Detailed forest information is increasingly desired not only for forest management purposes but also for maintaining and enhancing sustainable forest ecosystems. Although precise measurements of forests can be gathered by field measurements, they are labor intensive and time consuming especially when obtaining enough measurements over large and heterogeneous forest areas. Therefore we need automated and accurate methods which can supplement field measurements. High spatial resolution remotely sensed data can be applied for this objective because developing technologies keep increasing spatial resolution and make it possible to handle large amounts of remotely sensed digital data by powerful computers at reasonable prices. Although high spatial resolution remotely sensed data holds the potential to be a valuable source of information for forest characteristics, a number of challenges still exist in extracting the desired information from this data. Therefore, it is critical to develop and improve automated methods to extract forest information. In this dissertation, I develop and improve the automated methods of extracting individual tree level forest biophysical parameters using high spatial resolution remotely sensed data. While there are many new remote sensing technologies, such as digital aerial photographs, LiDAR (Light Detection and Ranging), radar, and multispectral (or hyperspectral) data, I mainly focus on small footprint LiDAR and aerial images (by digital frame camera) in this study, because these sensors can provide very high spatial resolution data, which are necessary to extract individual tree level biophysical characteristics.

This study consists of three parts, which are basic procedures to exploit high spatial remotely sensed data to extract individual tree level forest biophysical parameters. All three studies are conducted in a mixed-conifer forest at Angelo Coast Range Reserve on the South Fork of the Eel River in Mendocino County, California, USA. First, I develop a robust method to reconstruct Digital Terrain Model (DTM) by classifying raw LiDAR points into ground and non-ground points with the Progressive Terrain Fragmentation (PTF) method. PTF applies iterative steps for searching terrain points by approximating terrain surfaces using the TIN (Triangulated Irregular Network) model constructed from the ground return points. Instead of using absolute slope or offset distance, the proposed method utilizes orthogonal distance to and relative angle between a triangular plane and a node. For that reason, PTF was able to classify raw LiDAR points into ground and non-ground points on a heterogeneous steep forested area with a small number of parameters. The results show the robust performance of the proposed method even under complex terrain conditions. Second, I develop an automated method to detect individual tree tops and delineate individual tree-crown boundaries using airborne LiDAR data. Because of heterogeneous site conditions, I divide the study site into two height classes (high and low trees). For high trees (>= 25 m), I detect tree tops by using a progressive window-size local maximum filter and I conduct an additional verification procedure to reduce false tree top detection by using the shape of canopy profiles between trees. Then, I delineate tree-crown boundaries by marker-controlled watershed segmentation. For low trees (< 25 m), I apply a fixed window-size local maximum filter (1 m radius) to detect tree tops, and I apply the skeleton by influence zones (SKIZ) segmentation to delineate crown boundaries. Compared to fixed window-size local maximum filtering method, our method performed better for detecting and delineating individual trees regardless of tree sizes. Third, I combine aerial images and LiDAR data by means of automated registration procedures using tree tops as corresponding control points. A morphological operation (extended-maxima transformation) is applied to detect tree tops (as common control points) from aerial images and LiDAR data. I conduct the preliminary matching by using the small region of the image center, which was near the principal point. Then, I iteratively expand the control points to the entire images by using the backward projection of the tree top points of the LiDAR data over the aerial images. I employ a local transformation method (piecewise linear transformation) by using detected control points. The adjacent geo-rectified images are mosaicked into one large image by using the seam lines, which are created from the common control points between images. The result shows that the proposed approach enables us to register aerial images with airborne LiDAR data by using individual trees as common control points.

In this study, I develop and improve the automated approaches for extracting individual tree level forest information with high spatial resolution remotely sensed data. I expect the proposed approaches may contribute to cope with a number of challenges for forest information extraction from high spatial resolution remotely sensed data.