Classification method of mixed pixels does not affect canopy metrics from digital images of forest overstorey

Abstract

A great drawback of photographic methods for estimating canopy metrics such as leaf area index ( L) and cover has been the tedious and time consuming image processing step and the perceived sensitivity of the results to image processing. This paper describes an automatic method, the ‘two-corner method’, for detecting homogeneous regions of canopy and sky, and for quantifying the number of mixed pixels in canopy images. Mixed pixels are pixels of intermediate brightness value that do not very obviously belong to either sky or canopy. Four image classification methods were tested for classifying mixed pixels as canopy or sky. When applied to both fisheye and cover images of Eucalyptus forest, none of the more complicated classification methods yielded results that greatly differed from a simple global binary threshold classification, even if those metrics were derived from the zenithal distribution of gap fraction or gap size. Increasing photographic exposure by one stop reduced calculated L by 9–12%, but modern digital camera technology makes it much easier to correctly expose fisheye canopy images, either by examining the image histogram in the field or by taking multiple exposures and choosing the best exposures after automatic processing. This study is the first to systematically quantify the number of mixed pixels in canopy images and demonstrated that fisheye images contain more mixed pixels than cover images, and that the number of mixed pixels increases with increasing vegetation cover. In conclusion, the recent advances in digital camera technology, combined with robust and automated image analysis methods, are rapidly bringing the field of photographic analysis of canopy structure to maturity, where the field techniques and image processing aspects of the methodology are no longer significant factors limiting its application by non-experts. In the case of fisheye photography, research is still needed to improve the estimation of L in clumped canopies.