A multi-scale accuracy assessment of the MODIS irrigated agriculture data-set (MIrAD) for the state of Nebraska, USA

Figures & data

Figure 1. Study area map showing the spatial distribution of irrigated and nonirrigated croplands of Nebraska as classified in the Nebraska land use map (left; Dappen et al. 2007) and the MIrAD (right; Pervez and Brown 2010).

Figure 2. Validation sites for nonirrigated (a, n = 572) and irrigated (b, n = 1238) validation fields. Bold lines are the USDA NASS Agricultural Statistics District boundaries.

Figure 3. Examples of individual validation sites for (a) irrigated fields in western Nebraska and (b) rainfed fields in eastern Nebraska with the centroid (white), interior (transparent), and edge (hatched) pixels used in the various scales of field-level accuracy assessment.

Figure 4. Example of error clusters omitted because of their narrow linear shapes indicative of transitional, mixed pixels between irrigated and nonirrigated areas.

Figure 5. Examples of error types in the MIrAD overlaid on the 2002 Nebraska Cropland Data Layer. (a) The MIrAD irrigated omission clusters in mixed crop and grassland areas, and (b) the MIrAD irrigated commission clusters in soybeans and corn.

Figure 6. Frequency distribution of ‘core-containing’ and ‘noncore-containing’ ‘disagreement clumps’ by acreage over Nebraska. Numbers relating to ‘noncore-containing clumps’ are in bold.

Table 1. Overall, user’s (UA), and producer’s (PA) accuracy and Kappa results for the field centroids (a), all interior pixels (b), and both interior and edge pixels (c).

	Reference data (CALMIT)
		Irr	Non	Total	UA
(a) Field centroids
Classified data (MIrAD)	Irr	982	60	1042	94.2%
	Non	256	512	768	66.7%
	Total	1238	572	1810
	PA	79.3%	89.5%
	Overall accuracy	82.5%
	Kappa	0.63
(b) Interior pixels
Classified data (MIrAD)	Irr	4801	188	4989	96.2%
	Non	1921	2778	4699	59.1%
	Total	6722	2966	9688
	PA	71.4%	93.7%
	Overall accuracy	78.2%
	Kappa	0.56
(c) Edge pixels
Classified data (MIrAD)	Irr	9815	858	10,673	92.0%
	Non	9047	8213	17,260	47.6%
	Total	18,862	9071	27,933
	PA	52.0%	90.5%
	Overall accuracy	64.5%
	Kappa	0.34

Table 2. Overall, producer’s (PA), and user’s (UA) accuracy and Kappa results by field size.

	Reference data (CALMIT)
		Irr	Non	Total	UA
(a) Field size 480–640 acres
Classified data (MIrAD)	Irr	443	6	449	98.7%
	Non	300	1078	1378	78.2%
	Total	743	1084	1827
	PA	59.6%	99.4%
	Overall accuracy	83.3%
	Kappa	0.63
(b) Field size 320–480 acres
Classified data (MIrAD)	Irr	692	11	703	98.4%
	Non	316	461	777	59.3%
	Total	1008	472	1480
	PA	68.7%	97.7%
	Overall accuracy	77.9%
	Kappa	0.57
(c) Field size 160–320 acres
Classified data (MIrAD)	Irr	1611	68	1679	95.9%
	Non	587	659	1246	52.9%
	Total	2198	727	2925
	PA	73.3%	90.6%
	Overall accuracy	77.6%
	Kappa	0.52
(d) Field size 21–160 acres
Classified data (MIrAD)	Irr	2043	103	2146	95.2%
	Non	714	580	1294	44.8%
	Total	2757	683	3440
	PA	74.1%	84.9%
	Overall accuracy	76.3%
	Kappa	0.44

Dappen, P., J. Merchant, I. Ratcliffe, and C. Robbins. 2007. Delineation of 2005 Land Use Patterns for the State of Nebraska Department of Natural Resources: Final Report. Lincoln: Center for Advanced Land Management Information Technologies, University of Nebraska-Lincoln.

 Google Scholar

Pervez, M. S., and J. F. Brown. 2010. “Mapping Irrigated Lands at 250-m Scale by Merging MODIS Data and National Agricultural Statistics.” Remote Sensing 2 (10): 2388–2412. doi:10.3390/rs2102388.

 Web of Science ®Google Scholar