1

Appendix 1

Relationship of NDVI to field-collected vegetation data

A vegetation inventory on Ft. Benning was conducted on selected stands in 2007, and updated and completed in 2011. Although the forest inventory was not useful for the current study because it was only conducted in a subset of forest stands, we used the detailed vegetation data to validate the use of NDVI to describe vegetation on the installation.

During the inventory, ground, midstory and canopy vegetation were characterized at the stand level where stands were a minimum of 2 ha in size. Sampling points within each stand were allocated randomly within a stand at a rate of approximately 1 plot per acre. Ground vegetation was measured as the percent composition of herbaceous, bare ground, pine straw, and woody vegetation within a 3.4 m radius around sampling points. The height of the majority of hardwood midstory vegetation within an 11.3 m radius around sampling points was categorized as low (<2.1 m), medium (2.1 – 4.6 m), or tall (>4.6 m), and hardwood midstory density was categorized as sparse, moderate, or dense. Basal area of pine and hardwood trees at each sampling point was measured using a 10 BAF Prism Plot. Values from sampling points were then averaged to yield a single average value per stand for each ground cover category, midstory height, midstory density, pine basal area, and hardwood basal area.

To analyze the relationship between NDVI and these stand-level field measurements, we first calculated mean NDVI for each stand using the “Intersect Polygons With Raster” tool in the Geospatial Modeling Environment (Beyer 2012). We then regressed mean NDVI against all vegetation variables in a multiple linear regression, after performing transformations as needed to obtain normal variable distributions. Variance inflation factors (VIF) were calculated for all predictor variables and when two or more variables showed VIF greater than 3, we eliminated one variable with weakest bivariate support. We carried out these analyses separately for the two NDVI variables included in the resource selection probability function: winter NDVI, and the difference between winter and summer NDVI (NDVI difference).

Only one variable displayed a strongly skewed distribution: we used a logarithmic transformation to transform hardwood basal area per acre to normality.

Both winter NDVI and NDVI difference were significantly predicted by field vegetation measures (winter NDVI adjusted R2: 0.18, F6,2482 = 89.0, p < 0.001; NDVI difference adjusted R2: 0.25, F6,2316 = 130.2, p < 0.001). Winter NDVI was dominated by a strong positive association with pine basal area, but was also significantly predicted by midstory and some ground-level vegetation (Table 5). NDVI difference was also significantly predicted by canopy, midstory and ground vegetation but was most strongly positively associated with hardwood basal area, followed by positive associations with midstory height and density (Table 6).

Table 5. Coefficients in a multiple linear regression of winter NDVI on stand-level vegetation measurements.

Term / Estimate / SE / t / p
Intercept / 1.25 / 0.06 / 20.98 / < 0.001
Pine basal area per acre / 0.01 / 0.0006 / 17.51 / < 0.001
Midstory density / 0.10 / 0.02 / 5.73 / < 0.001
Bare ground (%) / -0.002 / 0.0008 / -2.54 / 0.01
ln(hardwood basal area per acre) / 0.03 / 0.01 / 2.38 / 0.02
Midstory height / 0.03 / 0.01 / 2.31 / 0.02
Pinestraw (%) / 0.002 / 0.001 / 1.24 / 0.21

Table 6. Coefficients in a multiple linear regression of NDVI difference, calculated as the difference between winter and summer NDVI, on stand-level vegetation measurements.

Term / Estimate / SE / t / p
Intercept / -0.09 / 0.01 / -8.54 / < 0.001
ln(hardwood basal area per acre) / 0.02 / 0.002 / 12.50 / < 0.001
Midstory height / 0.02 / 0.002 / 9.61 / < 0.001
Midstory density / 0.01 / 0.002 / 5.46 / < 0.001
Pine basal area per acre / -0.0004 / 0.0001 / -4.93 / < 0.001
Herbaceous cover (%) / -0.0007 / 0.0002 / -3.85 / < 0.001
Bare ground (%) / 0.0003 / 0.0001 / 2.22 / 0.03

References

Beyer HL (2012) Geospatial modelling environment (version 0.7.1.0) software. http://www.spatialecology.com/gme. Accessed 11 Jul 2012

Appendix 2. Parameter estimates for all candidate models of gopher tortoise habitat selection.

Table 7. Soil model (model I)

Covariate / Estimate / SE / z / p
Intercept / -8.45 / 0.36 / -23.58 / < 0.001
% Sand in top 1m / 0.10 / 0.004 / 22.51 / < 0.001
Soil drainage index / -0.04 / 0.003 / -13.84 / < 0.001

Table 8. Vegetation model (model II)

Covariate / Estimate / SE / z / p
Intercept / 16.70 / 0.74 / 22.61 / < 0.001
Winter NDVI / -22.16 / 0.86 / -25.80 / < 0.001
NDVI difference / -21.90 / 0.86 / -25.49 / < 0.001

Table 9. Soil and vegetation model (model III)

Covariate / Estimate / SE / z / p
Intercept / -1.50 / 0.95 / -1.57 / 0.12
% Sand in top 1m / 0.09 / 0.005 / 16.77 / < 0.001
NDVI difference / -9.01 / 0.80 / -11.28 / < 0.001
Winter NDVI / -7.44 / 0.73 / -10.19 / < 0.001
Soil drainage index / -0.04 / 0.004 / -9.70 / < 0.001

Table 10. Abiotic model (model IV)

Covariate / Estimate / SE / z / p
Intercept / -11.73 / 0.32 / -36.85 / < 0.001
Elevation / 0.03 / 0.001 / 26.90 / < 0.001
Distance to roads / 0.0004 / 0.00002 / 22.51 / < 0.001
% Sand in top 1 m / 0.07 / 0.003 / 20.58 / < 0.001
Slope / -0.19 / 0.01 / -13.96 / < 0.001
Soil drainage index / -0.02 / 0.002 / -11.17 / < 0.001
Distance to water / 0.0003 / 0.00004 / 7.48 / < 0.001