Classification of Digital Photography for Measuring Productive Ground Cover
Abstract
Productive ground cover (PGC) is often used as a measure of sward health and persistence. To measure PGC, a camera stand was constructed to provide diffuse lighting of grass swards for color digital photography; the photographs were classified into productive and nonproductive cover using Mahalanobis distance. The PGC measurement techniques were tested on a grazing experiment that used four forage types: Lakota prairie grass (Bromus catharticus Vahl.), Kentucky 31 endophyte (Neotyphodium coenophialum)-free tall fescue (Lolium arundinaceum [Schreb.] S. J. Darbyshire), Kentucky 31 endophyte- infected tall fescue, and Quantum (novel-endophyte) tall fescue. The accuracy of the PGC maps was assessed using a stratified subsample of 48 images, 12 from each of four productive cover classes (0%–39%, 40%–59%, 60%–79%, and 80%–100%). On each of these 48 images 100 random points were labeled by a single skilled interpreter. The PGC percentages thus derived had an 83.7% agreement with the PGC maps. However, the percentages derived from the PGC maps were not well correlated with the PGC percentages derived from either ocular estimation (r 5 0.22) or a simple digital point quadrat method (r 5 0.47). This experiment highlights the potential for semiautomated classification of ground-based digital photographs for estimating PGC, though further research (including more direct comparison with established field techniques) is warranted.