Laserfiche WebLink
SPDSS Memorandum -Final <br />To: Ray Alvarado, Ray Bennett <br />From: Riverside Technology, inc.: Nils Babel, Tim Martin <br />Subject: SPDSS Spatial System Integration Component Subtask 89.10 -Mapping of <br />Water Features <br />Date: 5 October 2005 <br />1. Introduction <br />This memorandum summarizes activities performed under Task 89 -Mapping of Current Land Use and <br />Irrigated Field Boundaries, Subtask 89.10 -Mapping of Water Features. These activities are described in <br />RTi's Scope of Work for SPDSS Phase 3 Spatial System Integration Component. <br />The deliverables for Subtask 89.10 are digital coverages of water bodies for the entire SPDSS study area <br />and this technical memorandum describing the procedures and results. <br />2. Approach <br />The maximum extent of water bodies three acres and greater in the South Platte study area were mapped <br />using the multidate 2001 Landsat satellite imagery. Water bodies were mapped in three major steps: <br />Step 1 -Hybrid classification for each individual Landsat scene <br />Step 2 -Multi-date water body analysis and refinement <br />Step 3 -Review of water bodies, manual editing, and QAQC <br />In the first step each Landsat scene was classified with a hybrid classification approach. Using ERDAS <br />Imagine the scene was first input into an unsupervised (Isodata) classification to create a thematic image <br />with 20-30 clusters or groups of pixels with similar spectral characteristics (depending on the scene). The <br />clusters containing water bodies were then identified and labeled accordingly. Any clusters containing a <br />mixture of water bodies and other cover types with signatures similar to water (e.g. cloud shadows or <br />mountain shadows) were further disaggregated using a technique called `cluster busting'. Cluster busting <br />splits one or more clusters containing similar cover types into several clusters for further analysis. The <br />resulting clusters were evaluated and pixels corresponding to water bodies were identified and labeled. <br />This procedure was iterative and continued until the maximum separation of mixed pixels was achieved. <br />Once homogenous water body classes were identified, a final signature file was created by combining the <br />signature file from the cluster busting procedure and the initial unsupervised classification. The <br />combined signature file was used to classify the Landsat scene a second time using a supervised <br />(Maximum Likelihood Classifier) classifier. The result for each scene was a single raster file with three <br />or four classes of water bodies and all other cover types in a single, separate class. <br />In the second step the multi-date water body classifications for each Landsat frame were overlaid to <br />determine the maximum extent of water bodies. In this step additional data also were used to refine the <br />water body classification. For example, rivers were removed, existing lake data were used to aid in <br />isolating water bodies, and slope data were used to remove mountain shadows. After the overlay and <br />refinement of the water body mapping the different satellite frames were merged and the raster data was <br />converted to vector data. The water bodies were then filtered to remove water bodies less than three acres <br />in size. Water bodies were then named using the `Lake and Reservoir' dataset available from the <br />Colorado DWR (lak3co.shp). The water bodies from our analysis were joined with the centroids of the <br />DWR data and names in the DWR `Lake and Reservoir' dataset were then transferred to the water bodies. <br />Page 1 of 3 ~fxlversfde Technology, fnc. <br />V.Lrrc- Aesourc.~s Frr4r.~ec.•mq .m~Y Corzsuxrnq <br />