Comparison of individual tree crown delineation method for carbon stock estimation using very high resolution satellite images

Greenhouse gas inventories and emission reduction program requires scientifically robust methods to quantify forest carbon storage in forest. Remote sensing techniques are accurate and low-cost alternatives to the field based assessment. High spatial resolution remotely sensed imagery provides viable sources and opportunities for forest inventory at an individual tree level. The study aims to compare two methods to delineate the individual tree crown to develop a model to estimate carbon stock obtained from field survey and the crown projection area derived from high resolution satellite data of 0.5 m spatial resolution (GeoEye-1). The performance of two algorithms, namely region growing and valley following of object oriented classification was compared in the dense broadleaf forest of Ludhikhola watershed, Gorkha, Nepal. The region growing of eCognition and valley following of ITC algorithms were used to delineate individual tree crowns produced a segmentation accuracy of 68% and 58% respectively. The region growing is based on finding the local minima to create the boundary and local maxima to locate the potential tree top. The valley following algorithm is also based on finding the local minima by following the shades between the valleys of the trees canopies. Shorea robusta trees were identified in the image using a nearest neighbor classification approach with an overall accuracy of 74%.