70 <strong>Ground</strong>-<strong>Water</strong> <strong>Hydrology</strong> <strong>of</strong> <strong>the</strong> <strong>Upper</strong> <strong>Klamath</strong> <strong>Basin</strong>, <strong>Oregon</strong> <strong>and</strong> California This page intentionally left blank.
Appendix A. 71 Appendix A. L<strong>and</strong>sat Analysis Methods The imagery classification was based on two levels <strong>of</strong> resolution: (1) a generalized nonirrigated level that consisted <strong>of</strong> four Anderson Level I/II classes: ice <strong>and</strong> snow, evergreen forest, water, <strong>and</strong> shrub l<strong>and</strong>s that represented a combination <strong>of</strong> bare soil, rock, sagebrush <strong>and</strong> o<strong>the</strong>r nonirrigated vegetation (Anderson <strong>and</strong> o<strong>the</strong>rs, 1976); <strong>and</strong> (2) irrigated l<strong>and</strong>s clustered into five vegetative classes based on <strong>the</strong>ir spectral similarity <strong>and</strong> potential crop water needs. These classes were labeled alfalfa <strong>and</strong> irrigated grasses, small grains, onions-garlic, potatoes-corn, <strong>and</strong> strawberries. Although it would have been desirable, identification <strong>of</strong> individual crop types proved to be impossible because <strong>of</strong> <strong>the</strong> lack <strong>of</strong> unique spectral signatures (Paul Seevers, EROS Data Center, written commun., 2000). The use <strong>of</strong> three satellite images from different times during <strong>the</strong> growing season increased <strong>the</strong> probability that certain croptypes could be distinguished based on <strong>the</strong> development <strong>of</strong> <strong>the</strong>ir spectral signatures. To aid in <strong>the</strong> classification <strong>of</strong> <strong>the</strong> imagery, field work was conducted to map directly crop types in areas totaling about 17,000 acres. A variety <strong>of</strong> representative l<strong>and</strong>cover types were observed, including native trees, pasture <strong>and</strong> most <strong>of</strong> <strong>the</strong> agricultural crops. Most data were collected in mid-July. Data on crop type <strong>and</strong> height, percentage <strong>of</strong> crop cover, sprinkler type, <strong>and</strong> o<strong>the</strong>r parameters also were collected. The three images were processed in succession. The May 21, 2000, scene (from L<strong>and</strong>sat 5) showed full canopy cover <strong>of</strong> <strong>the</strong> perennial crops such as alfalfa, irrigated grasses, <strong>and</strong> any winter wheat that may have been planted <strong>the</strong> previous fall. The timing <strong>of</strong> <strong>the</strong> scene put it before any annual row crops had enough growth to show a vegetative signature. Areas with vigorous growth in <strong>the</strong> May scene served to mask over <strong>the</strong> same areas in <strong>the</strong> August scene. This step reduced <strong>the</strong> amount <strong>of</strong> data in <strong>the</strong> August scene that required analysis. The irrigation district boundaries were used to segregate irrigated areas outside <strong>of</strong> <strong>the</strong> <strong>Klamath</strong> Project. The August 1 scene (from L<strong>and</strong>sat 7) revealed <strong>the</strong> full canopy <strong>of</strong> <strong>the</strong> annual row crops planted during <strong>the</strong> spring. The September 18 scene (also from L<strong>and</strong>sat 7) was used to show any crops that might have a vigorous vegetative signature beyond <strong>the</strong> harvest dates <strong>of</strong> <strong>the</strong> small grains. A review was performed on <strong>the</strong> results <strong>of</strong> <strong>the</strong> final l<strong>and</strong>cover classification by creating an error matrix to evaluate how well <strong>the</strong> classification <strong>of</strong> <strong>the</strong> imagery matched what was actually mapped on <strong>the</strong> ground. The review looked at how accurately <strong>the</strong> classification identified specific crop-types <strong>and</strong> how accurately <strong>the</strong> classification did with respect to all irrigated crop-classes. To do this, <strong>the</strong> new l<strong>and</strong>-cover map created from <strong>the</strong> classified L<strong>and</strong>sat imagery was converted to a polygon dataset, with each polygon being 30 meters, <strong>the</strong> resolution <strong>of</strong> <strong>the</strong> original imagery. Each polygon contained a code for crop class determined by <strong>the</strong> classification process. The ground reference boundaries were <strong>the</strong>n used to clip out <strong>the</strong> same areas in <strong>the</strong> classified l<strong>and</strong> cover. The clipped l<strong>and</strong>cover polygons were <strong>the</strong>n evaluated by crop class against <strong>the</strong> ground reference. For example, if a ground reference area was identified as alfalfa, <strong>the</strong>n <strong>the</strong> same area was compared in <strong>the</strong> l<strong>and</strong> cover. The results <strong>of</strong> <strong>the</strong> review showed that largeacreage crops such as alfalfa had a correct classification ratio <strong>of</strong> about 64 percent. However, if all irrigated crop classes were included <strong>the</strong>n <strong>the</strong> accuracy <strong>of</strong> identifying irrigated l<strong>and</strong>s within that area increased to about 73 percent. For crops grown on smaller fields, <strong>the</strong> accuracy <strong>of</strong> <strong>the</strong> classification to identify specific crop class decreased to a range <strong>of</strong> 20 to 30 percent. If all irrigated l<strong>and</strong>s were included <strong>the</strong>n <strong>the</strong> accuracy increased to <strong>the</strong> 50 to 70 percent range. Sugar beet fields for example, were correctly classified 34 percent <strong>of</strong> <strong>the</strong> time, but if all irrigated l<strong>and</strong>s for <strong>the</strong> same area were included <strong>the</strong>n <strong>the</strong> accuracy increased to 62 percent. All mint <strong>and</strong> strawberry fields were located during <strong>the</strong> ground truth <strong>and</strong> were included in <strong>the</strong> final classification.