A pixel‐based semi‐empirical system for predicting vegetation diversity in boreal forest

Abstract

We have designed a system that predicts species richness in the mixed wood boreal forest of Canada. The system is based on a simple multivariate linear model that uses four landscape characteristics as independent variables: canopy species type, distance from the nearest ridgeline, time since the last fire and canopy stem density. The model is shown to provide statistically significant estimates of richness when using observed independent variables. We developed models for estimating the four landscape characteristics from geospatial data consisting of remotely sensed imagery and a digital elevation model. We ran the model at the stand scale and the pixel scale and found that stand scale predictions were be more accurate that pixel scale predictions. We produced a map of vegetation species richness for Prince Alberta National Park in central Saskatchewan Canada that is consistent with our expectations. We also estimated the uncertainty in the four landscape characteristic estimates and developed a methodology for propagating this uncertainty through the system to produce estimates of uncertainty in the pixel‐based richness predictions. While the uncertainty is significant, the estimation and management of uncertainty in a mapping system of this type represents an innovation.

Acknowledgements

We wish to acknowledge our collaborators, Dr. E.A. Johnson of the Department of Biological Sciences at the University of Calgary, Sylvia Taylor (née Chipman) of Alpine Environmental Ltd. of Calgary and Dr. P. Ehlers of the Department of Mathematics and Statistics of the University of Calgary. Data was provided by Parks Canada and the Canada Centre for Remote Sensing. Funding came from the University of Calgary, NSERC, and two Networked Centres of Excellence (www.nce.gc.ca): Geomatics for Informed Decisions: GEOIDE (www.geoide.ulaval.ca) and the Sustainable Forest Management Network (sfm‐1.biology.ualberta.ca).