Ecological Archives A022-122-A2

Amanda E. Martin, Lenore Fahrig. 2012. Measuring and selecting scales of effect for landscape predictors in species–habitat models. Ecological Applications 22:2277–2292. http://dx.doi.org/10.1890/11-2224.1

Appendix B. Summaries of the best single- and multi-scale habitat models for each species and each statistical framework.

Table B1. Summaries of single-scale (SS) and multi-scale (MS) occupancy models selected through multi-model inference, where all landscape predictors were measured at the same scale in SS models and predictors were each measured at their individual scale of effect in MS models. We include the Nagelkerke R2 (RN2), number of parameter estimates per model (k), change in Akaike Information Criterion (corrected for small samples) from the most supported model (ΔAICc) and Akaike weight (wi) for each model with ΔAICc = 2. For included predictors, p() indicates effects on the detection probability,  ψ() indicates effects on the probability of occurrence. LC = local cover type, F = forest, W = wetland, G = grass, C = crop, H = housing, R = road, E = edge, S = stream.

Species Model Included predictors RN2 k ΔAICc wi
Striped skunk SS p(H, R)  ψ(F, W, G, C, E) 0.46 9 0.00 0.18
p(H)  ψ(F, W, G, C, E) 0.43 8 0.16 0.16
p(G, H)  ψ(F, W, G, C, E) 0.45 9 0.80 0.12
 ψ(F, W, G, C, H, E) 0.42 8 0.96 0.11
p(W)  ψ(F, W, G, C, H, E) 0.45 9 1.20 0.10
p(W, G)  ψ(F, W, G, C, H, E) 0.47 10 1.23 0.10
p(W, H)  ψ(F, W, G, C, E) 0.44 9 1.33 0.09
p(H, R)  ψ(F, W, G, C, H, E) 0.47 10 1.44 0.09
p(E)  ψ(F, W, G, C, H, E) 0.44 9 2.00 0.07
MS  ψ(LC, F, W, H, E) 0.54 10 0.00 0.50
 ψ(LC, F, W, H, R, E) 0.56 11 0.87 0.32
 ψ(LC, F, W, G, E) 0.52 10 1.97 0.19
Ermine SS p(H)  ψ(W, S) 0.38 5 0.00 0.29
p(R)  ψ(W, S) 0.36 5 0.96 0.18
p(H)  ψ(W, R, S) 0.39 6 1.26 0.15
p(H)  ψ(W, G, S) 0.39 6 1.38 0.15
p(R) ψ(W, G, S) 0.38 6 1.75 0.12
p(W, H) ψ(W, S) 0.38 6 1.94 0.11
MS ψ(W, G, R) 0.38 5 0.00 0.13
ψ(W, R) 0.34 4 0.59 0.10
p(W)  ψ(W, R) 0.37 5 0.63 0.09
p(F)  ψ(W, G, R) 0.40 6 0.97 0.08
p(H)  ψ(W, G, R) 0.40 6 0.98 0.08
 ψ(G, C, R) 0.37 5 1.01 0.08
p(H)  ψ(F, W, G, R) 0.43 7 1.13 0.07
p(G)  ψ(W, G, R) 0.40 6 1.29 0.07
p(H)  ψ(G, C, R) 0.40 6 1.35 0.07
p(C)  ψ(W, G, R) 0.40 6 1.37 0.06
 ψ(F, W, G, R) 0.40 6 1.39 0.06
 ψ(F, W, H, R) 0.40 6 1.39 0.06
p(F)  ψ(F, W, H, R) 0.43 7 1.68 0.06

 

Table B2. Summaries of single-scale (SS) and multi-scale (MS) generalized linear models selected through multi-model inference, where all landscape predictors were measured at the same scale in SS models and predictors were each measured at their individual scale of effect in MS models. We include the Nagelkerke R2 (RN2), number of parameter estimates per model (k), change in Akaike Information Criterion (corrected for small samples) from the most supported model (ΔAICc) and Akaike weight (wi) for each model with ΔAICc = 2. LC = local cover type, F = forest, W = wetland, G = grass, C = crop, H = housing, R = road, E = edge, S = stream.

