Incorporating tree competition in individual tree crown delineation from airborne laser scanner (ALS) data.

摘要

Crown characteristics are significant individual tree-based measurements used in forest inventory for predicting responses to silvicultural treatments and for incorporation in growth and yield models to estimate tree growth. As an alternative to measure crown size in the field, individual tree crown delineation (ITCD) from remotely sensed data plays a critical role in modern forest inventory. Although various ITCD algorithms have been developed, there is still development potential for improving the accuracy of ITCD. The overarching goal of this dissertation is to propose a novel algorithm that integrates ecological processes into traditional image processing for enhancing individual tree crown delineation. In the first manuscript, ITCD related studies published in the past two decades (1990-present) were reviewed from different perspectives (i.e., remotely sensed data type, methodology, research purpose, and forest type) and the trends in this field were identified. In the second manuscript, the impact of combining ALS data and orthoimagery on the accuracy of treetop detection using local maximum filtering was investigated; and the impact of growth order (i.e., sequential, independent, and simultaneous) on the accuracy of crown delineation from marker-controlled region growing (MCRG) using ALS data was also explored. The results showed that complementary data from the orthoimagery reduced omission error associated with small trees in treetop detection. Growth order in MCRG influenced crown delineation accuracy and simultaneous growth had better performance than sequential growth. In the third manuscript, a novel agent-based region growing (ABRG) algorithm using ALS data was proposed to capture both growth and competition processes (i.e., one- and two-way competition). Results showed that ABRG improved the overall accuracy of tree crown delineation compared to MCRG. One-way competition was likely to improve overall accuracy in coniferous plots when tree height variability was relatively large; while two-way competition worked more efficiently for deciduous trees, especially when trees had intensive competition. The degree of improvement by ABRG was related to the characteristics of trees and density in the plots. Automatic ITCD algorithms that integrate geospatial technologies and ecological processes will continue to be an attractive forest management topic for both the forestry and remote sensing communities.