Digital image analysis to estimate leaf area

Digital image processing was evaluated as a nondestructive technique to estimate leaf area of Douglas-fir trees (Pseudotsuga menziesii). Four photographs were obtained per tree from side two angles approximate to 90 degrees apart, with separate white and black backgrounds for each angle. Photographs were acquired using a digital camera with computer images (spatial resolution of 1012 by 1524 pixels) recorded directly to a magnetic hard drive, or a standard film camera with photos scanned to produce computer images (spatial resolution 800 by 1200 pixels). Images were separated into red, green, and blue bands with the intensity of the signal digitized into 256 levels (8 bits of accuracy) in each band. Images were processed through a series of operations: choosing of appropriate band for analysis, finding an intensity threshold below which most of the pixels were noise, separating foreground and background pixels, counting the foreground pixels, and converting the number of foreground pixels to a silhouette leaf area (SLA) through comparison with a reference area included in the image. Projected leaf area (PLA) per tree was measured through destructive harvest. Two experiments were carried out: the first to provide an initial function relating SLA to PLA using a small, select, group of trees in pots under diffuse, i.e., scattered, light; and the second using data from trees growing in situ to validate the function under field conditions. In the first experiment, r(2) for SLA vs. PLA was high (0.907) and an initial function was calculated. In the second experiment, predicted PLA for individual plants based on image analysis and the initial function, overestimated measured PLA by approximate to 19%. Thus, a new function for SLA vs. PLA (r(2)=0.861) was calculated based on the larger data set in experiment two. With additional validation, this technique may provide a valuable tool for estimating leaf area nondestructively for studies of environmental stress and vegetation.