Table 2 Summary and evaluation of decile distribution models.

From: Stacked distribution models predict climate-driven loss of variation in leaf phenology at continental scales

Modeled trait decile

Nocc

Predicted suitable

Model evaluation

Pixel

% area

AUCtest ± sd

AUCtrain

Omission rate

D1

39

16,840

10.7

0.797 ± 0.090

0.950

0.077

D2

46

47,386

30.2

0.875 ± 0.056

0.933

0.043

D3

41

52,064

33.2

0.845 ± 0.071

0.913

0.098

D4

39

41,900

27.2

0.884 ± 0.049

0.920

0.077

D5

38

49,687

31.7

0.812 ± 0.067

0.896

0.079

D6

39

32,359

20.6

0.848 ± 0.050

0.911

0.077

D7

39

40,574

25.9

0.831 ± 0.050

0.900

0.077

D8

44

29,752

19.0

0.886 ± 0.042

0.940

0.091

D9

41

32,326

20.6

0.856 ± 0.052

0.902

0.098

D10

34

17,959

11.5

0.879 ± 0.049

0.943

0.088

Stacked Models: 400

53,727

34.3

 
  1. Predicted suitable climatic conditions on the landscape are listed both as pixel number and as percent area of the total extent of the landscape (total pixels = 156,659). Each pixel has an area of ~1 km2, a resolution of 30 arc seconds. Test AUC (area under the receiving operating characteristic curve) +/−standard deviations are given from 5-fold cross-validation. Training AUC is for final models. Training omission rate is for the 10-percentile training presence threshold rule.
Back to article page