Laserfiche WebLink
MEMO -TASK 89.1 <br />classification method, which automated the interpretation process resulting in a less expensive and <br />timelier product. The resulting average classification accuracy for the likelihood of corn classification was <br />89 percent. <br />In a study conducted in a semiarid zone in La Mancha, Spain, Martinez-Beltran and Calera-Belmonte <br />(2001) developed a methodology to estimate the area of irrigated crops with Landsat TM imagery based <br />on the different crop spectral responses through time according to their phenological evolution. The <br />multi-temporal classification included the development of Normalized Difference Vegetation Index <br />(NDVI) (Jackson and Huete, 1991) of Landsat TM imagery acquired on four different dates during the <br />crop season as well as the combination of maximum likelihood, decision tree criteria and context <br />classifiers algorithms. The following four land cover classes were mapped: spring irrigated crops, summer <br />irrigated crops, alfalfa, and not irrigated areas. The resulting overall classification accuracies for irrigated <br />areas were 93.1 percent in 1996 and 90.2 in 1997. On a larger scale, Qi et al. (2002) developed a <br />methodology to map irrigated lands on a 174,000-square-mile area in parts of Colorado, Kansas, <br />Nebraska, New Mexico, Oklahoma, South Dakota, Texas, and Wyoming. Landsat scenes were acquired <br />in summer months (leaf on) and in winter months (leaf off). Using a ratio between the Landsat TM bands <br />4 and 3, a brightness theme was created that enhanced the vegetation signature. Subsequently a brightness <br />value was selected as a threshold to segregate irrigated and non-irrigated land. The reported overall <br />accuracy for the irrigated lands vs. non-irrigated lands classification ranged between 77.5 and 79.8 <br />percent. <br />Cipra et al. (2002) developed a center pivot map for the entire State of Colorado using 15-meter <br />resolution satellite imagery obtained from Digital Globe between July and November 1999. The source of <br />satellite data is not specified, but presumably is Landsat TM panchromatic. Center pivots that were <br />visible on the satellite image were digitized on-screen and saved as apivot-boundary GIS file. The <br />accuracy of the method, evaluated by a field crew in selected areas of the state, was reported to be within <br />1.5%. However, it is not clear what this value represents, or how it was obtained. <br />3. Pilot study for SPDSS <br />The use of carefully selected multi-date satellite imagery has been shown to improve the classification <br />accuracy over single date classification of agricultural crops (Maxwell and Hoffer 1996a). In selecting the <br />multi-date imagery it is necessary to consider regional crop phenology (Maxwell and Hoffer 1996b) as <br />well as cloud cover limitations. For evaluating satellite image needs for this SPDSS project that covers six <br />Landsat image frames, seven Landsat images representing different times for one frame (32/32) were <br />obtained for the 2001 growing season (Table 2). <br />Table 2: Landsat Images Used for a Pilot Area to Assess the Image Requirements for the Entire <br />SPDSS Study Area <br />FRAME DATE CLOUD REMARKS SENSOR <br />(Path_Row) (2001) COVER <br />33_32 Apr26 2 Wedge has cloud cover and patches of haze. ETM7 <br /> May12 0 Good image. 2 small patches of haze. ETM7 <br /> Jun5 0 Good image. Very little scattered cloud cover. TM5 <br /> Jul? 0 Good image. Very little haze in SW corner. TM5 <br /> Aug24 0 Good image. Very little scattered cloud cover, TM5 <br /> mostly over mountains. <br /> Sep25 0 Good image. 3 very small patches of TM5 <br /> scattered cloud cover. <br /> Novo 0 Good image. ETM7 <br />Page 4 of 13 ~-~~versfde TecAnafagy, lnc. <br />W:rler Resources Frr~mec.•rng ,end L'nr+surlrn2 <br />