Laserfiche WebLink
MEMO 93 <br />six historic period spectral signatures as corn and sugar beets. The remaining two signatures were labeled <br />as unknown. <br />This graphical comparison was repeated for each of the 50 spectral signatures from the unsupervised <br />classification. <br />Spectral signatures were analyzed to separate the known from unknown for each historic period. The <br />following shows how unknown spectral signatures for each historic period were further investigated in an <br />attempt to understand the crop type that the signature represented: <br />(1) The location of unknown classes within fields was analyzed in relation to known crop types <br />for each historic period. <br />(2) The presence or absence of clouds, scan line issues, and saturation within a single or multiple <br />images of a frame for an unknown historic period spectral class was analyzed. This step was <br />especially useful when classifying 1976 irrigated lands because Landsat MSS was affected <br />sporadically by scan line issues (Table 1). In addition, the native 4-bit radiometric resolution of <br />Landsat MSS was a consideration when comparing spectral signatures from 1976 to signatures <br />2001. <br />(3) Classification results between overlapping areas of adjacent Landsat frames was compared (if <br />overlap existed). <br />(4) The irrigated acreage of known and unknown crop type classes within a county was compared <br />to agricultural statistics. Temporal trends and relative proportions of each crop type were <br />compared between the agricultural statistics and classified irrigated acreages. <br />If this information did not clarify the identity of the unknown signature, it remained labeled as unknown <br />and was changed during post-classification revisions as described below. Once all classes of the <br />unsupervised classification were identified using these procedures, the thematic classified NDVI raster <br />was used as "training data" for the MLC classification. The irrigated lands mask described in Section <br />4.1.1 of this memo was applied to amulti-temporal Landsat composite (layer stack of all Landsat dates) to <br />create amulti-temporal composite of irrigated areas only. This irrigated lands Landsat composite was the <br />base layer for the MLC classification. The resultant preliminary irrigated crop type classification was <br />then attributed to the irrigated parcel using zonal statistics to identify the majority crop type within a <br />parcel. <br />4.2.1.2 Post Classification Refinement and Review <br />In the third step, post classification refinement was performed to examine individual crop types of <br />irrigated parcels for errors. These classification refinement rules were based upon the zonal statistics <br />tabulation of area of crops per parcel and majority fraction, as well as a measure of homogeneity per <br />parcel. Additional rules applied included the use of parcel types and shapes (e.g., center pivots). <br />The fourth and final step of the crop type classification process consisted of a QA/QC review of per crop <br />irrigated acreage totals. The final classification review step was used to verify that high-level agricultural <br />trends matched with expectations based upon reference data sources. The tabular and graphical <br />comparisons displayed in Table 2 and Figure 9 assisted with the process of evaluating irrigated acreage <br />classification results versus agricultural statistics trends. <br />Page 14 of 59 ,Rereradde ~eshn~P~gy, ~os~ <br />