------- ------- a nominal ground pixel resolution of 0.20 m. The data collected aerially also were compared to ground survey data obtained during the summer of 1997 within one week of the aerial survey. We selected 160 reference stations within one of two habitat classes: vegetated and bare substrate, situated below the level of the supra-intertidal salt marsh. At each station a 1 m1 area was surveyed for the presence/absence of each SAV taxa. A detailed description of the ground survey procedure is presented elsewhere (Young etal., 1998). SAV Classification Traditional Photo Interpretation A traditional photo interpretation procedure was utilized to detect SAV presence in both CIR and TC aerial photographs. Specifically, each of the 160 reference stations were viewed electronically at a scale of 1:500 using ArcView (Version 3.0a) by three of the authors (Young, Specht, and Robbins) who independently determined the presence or absence of SAV from the remote images. Algorithm Development and Image Analysis The digital SAV classification of both the CIR and TC orthorectified aerial photographs was conducted using a mosaic of 12 digital images within ERMapper (Version 5.5). The development of an algorithm to classify SAV from CIR utilized a large (-100 ha) intertidal area with known distributions of SAV. Our algorithm, which proved to be highly accurate, had minimal edge effects among the mosaic's orthophotographic boundaries, and is similar to a combination of an Infrared Percentage Vegetation Index (IPVI) and an Atmospherically Resistant Vegetation Index (ARVI) (see a description of thes>e indices in Ray, 1994). The resulting classified image was exported into Arclnfo (Version 7.1.2) Geographic Information System (GIS) and converted into a raster grid at the original scale of 0.2 m pixel size. The grid then was converted into a vector polygon coverage and, using an overlay technique in Arclnfo, was compared to a point coverage of ground survey data. In the overlay process, a polygon classified as SAV was considered to be classified correctly if it included, or was within 1.28 m of, a ground station location representing SAV. The 1.28 m distance function represents the combined spatial accuracy of the point and polygon coverages (RMSE of 0.78 m) plus the nominal distance from the center to the edge of a 1 m2ground station area (0.5 m). The application of our algorithm to TC digital orthophotography was less successful, as were our attempts to develop a new algorithm. We didn't succeed because the spectral characteristics of macroalgae in the visual range are shared with those of open water, and because the spectral characteristics of seagrasses in the visual range are shared with those of bare substrate, particularly sandy sediment. Data Analysis To compare the results of our several classifications of SAV versus bare substrate in the study area (Figure 1), we used the error matrix approach described by Congalton (1991). In this approach, the percentage of stations with a given classification obtained via the remote sensing approach that agree with the reference (ground survey) results is termed Users Accuracy. Alternatively, the percentage of stations with a given classification obtained from the ground survey that are correctly classified by the remote sensing is termed Producers Accuracy. A third commonly used category, termed Overall Accuracy, also is included. ------- Results and Discussion Our results did not yield a significant difference (p > 0.05; two-tail Student's t-test) between the CIR and TC Overall Accuracy values (Table 1). Thus the use of either CIR or TC for the detection and quantification of SAV and bare substrate can be recommended. However, equally clear is the qualitative indication that the CIR approach is more useful when distributions of exposed SAV within intertidal flats are digitally classified. Specifically, a visual comparison of orthorectified aerial photographs of the study area using both CIR and TC film suggests that the digital classification of the TC film over-represents SAV distributions (Figure 2). Thus, it appears that the algorithm we developed for the classification of the CIR images was inadequate when applied to the TC images. We therefore conclude that our digital classification obtained for Yaquina Bay using the CIR film is preferable to that obtained from the TC film. Acknowledgments The ground reference survey was conducted by employees of DynCorp/TAI, Inc. stationed at Newport, OR and EPA scientists and staff of the Western Ecology Division's Coastal Ecology Branch. The aerial photography was conducted by Bergman Photographic, Inc. (Portland, OR); the orthorectified digital photographs were produced by Photogrammetric Digital Services, Inc. (Eugene, OR). Mention of trade names or commercial products does not constitute endorsement or recommendation for use. This paper has been subjected to EPA's peer review and administrative review process and has been approved for publication as an EPA document. References Congalton, R.G., (1991) A review of assessing the accuracy of classifications of remotely sensed data. Remote Sensing of the Environment. 37:35-46. Dobson, I.E., Bright, E.A., Ferguson, R.L;, Field, D.W., Wood, L.L., Haddad, K.D., Iredale HI, H., Jensen, J.R., Kelmas, R.J., Orth, R.J., and Thomas, J.P. (1995) NOAA Coastal Change Analysis Program (C-CAP): Guidance for Regional Implementation. NOAA Technical Report NMFS 123. U.S. Department of Commerce. Ray, T.W. (1998) Vegetation in Remote Sensing FAQs (Version 1.0). ERMapper Applications, pp. 99-124, San Diego, CA. Young, D.R, Specht, D.T., Clinton, P.J., and Lee II, H. (1998) Use of color infrared aerial photography to map distributions of eelgrass and green macroalgae in a non-urbanized estuary of the Pacific Northwest U.S.A. Proceedings of the Fifth International Conference on Remote Sensing for Marine and Coastal Environments, Vol. II, pp. 37-45, ERIM International, Inc., Ann Arbor, MI. ------- Table 1: Accuracy assessment error matrices (%) of SAV versus bare substrate classification by traditional photointerpretation and digital classification (Congalton, 1991). Accuracy Users -CIR Producers - CIR Overall - CIR Users -TC Producers - TC Overall -TC Mean Overall -CIR Mean Overall -TC Young SAV 89.7 78.9 Bare 68.2 83.3 80.5 82.8 90.2 76.8 63.2 81.1 Specht SAV 85.2 82.0 Bare 68.8 73.6. 79.0 84.1 87.2 73.0 67.7 80.6 Robbins SAV 87.5 84.2 Bare 72.7 77.8 81.9 78.3 84.2 63.8 54.4 74.1 80.5 78.6 Digital SAV 92.4 72.9 Bare 64.0 88.9 78.5 77.2 71.4 513 58.8 67.2 ------- Central Yaquina Estuary Figure 1: Site map showing location of Yaquina Bay on the west coast of Oregon. ------- c. s/ > D. k. '• » V ' •;./ figure 2: A comparison of orthorectified aerial photographs using CIR (A) and TC (B) and their issociated digital classifications (C and D). ------- ------- |