PARENT SESSION
Oral Session 7: GIS / Remote Sensing I.
Presiding: E Ellis and P Valko
Monday, August 2, 8:00 AM to 11:30 AM, Meeting Room B 116.

Incorporating autecological information in image-based vegetation mapping.

Dobrowski, Solomon^*,1, Greenberg, Jonathan¹, Ramirez, Carlos¹, Ustin, Susan¹, ¹ Center for Spatial Technologies and Remote Sensing, Davis, CA, USA

ABSTRACT- Accurately mapping vegetation floristics at high thematic resolution using remotely sensed data remains an elusive target. Recently, much attention has been placed on the use of ancillary information layers in image classification (e.g. digital elevation models) in that they provide indirect links to information that is ecologically relevant to species distributions. Incorporating direct autecological information in image-based vegetation mapping remains a challenge. We used spatially explicit non-parametric regression modeling in order to incorporate autecological information in the production of an existing vegetation map for the Lake Tahoe basin. Probability surfaces for 23 species were produced using generalized additive modeling (GAM) as implemented by a group of S-plus functions called generalized regression analysis and spatial prediction (GRASP). Models were fit to plot data obtained from multiple resource agencies, using climatic and land-form based explanatory variables derived from a DEM. A logistic link function and a binomial error term were used in the GAM modeling. Model selection was based on a step-wise procedure using the AIC statistic. Model evaluation was assessed by cross-validation resulting in a range of accuracies for individual species (ROC values from 0.58 to 0.85) depending primarily on the efficacy of the plot sampling. Probability surfaces for the basin were subsequently generated within a GIS. These surfaces were spatially re-sampled and used in conjunction with IKONOS imagery for use in vegetation mapping. The GAM surfaces were used as both independent information sources within the context of multi-source classification, as well as an attribute layer in image classification using both maximum-likelihood and decision tree modeling. Results from the analysis demonstrate that the inclusion of the GAM surfaces improved individual class accuracies significantly. This analysis suggests the need for implementing standardized and objective species modeling techniques for improving vegetation maps.

Key words: generalized additive modeling, image classification, vegetation mapping, ikonos