Species Model Included predictors RN2 k ΔAICc wi
Striped skunk SS F, W, G, C, H, E 0.46 7 0.00 0.27
F, E, W 0.34 4 0.81 0.18
F, W, C, H, E 0.41 6 1.20 0.15
F, W, H, R, E 0.41 6 1.21 0.15
F, W, H, E 0.37 5 1.35 0.14
F, W, G, C, H, R, E 0.47 8 1.95 0.10
MS F, E, W 0.34 4 0.00 0.18
F, W 0.30 3 0.16 0.16
F, W, H, E 0.38 5 0.24 0.16
F, W, G, E 0.36 5 1.29 0.09
F, W, C, E 0.35 5 1.65 0.08
F, W, H 0.31 4 1.75 0.07
F, G, E 0.31 4 1.89 0.07
G, C, E 0.31 4 1.91 0.07
F, W, C 0.31 4 1.96 0.07
F, G, C, E 0.35 5 1.98 0.07
Ermine SS R 0.36 2 0.00 0.37
W, R 0.39 3 0.29 0.32
F, R 0.36 3 1.80 0.15
R, S 0.36 3 1.83 0.15
MS W, G, R, S 0.52 5 0.00 0.17
C, H, R, S 0.51 5 0.46 0.14
W, G, H, R, S 0.55 6 0.74 0.12
F, W, G, R, S 0.55 6 1.02 0.10
C, H, R, S 0.51 4 1.18 0.10
H, R 0.41 3 1.46 0.08
H, R, S 0.45 4 1.50 0.08
MS G, H, R, S 0.50 5 1.50 0.08
G, C, H, R, S 0.53 6 1.78 0.07
C, R 0.40 3 1.98 0.06

 

Table B3. Summaries of single-scale (SS) and multi-scale (MS) generalized additive models selected through multi-model inference, where all landscape predictors were measured at the same scale in SS models and predictors were each measured at their individual scale of effect in MS models. We include the Nagelkerke R2 (RN2), number of parameter estimates per model (k), change in Akaike Information Criterion (corrected for small samples) from the most supported model (ΔAICc) and Akaike weight (wi) for each model with ΔAICc = 2. LC = local cover type, F = forest, W = wetland, G = grass, C = crop, H = housing, R = road, E = edge, S = stream.

Species Model Included predictors RN2 k ΔAICc wi
Striped skunk SS F, W, G, C, H, E 0.89 7 0.00 0.63
F, W, G, C, H 0.80 6 1.08 0.37
MS F, E 0.51 3 0.00 0.18
F, E, W 0.62 4 0.54 0.14
F, W, G, E 0.71 5 0.58 0.13
LC, F, W, G, E 0.82 9 1.02 0.11
G, C, H, R 0.71 5 1.03 0.11
G, C, H 0.61 4 1.12 0.10
F, W, G, H, R 0.80 6 1.22 0.10
F, H, E 0.60 4 1.76 0.07
F, G, E 0.60 4 1.79 0.07
Ermine SS H, R 0.58 3 0.00 0.64
C, H, R 0.69 4 1.13 0.36
MS W, H, S 0.84 4 --- ---
Raccoon SS C, H 0.35 3 0.00 0.52
LC, C, H 0.49 7 0.20 0.48
MS G, C, H 0.54 4 0.00 0.43
LC, G, C, H 0.64 8 1.51 0.20
G, C, H, E 0.63 5 1.53 0.20
F, E 0.40 3 1.84 0.17

 

Table B4. Summaries of single-scale (SS) and multi-scale (MS) classification and regression tree models, where all landscape predictors were measured at the same scale in SS models and predictors were each measured at their individual scale of effect in MS models. We include the model misclassification risk and proportion of correct classifications for each level of the response variable (i.e. detection, non-detection). LC = local cover type, F = forest, W = wetland, G = grass, C = crop, H = housing, R = road, E = edge, S = stream.

Species Model Included predictors Risk Correct classification
Detection Non-detection
Striped skunk SS LC, F, W, G, R, E 0.02 0.95 1.00
MS LC, F, G, E 0.05 0.84 1.00
Ermine SS LC, W, G, C, R, S 0.00 1.00 1.00
MS LC, F, R, S 0.05 0.75 1.00
Raccoon SS LC, F, W, G, H, R 0.02 0.97 1.00
MS LC, F, G, C, H, E 0.05 0.97 0.92

[Back to A022-122]