Submitted to: Weed Science Society of America Meeting Abstracts
Publication Type: Proceedings
Publication Acceptance Date: 10/6/2007
Publication Date: 2/1/2008
Citation: Altland, J.E. 2008. Comparing digital software to human observation for estimating weed cover in nursery containers. Weed Science Society of America Meeting Abstracts. Interpretive Summary:
Technical Abstract: Researchers who study weed management in nursery crops often rely on visual ratings to assess weed growth in response to some treatment effect. Visual weed ratings are easy to perform, non-destructive, and do not require any special equipment. However, visual ratings are prone to bias and skewed judgment. Digital photography has become commonplace in the agricultural sciences over the past 10 years. Digital cameras with high resolution are inexpensive and commonly used by weed scientist. A program called Assess, Image Analysis for Plant Disease Quantification (American Phytopathological Sociey Press, St. Paul, MN), has been developed for the primary purpose of quantifying the size of lesions on plant leafs. However, the software can be used to measure the absolute area or percent area of any object against the background of another object with different color characteristics. The objective of this research was to determine the ability of Assess software to accurately measure weed cover in nursery containers from a digital camera image. Green disks of various diameters were placed on the surface of 20 different containers to emulate weed cover. The number and diameter of the green disks were recorded so that the exact surface area covered by the disks was known. Percent area covered by the disks ranged from 0 to 70%. A digital image was taken using a Nikon Coolpix 8800 digital camera with eight megapixel resolution. The digital image was analyzed with Assess software, while four technicians estimated coverage visually. In a second experiment, 40 containers with varying amounts of bittercress (Cardamine oligosperma) were arranged randomly. Bittercress coverage of the container surface ranged from 0 to 75%. Percent weed cover in each container was determined with the Assess software, and estimated visually by four technicians. Computer analysis with Assess software was nearly perfect in estimating coverage of the green disks (y = 1.08x + 1.3, r2 = 0.998). Estimation by the four technicians was also highly correlated to the known area covered, with r2 ranging from 0.91 to 0.96. However, each of the technicians tended to overestimate actual coverage. Estimated y-intercepts and slopes of the lines of best fit between known area covered and technician estimates were all greater than 0 and 1, respectively. Computer estimation of bittercress cover was highly correlated to each of four technician evaluations (r2 ranged from 0.73 to 0.88). Based on these results, the Assess software has potential to be used as a rapid method to quantitatively analyze weed cover in nursery containers. The software is best suited for weed species that grow prostrate or decumbent. Species that grow upright would be more difficult to analyze considering the 2-dimensional nature of the software.