Laserfiche WebLink
MEMO -TASK 89.1 <br />5.5 Crop classification <br />The crop types to be classified will be determined according to the following criteria: (1) crops ad acreage <br />reported by the Colorado Agricultural Statistics, (2) spectral characteristics of the crops obtained from <br />expert knowledge and research literature, and (3) input from Consumptive Use contractor on categories of <br />crops with similar water requirements. Based on this information, we anticipate a final classification of <br />approximately 9 to 10 crops or crop groups. <br />The crop classification algorithm will be hierarchical, combining multi-temporal NDVI or Tasseled Cap <br />techniques, multi-spectral information from the individual Landsat TM images, GIS data (i.e., slope, <br />proximity to stream), as well as other ancillary data in a knowledge-based classification algorithm. In <br />addition, inputs of water commissioners, water users, ditch companies, and field observation will be used <br />to refine the classification map. <br />5.6 Accuracy assessment <br />Using fifty percent of the "ground truth" or reference data, a quantitative evaluation of the classification <br />will be performed, consisting of an analysis of the overall accuracy, the producer's accuracy, the user's <br />accuracy, and the individual and overall Kappa coefficients (Jensen 1996, Congalton and Green 1999, <br />Lillesand and Kiefer 2000). In delineating polygons for extraction of reference data, a buffer distance will <br />be excluded as recommended in a recent South Platte River basin study (Cipra, 2003). <br />For details on the standard remote sensing accuracy assessment procedures to be applied in this project, <br />please refer to Jensen 1996, Congalton and Green 1999, and Lillesand and Kiefer 2000. <br />6. Comments/Concerns <br />This memorandum describes the most relevant existing studies and the recommended approach for the <br />classification of irrigated crops. However, there are some concerns that will need to be addressed as we <br />develop and apply the definitive classification procedure. The following issues are particularly relevant: <br />• Acquisition of reliable reference data for the entire project area is tenuous. We have identified the <br />FSA as the primary source for reference data. However, FSA has yet to give full access to data <br />from all the county offices in the SPDSS study area. <br />The spectral complexity of the some of the cover types to be classified will generate classification <br />confusion problems. For example, recently cut alfalfa hay fields can appear spectrally similar to <br />fallow, non-irrigated lands. Another example is the spectral similarity of some riparian vegetation <br />to certain irrigated agriculture cover types. There are different approaches to reduce the impact of <br />spectral similarities on the classification and these will be applied on a specific case basis, <br />including, for example use of a ancillary GIS data, masking operations, texture/spatial pattern <br />layers, etc. However, errors will not be completely eliminated. <br />• Although most of the imagery that we expect to use in this study is nearly cloud-free, there are <br />some areas of scattered cloud and areas of excessive haze. One of the advantages of the multi- <br />temporal approach that we will use is the possibility to replace areas obscured by clouds on a <br />particular date with data from another date. Nevertheless, there may be extreme cases where <br />cloud-covered areas may compromise the classification method. <br />• Although the root mean square (RMS) errors obtained in the co-registration were minimal (i.e., <br />below 0.5 pixels or 15 meters), due to the 30-meter spatial resolution of Landsat imagery there <br />will be mixed pixels (pixels that contain information of multiple cover types) along the edges or <br />boundaries of fields. This can create problems when developing training statistics for the <br />classification algorithms and is particularly problematic in the case of long and narrow <br />Page 11 of 13 ~,~tversFde TecAnafagy, lnc. <br />4'J:.fr: !?es'ourccs Frrr~mec.•r.?g ~~nd L'nr+suilrnq <br />