The Most Correlated Variables Are Open Land-Use in 1971 and 1999, Forested Land-Use In

Appendix 1

The most correlated variables are open land-use in 1971 and 1999, forested land-use in 1971 and 1999, forested land-use and open land-use in 1971, forested land-use and open land-use in 1999 (Fig A1.1). Due to the little land-use change that occurred from 1971 to 1999 the correlated variables are too similar, therefore most of them were excluded from the models. The distances to residential areas in 1971 and 1999 were also highly correlated. Distance to other roads and distance to residential areas were also correlated since more roads are built in highly populated areas. Lastly, stream length and distance to streams are negatively correlated, the greater the distance away from a stream the less stream length.

Figure A1.1: Correlation matrix for predictor variables. The correlation coefficients (r) for correlations of 0.6 or greater are shown in the upper portion of the matrix. Plots along the diagonal are histograms of the data values. All distance and length variables have been log+1 transformed. Variable codes are explained in Table 2.

Appendix 2

This appendix shows fitted and residual plots for the FSP and IPANE presence/absence models. Only significant predictors are presented. Residual plots for FSP (Figure A2.1) and IPANE (Figure A2.2) show little influence of outliers on the overall results. Fitted estimates for FSP (Figure A2.3) and IPANE (Figure A2.4) are based on very different predictor variables.

Figure A2.1. Residual plots for significant FSP binomial model predictors show little influence of outliers.

Figure A2.2. Residual plots for significant IPANE binomial model predictors show little influence of outliers.

Figure A2.3. Fitted estimates for significant FSP binomial model predictors.

Figure A2.4. Fitted estimates for significant IPANE binomial model predictors.

Appendix 3

This appendix shows predictions relative to observed data as well as fitted and residual plots for the FSP and IPANE count models. Figure A3.1 shows box plots of predicted vs. observed count for the FSP and IPANE models. Residual plots for FSP (Figure A3.2) and IPANE (Figure A3.3) show little influence of outliers on the overall results. Only significant predictors are presented in the residual and fitted plots. Fitted estimates for FSP (Figure A3.4) and IPANE (Figure A3.5) are both more complex than the presence/absence models. In this case, distance to residential areas is influential in both models. But, FSP is further influenced by open land use in 1999, while IPANE is further influenced by distance to roads and distance to streams.

Figure A3.1. Box plot comparison of predicted vs. observed counts for FSP (left) and IPANE (right). The FSP model had a higher correlation between predicted and observed (r = 0.34) than the IPANE model (r = 0.23), but both relationships were weak.

Figure A3.2. Residual plots for significant FSP count model predictors show little influence of outliers.

Figure A3.3. Residual plots for significant IPANE count model predictors show little influence of outliers.

Figure A3.4. Fitted estimates for significant FSP count model predictors. Open lands in 1999 and distance to residential areas in 1971 are the two most influential variables in the model.

Figure A3.4. Fitted estimates for significant IPANE count model predictors. Distance to other roads, distance to streams and distance to residential areas in 1971 are the most influential variables in the model.