investigator. To develop a meaningful classificalion of the data, the investigator must work with a computer at several stages, particularly to identify training sites and determine which of the pixels in the casi image are most similar to those in the training sites. The classification process is therefore not completely automated and is in some degree as subjective as air photo interpretation.
ii. Subtidal
Shallow water habitats, below the limit of lowest tide, can only be mapped with remote technology. There are some exceptions to this generalization. SCUBA can be used to make direct underwater observations and the diver's position can be plotted using DGPS in a surface vessel. This method was recently used to develop a habitat map for Vancouver Harbour. Kelp and seagrass habitats occur to the limit of light penetration into coastal waters, which in turn depends on local conditions such as turbidity. If the water is very clear it is possible to use aerial photos and casi for sensing below the water's surface. For very large kelp (e.g. Macrocystis pyrifera) where the blades float on the surface, aerial photos and multispectral imagery can also be used. The former method has been developed for mapping biomass of kelp on the west coast of Vancouver Island by Foreman (1975).
A variety of acoustic methods have been used for mapping subtidal habitats. In an early application, a 200 kHz echosounder was used to map seagrass beds in San Francisco Bay (Echeverria and Rutten 1989). Water depth was shallow enough so that ground-truthing could be done from a helicopter and boat. In our region, depth and habitat type as defined by sediment type have been mapped using Acoustic Substrate Classification (ASC) ("RoxAnn" or "Quester Tangent" technologies), which also provides accurate positioning using DGPS (Curran 1995). Cripps (1996) reported that subtidal clam habitat, which is characterized by a particular combination of sand and mud, was successfully mapped with ASC. He also reported that eelgrass beds could be mapped with this technology. Bornhold et al (1996) found that sidescan sonar, another example of ASC, was an effective technique to investigate subtidal habitats off the Oregon coast. As with the other remote sensing techniques described above, ASC methods require considerable ground-truthing, using SCUBA, submersible, video or still camera, ROV, or core sampling.
E. Ecological Classifications
A basic principle in an ecological accounting system is the use of a standardized "currency" for keeping track of habitats. Alternatively if a standard currency is not available, data should be collected in comparable units that when aggregated will allow comparisons through time in monitoring programs. On the northeast Pacific coast, it is very difficult to compare habitat data because there is no agreed-upon ecological classification system in use. In the Strait of Georgia-Puget Sound inland sea, for example, there are at least 8 systems in place (Levings and Thom 1994). In order to overcome this problem, and match with the capability of remote sensing systems, we have decided to use a relatively simple system that relies on relatively easily recognizable vegetation units (e.g. eelgrass beds) and sediment features (e.g. sand flats) as habitat units.
F. Ground-truthing Surveys to Confirm Identification of Habitat and Estimates of Area
The effort expended on ground-truthing before, during, or after remote sensing surveys is obviously related to the complexity and extent of the study area. Site specific forestry surveys (scale 1: 5000 to 1:20000) require that between 76 and 100 % of the polygons from air photo interpretation are checked by ground-truthing (Resources Inventory Committee 1995). We could not find similar criteria to use for coastal work. However air photo interpreters can provide guidance for field verification by noting areas where they had difficulties (e.g. sun glare or shadow).
Because the extent of vegetation can change rapidly owing to seasonal changes, the ground truthing should be done at the same time as the remote sensing, if at all possible.
As noted above, habitat managers in our agency need to account for habitats > 100 m2 in area, which is quite similar to the 300 m2 minimum specified in the protocol for submerged aquatic vegetation mapping given by Dobson et al (1995). However Durance (1996) found that the minimum polygon size for seagrass observable in aerial photos from Saanich Inlet was 1000 ? . In addition to the obvious matters such
