Image analysis and modeling of macro algae densities in intertidal areas of Narragansett Bay

Andrew J Bird, University of Rhode Island


Prior to 2011 analysis of images collected from aerial surveys conducted in the Narragansett Bay area in Rhode Island have been analyzed by human for macro algal percentage cover. This process of visual inspection for percentage cover analysis was found to be very subjective in nature highlighting a humans tendency to over estimate percentage cover. A tool was needed to remove subjectivity in percentage cover analysis that caused poor macro algal estimates, while maintaining the positive benefits of human interaction. ^ In order to assess the percentage cover of both red and green algae seen in an image a Java based image processing plugin was created using the ImageJ library as a foundation. The tool performs a targeted color segmentation from a user defined mask, with the mask specifying non-algae and probably algae observed. The mask is created as part of a two step process that initially performs a local histogram operation (Contrast Limited Adaptive Histogram Equalization) that is aimed at removing some of the illuminations effects seen in an image. In an experiment conducted using mock images to assess the accuracy of users percent cover estimation the results yielded a maximum variance between observed and expected percentage cover of 21\%. ^ The testing conducted on the image processing tool using created checkerboard test patterns showed that the image processing tool was consistently within 2\% of the actual observed percentage cover. The results from the comparison between visual inspection and image processing tool results showed a high correlation in green and red percent coverage. The mean difference between both methods for green algae was identified at 2\%, and for red was identified at 2.6\%. Images that were found to be outside the 21\% bias offset used were found to be erroneous as a potential result of failure in the image processing work flow, examples of these failures were reexamined and found to be closer to the visual inspection results. In another test conducted with actual aerial images processed by users, the percentage variance between the visual inspection method and image processing tool were compared. The image processing tool yielded a 5 percent reduction in variance when compared with the simple visual inspection method. The percentage cover results from the image processing tool were converted to an area estimation using a simple linear trigonometric function that utilized the horizon line for pitch estimation. The area estimation results were compared with macro algal growth model runs performed on several areas in or around the Greenwich Bay area within Narragansett Bay. The growth model results showed potential growth of macroalgae and were compared visually to the results from the image processing tool for the months May, June and July for the year 2010. The results from both the growth model and image processing tool appear to correlate greatly when compared. ^ The project revealed a number of issues with the previous and current methods for assessing the percentage cover, and utilizing this percentage cover to create an area estimation. Future developments would see the inclusion of additional sensors to capture roll and pitch of the camera. This coupled with a non-linear area estimation approach would improve the accuracy of the process greatly. Also at present there is no means to directly compare the results from the image processing tool to the growth model output, this could be achieved by converting area to dry weight.^

Subject Area

Biology, Oceanography|Engineering, Marine and Ocean

Recommended Citation

Andrew J Bird, "Image analysis and modeling of macro algae densities in intertidal areas of Narragansett Bay" (2011). Dissertations and Master's Theses (Campus Access). Paper AAI1502430.