover from 346 field sites in a semi-arid shrub-steppe ecosystem of southern Idaho to look at the changes in spatial dependence of regression predictions and residuals for 10 different levels of image segmentation. We found that the segmentation level whose regression predictions had spatial dependence that most closely matched the spatial dependence of the field samples also had the strongest predicted-to-observed correlations. This suggested that for percent bare ground cover in our study area an appropriate scale could be defined. With the incorporation of a geostatistical interpolator to predict the value of regression residuals at unsampled locations, however, we achieved consistently strong correlations across many segmentation levels. This suggests that if spatial dependence in percent bare ground is accounted for, a range of appropriate scales could be defined. Because the best analysis scale may vary for different ecosystem attributes and many inquiries consider more than one attribute, methods that can perform well across a range of scales and perhaps not at a single, ideal scale are important. More work is needed to develop methods that consider a wider range of ways to segment images into different scales and select sets of scales that perform best for answering specific management questions. The robustness of ecological landscape analyses will increase as methods are devised that remove the subjectivity with which observational scales are defined and selected.