SPDSS_Task89-2_CropLandUseClassificationProcedures_20060929

• Title: SPDSS Task 89.2 - Crop and Land Use Classification Procedures for Year 2001
• Description: This memorandum describes the activities conducted under Task 89: ‘Mapping of Irrigated Land Use and Irrigated Parcel Boundaries’ for year 2001 and complements the SPDSS Memoranda for Task 89.1 and Task 90.2. This memorandum also provides details on the methods used to conduct a number of Task 89 activities, including Task 89.3: Determine Irrigated Vs. Non-irrigated Lands, Task 89.4: Identify Crop Types In Each Polygon, Task 89.5: Review, Revision and Final Classification, and Task 89.6: Conduct Accuracy Assessment, as well as the results obtained from these activities for year 2001.
• Decision Support - Doc Type: Task Memorandum
• Date: 9/29/2006
• DSS Category: GIS
• DSS: South Platte
• Basin: South Platte
• Contract/PO #: C153960
• Grant Type: Non-Reimbursable
• Bill Number: SB01-157, HB02-1152, SB03-110, HB04-1221, SB05-084, HB06-1313, SB07-122
• Prepared By: Riverside Technology inc.

MEMO 89.2
correctly identifying the mislabeled training data while not discarding training data that are accurate but
difficult to classify. This identification of mislabeled reference data was conducted though the spectral
analysis described in Section 2.5.
2.7.1 Irrigated Lands Accuracy Assessment
The accuracy assessment for the irrigated lands classification used data collected from FSA, and
complemented with information obtained through photo interpretation of DOQQs and NDVI composites.
According to Congalton and Green (1999), the collection of reference data by photointerpretation is
acceptable for simple classification schemes, as long as a subset of the photointerpreted data is verified in
the field. The objective of the photointerpretation was to obtain information on cover types not available
through FSA. Therefore, non-irrigated cover types riparian trees, rangeland, and other general classes of
native vegetation were obtained through photointerpretation across the entire SPDSS area. In conjunction
with photointerpretation and image analysis, RTi personnel conducted an extensive field verification
campaign in 2004. Special attention was given to inconclusive areas, such as marginally irrigated lands
for improving discrimination between irrigated and non-irrigated lands. In performing this activity, the
three-year time difference between the verification and the year 2001, was taken into account.
Nevertheless, valuable information was obtained due to the fact that some of these areas displayed little
change in such a short period of time.
Figure IS shows the process for the irrigated lands accuracy assessment. After signature analysis and
error minimization procedures, 456 non-irrigated parcels were determined as adequate for the accuracy
assessment. Therefore, since irrigated parcels almost doubled in number the non-irrigated parcels, 456
irrigated parcels were selected from a total of 800 using a randomization algorithm to minimize bias.
Reference parcels
ii
~~
Irrigated Lands
Classification
Overal I,
Producer', and
User' Accuracies,
Kappa Statistics
Error Matrix
Figure 15. Irrigated Lands Accuracy Assessment
These test parcels were compared to several iterations of the irrigated lands classification (Figure 1 ~
until optimal results were obtained. The accuracy assessments included the overall accuracy, the
producer's accuracy, the user's accuracy, and the individual and overall Kappa coefficients. For more
details on the standard remote sensing accuracy assessment procedures applied in this project, please refer
to Jensen 1996, Congalton and Green 1999, and Lillesand and Kiefer 2000.
Page 23 of 45 ~R~versfde FecAnotogy, fnc.
4'JaYer Resources Errgi~ecr:np an~i CansuFlrnp