Model Validation
To validate the dynamic landcover model, we iteratively withheld 1/10th of the training data in a k-fold crossvalidation exercise on 50,000 training locations distributed across the Cascadia region. Each fold was spatially independent, so the withheld fold did not overlap in space with the training folds. The table below shows the producers, users, and overall accuracy averaged across the 10 folds in the crossvalidation.
water
2987
0
1
11
0
0
0
0
0
0
0
0
0
0
0
0
1
0
99.57%
snow_ice
0
974
24
0
0
0
0
1
0
1
0
0
0
0
0
0
0
0
97.40%
barren
3
18
2750
2
1
0
9
0
51
1
2
0
4
1
1
12
115
30
91.67%
emergent wetland
27
0
10
1960
168
46
25
42
15
12
5
92
17
36
4
5
15
21
78.40%
woody wetland
0
0
7
148
1783
27
0
74
2
161
268
6
3
10
0
0
2
9
71.32%
mesic grass forb
0
0
0
43
28
2799
0
953
1
31
82
24
9
3
7
2
0
18
69.98%
xeric grass forb
0
0
2
0
0
11
2096
0
354
0
0
0
1
0
6
28
0
2
83.84%
mesic shrub
0
0
0
36
28
1036
0
2478
1
113
267
14
3
10
1
0
0
13
61.95%
xeric shrub
0
0
62
1
0
4
275
2
2142
0
0
0
0
1
0
6
0
7
85.68%
deciduous forest
0
0
0
8
71
17
0
21
0
3867
16
0
0
0
0
0
0
0
96.68%
coniferous forest
0
0
9
9
207
76
1
345
6
120
4210
1
0
1
0
0
0
15
84.20%
pasture
0
0
0
119
5
77
9
14
2
0
2
1180
32
23
1
4
2
30
78.67%
mesic row crops
0
0
2
16
3
29
13
3
2
0
1
35
1594
121
121
50
0
10
79.70%
orchard
0
0
6
22
1
8
12
12
1
0
1
22
158
1199
0
30
0
28
79.93%
xeric row crops
0
0
11
13
1
29
24
8
7
1
1
4
165
8
827
394
0
7
55.13%
fallow
0
0
14
13
0
60
99
5
45
1
0
31
112
56
303
741
1
19
49.40%
developed high intensity
0
0
45
5
1
1
1
0
0
0
0
5
6
4
0
1
3458
473
86.45%
developed medium intensity
0
0
55
13
8
9
2
8
6
0
5
33
2
28
1
6
441
3383
84.58%
producer's accuracy
99.01%
98.19%
91.73%
81.03%
77.35%
66.19%
81.68%
62.48%
81.29%
89.76%
86.63%
81.55%
75.69%
79.88%
65.02%
57.94%
85.70%
83.22%
overall accuracy
80.86%
Last updated