Northern Prairie Wildlife Research Center
Using GPS data to organize video images by transect segment provided a rapid method of preparing data for analysis. Without GPS data, video would have to be manually inspected to determine location of transect segments and this would increase labor required to process images. The optical disk with the ability to store thousands of video images and random access recall was an efficient means to store and retrieve video images for this survey.
An improvement of the aerial video system would be to estimate the scale of each video image. Scale varies among video images due to changes in ground and aircraft elevation along the transect. The variable image scale requires verification of transect boundary locations. The time required to verify the transect boundary locations (an average of 20 min for an air-ground segment) could be reduced or eliminated by using a laser ranging device to estimate distance from the camera to the ground (Ritchie et al. 1992). Distance could be recorded on the GPS data logger and used in the Prepare Video Disk software to improve the estimate of video image scale and transect boundary locations.
Use of aircraft attitude systems to estimate yaw, pitch, and roll or a gyro-mounted camera would enable more accurate georegistration of imagery (Bobbe 1992; Trimble Navig. Ltd. 1993). This enhancement would be useful if digital National Wetlands Inventory maps were available for overlay of video images. Omission and commission error in the video interpretation could be reduced by analysis of video images in conjunction with digital National Wetlands Inventory data.
We selected the near infrared spectral region on the basis of its successful use in estimating the surface area of water bodies (Salomonson et al. 1983). Accuracy of pond detection would increase if a solid state video camera with a shortwave (1.55-1.75 Ám) spectral band was used because of increased contrast between water and terrestrial features in these wavelengths. However, measuring the reflected radiance in these wavelengths requires costly modifications of solid state video cameras (Neville and Powel 1992). There are tube cameras with shortwave infrared capability (Everitt et al. 1986), however these cameras are sensitive to radiance overload, which could occur from the specular reflectance from a water surface. A small increase in pond detection accuracy would occur using a multispectral video camera measuring reflected radiance in the visible and near infrared wavelengths. A multispectral video camera would also enable estimation of additional wetland attributes under certain conditions (Lyon et al. 1992) and a more detailed classification of upland cover and land use than is possible using only near infrared wavelengths (King and Vlcek 1990).
Image thresholding and manual interpretation combine strengths of computer analysis of image tone and human perception and reasoning. Variability in reflectance due to wetland characteristics including water turbidity and depth, physical substrate, aquatic and emergent vegetation, adjacent land cover, and imaging geometry make developing a fully automated system for detecting ponds difficult. The time required to analyze the number of ponds on a 0.2- x 29-km transect segment ranged from 9 to 59 minutes. Time required for interpretation varies with abundance and size of ponds and their contrast with the background.