A Multi-scale Approach for Delineating Individual Tree Crowns with Very High Resolution Imagery

This paper presents a multi-scale approach for delineating individual tree crowns (ITC) from high spatial resolution imagery. By analyzing the evolution of image gradients over the scale-space constructed with orthogonal wavelets, tree crown boundaries are effectively strengthened while the textures resulted from tree branches and twigs are largely suppressed. Two scale consistency checks, a scale and a geometric consistency check, were devised to account for tree crown’s radiometric and geometric characteristic. After an edge-enhanced image was acquired, a previously developed marker-controlled watershed segmentation method was adopted to delineate ITC. An experiment was carried out in a study site in California. Field measurements of crown size of 58 trees were compared with those derived from aerial imagery. An R square value of 0.68 was achieved. It was found that crown size was underestimated from the photo interpretation compared to that from the ground survey. The result can be attributed to the fact that pixels lying on the tree crown boundaries are poorly represented in the image. Introduction A common practice in forest studies is to stratify forests into different stands. Each stand occupies a contiguous area that contains a number of trees that are relatively similar in species composition or age and different from adjacent areas (Lindenmayer and Franklin, 2002). At the stand level, various parameters are of interest including tree density, stand diameter, stand table, stand height, crown closure, stand volume, and site index. Timely and accurate acquisition of these parameters is not only critical to updating forest inventory, which normally deals with the estimation of spatial distribution of wood volume (Spurr, 1948), but is also vital for ecological studies, in which quantitative modeling of ecosystem processes can be made possible with such parameters as inputs (Palace et al., 2008). Compared to traditional labor-intensive field surveys, remote sensing opens up an effective and unique avenue to acquire these parameters in a more automatic fashion. This paper aims at automatically deriving one of these important stand-level parameters, tree crowns, from remote sensing imagery. The increasing availability of Very-High-Resolution (VHR: one meter or sub-meter level) imagery from either airborne or PHOTOGRAMMETRIC ENGINEER ING & REMOTE SENS ING Ap r i l 2010 371 Department of Geography, University at Buffalo, the State University of New York, 110 Wilkeson Quad., Buffalo, NY 14261, ([email protected]). Photogrammetric Engineering & Remote Sensing Vol. 76, No. 4, April 2010, pp. 371–378. 0099-1112/10/7604–0371/$3.00/0 © 2010 American Society for Photogrammetry and Remote Sensing A Multi-scale Approach for Delineating Individual Tree Crowns with Very High Resolution Imagery AAG Remote Sensing Specialty Group 2008 Award Winner1