A geographically weighted regression analysis of Douglas-fir site index in north central Idaho

Abstract

An analysis of Douglas-fir (Pseudotsuga menziesii [Mirb.] Franco var. glauca) site index (SI) was performed in north central Idaho using two forms of linear regression: standard multiple linear regression (MLR) and geographically weighted regression (GWR). The hypothesis was that the GWR model would provide better estimates of SI using edaphic, topographic, and climatic predictor variables than ordinary MLR. Elevation, volcanic ash depth, slope, and aspect were significantly correlated with Douglas-fir SI (RA2 = 0.5). GWR accounted for an additional 29% of the variation in SI and reduced the error sum of squares by approximately 53%. The geographically weighted parameter estimates of the model intercept and elevation were nonstationary, which indicates that geographic location determines the influence of a variable on SI. Residual analysis from a MLR model showed large SI prediction error throughout the study area. This prediction error was significantly reduced using the GWR model. GWR shows promise for biological modeling. Copyright © 2008 by the Society of American Foresters.