Premium
Using digital photographs and object‐based image analysis to estimate percent ground cover in vegetation plots
Author(s) -
Luscier Jason D.,
Thompson William L.,
Wilson John M.,
Gorham Bruce E.,
Dragut Lucian D.
Publication year - 2006
Publication title -
frontiers in ecology and the environment
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 3.918
H-Index - 164
eISSN - 1540-9309
pISSN - 1540-9295
DOI - 10.1890/1540-9295(2006)4[408:udpaoi]2.0.co;2
Subject(s) - forb , vegetation (pathology) , land cover , object based , vegetation cover , ground truth , cover (algebra) , plant cover , shrub , digital image analysis , pixel , remote sensing , statistics , environmental science , cartography , ecology , geography , mathematics , computer science , artificial intelligence , grassland , canopy , land use , computer vision , biology , medicine , mechanical engineering , pathology , engineering
Ground vegetation influences habitat selection and provides critical resources for survival and reproduction of animals. Researchers often employ visual methods to estimate ground cover, but these approaches may be prone to observer bias. We therefore evaluated a method using digital photographs of vegetation to objectively quantify percent ground cover of grasses, forbs, shrubs, litter, and bare ground within 90 plots of 2m 2 . We carried out object‐based image analysis, using a software program called eCognition, to divide photographs into different vegetation classes (based on similarities among neighboring pixels) to estimate percent ground cover for each category. We used the Kappa index of agreement (KIA) to quantify correctly classified, randomly selected segments of all images. Our KIA values indicated strong agreement (> 80%) of all vegetation categories, with an average of 90–96% (SE = 5%) of shrub, litter, forb, and grass segments classified correctly. We also created artificial plots with known percentages of each vegetation category to evaluate the accuracy of software predictions. Observed differences between true cover and eCognition estimates for each category ranged from 1 to 4%. This technique provides a repeatable and reliable way to estimate percent ground cover that allows quantification of classification accuracy.