Abstract:
Geographic Information Systems and Remote Sensing are important contributors to Sustainable
Forestry Management Plans. Remote sensing techniques for image interpretation provides the
means to extract valuable information that could be expensive and time-consuming to obtain
through field observations (Franklin et al. 2001).
Spatial Products derived from the interpretation of airborne and satellite borne images feed
Geographic Information Systems to develop strategies and methodologies for resource
management, harvest planning, fire management, map production, and model predictions.
(Yusmah et al. 2015)
This study has three important objectives: to test the feasibility of template matching for the
identification of single pine tree crowns, to conduct a delineation of pine plantations using
relational features and to evaluate how single tree crown size affects the accuracy of the
proposed method.
Templates of single trees were produced in the software eCognition Developer. The sampling
process comprised the random selection of 3000 single pine trees in 7 different test sites (test sites were grouped in 3 categories according to the single tree sizes). A first rule set to detect
single tree crowns was developed in eCognition Developer, using three different template groups (4, 8 and 16 templates). Through an analysis of variance, the number of single tree
crowns detected was compared for the different template groups.
Using a second rule set in eCognition, the template matching algorithm combined with
relational, spectral and contextual information were applied to delineate pine plantation areas.
An accuracy assessment was performed in the test sites for all thematic classes identified.
Finally, an Analysis of Variance evaluated the influence of single tree crown size on the overall
accuracy.
Potential applications and improvements to the proposed methodology for single tree crown detection and plantation delineation are proposed at the end of the document.
