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.

Landcover Class	water	snow_ice	barren	emergent wetland	woody wetland	mesic grass forb	xeric grass forb	mesic shrub	xeric shrub	deciduous forest	coniferous forest	pasture	mesic rowcrops	orchard	xeric rowcrops	fallow	developed high intensity	developed medium intensity	user's accuracy
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%