> Research

Classifying Forest Structure Using

Small-footprint Multi-return Airborne LIDAR


Background

The conservation of biological diversity within the multitude of existing ecosystems throughout the planet requires a wide variety of management strategies. Unfortunately, planners that design and implement such strategies are at the mercy of information that is available to them.Furthermore, the art and science of ecosystem management is constantly becoming more complex as new studies continue to provide insights into the dynamic interdependencies between various combinations of species and their environment(s) that cross social ecological, and economic dimensions.In order to handle such complex management demands, decision support systems have evolved to take advantage of geospatial information.Efforts to discover, describe, and utilize a wide variety of environmental information must continue to satisfy increasing demands into intelligent decision support systems.

Environmental variables of particular interest in forest management are those that are capable of describing, in detail, the three-dimensional organization of vegetation (or physical structure) within forested ecosystems.Ecosystem management and the decision-making that determines how we affect the long-term sustainability of an ecosystem can be greatly enhanced by the incorporation of higher quality information.However, relying on field-based inventories of vegetation alone, within a forested ecosystem, would require an enormous investment of resources and would not adequately capture enough information necessary for effective decision-making.In fact, impending concerns over declining biological diversity across multiple scales encompassing individual species, communities, ecosystems, and landscapes have precipitated efforts to develop and implement more robust planning strategies.Information relating directly to the continuous nature of biological diversity across entire landscapes, at the species level, is certainly cost prohibitive considering the rate at which we should be implementing more intelligent management plans. However, research into ecological relationships have yielded information that can be used to guide decision-makers about policies for land management. Such research suggests using proxies to biological diversity, such as continuous detailed information on the vegetation structure within a forest, that may provide information useful in coarse filter approaches to conserving biodiversity (Lippke, 1999).In fact, for forested systems, it has been generally accepted that the more structurally diverse a forest is the more species rich that forest is (Brokaw, 1999). Understanding such relationships and leveraging geospatial information describing such information will help devise more suitable and sustainable management plans.

Remote sensing technologies such as light detection and ranging (LIDAR) systems are able to discern vegetation structure, in great detail, within forested ecosystems (Ritchie et al., 1993). Small foot-print, multiple-return LIDAR systems should be able to detect the vertical and horizontal distribution of forest canopies (Means, 2000). An investigation of this technology was conducted to determine the potential for developing information about vegetation structure that could improve the performance of the ecological diversity matrix (EDM), a decision-support system already in use (Haufler et al., 1999).

The study was conducted near McCall, Idaho (Figure 1.0) where the primary goals were to determine how well LIDAR could discern various structural characteristics within the predefined landscape.The study site was situated within the Idaho Southern Batholith Landscape (ISBL) which was the EDM test bed.The ISBL was derived from research which described holistic, relatively homogeneous sections and subsections within the Idaho Batholith landscape (Mehl et al., 1998).

The project is a United States Forest Service and Mississippi State University cooperative research initiative with support from the Boise Cascade Corporation.At the time of this study the Boise Cascade Corporation was utilizing the Ecosystem Diversity Matrix to manage their lands within the Southern Idaho Batholith Landscape.The EDM employed by Boise Cascade Corporation combines both habitat classification and vegetation growth stages found within the ISBL towards the creation of an ecosystem diversity matrix, which is then used as a guide for designing and planning forest management activities.LIDAR systems may help improve the cost-effectiveness of such a management system by reducing the amount of time required to collect information on the ground as well as improve the overall performance of the decision support system.

LIDAR overflights of the McCall, ID sudy site were conducted by EarthData Technologies on October 12, 1998.EarthData employed a custom built Azimuth Aeroscan small foot-print, multi-return LIDAR system.Field data collection was conducted from mid-June through mid-July during the summer 2000 field season.Field efforts concentrated on sampling the variation of vertical and horizontal structure as well species composition within the study site.

Figure 1.0 The location of the NE and SW research blocks (black outlined boxes) within the ISBL study site (red outline).

Methods

Various field/computer processing techniques were used to derive information on vertical and horizontal forest structure.For statistical purposes, vertical structure was defined by using tree height variance  This statistic was chosen due to it being easily tested using an analysis of variance. For representing vertical structure spatially, the percent coefficient of variation was chosen due to the more normalized nature of the metric.Both metrics were derived and tested from tree height information using field and LIDAR data.LIDAR tree heights were determined from a series of processing steps (see Figure 2.0) 

Horizontal structure was defined as percent canopy closure as determined from multiple observations on each plot.This metric was used to help define the habitat type and vegetation stage of the plots as well as a means for excluding areas that were not considered to be ‘forest’ as defined by the EDM ( < 10% canopy closure).LIDAR derived metrics of percent canopy closure were based on thresholding LIDAR data to 2.0 m above ground (assuming canopy closure observations were from the maximum height of person taking the reading).  Remaining features were grouped into one class.The binary image produced resulted in trees and not-trees. Theoretically, the tree class represents interpolated canopy LIDAR shots.The tree area divided by the total area per plot was estimated percent canopy closure per plot.

Figure 2.0 Processes used to derive tree height information from LIDAR data: a. Subtracting the interpolated (0.2 m) 1st return LIDAR from the interpolated (0.2 m) bare earth LIDAR yields an absolute forest height surface, b. The absolute forest height surface is processed using a tree height finding model (developed by John McCombs, 2001) to yield the highest point within an individual peak where red dots represent assumed individual trees, and c. The tree heights data set is clipped using a polygon coverage (representing plot boundaries) to yield height of individual trees detected within each plot (red points).

Results

According to the analysis, there were significant differences between tree height variances between single-story and multi-story plots.The results suggested that a spatial analysis of tree height variances might provide a suitable metric for classifying vertical structure within the study area.Classification of the entire study area was performed for vertical structure using the tree height variance statistic (Figure 3.0).

Figure 3.0 Various perspectives of single-story vs. multi-story vertical structure classes:
a. 3D perspective of relative LIDAR-derived tree heights within a single-story plot,
b. 3D perspective of interpolated 1st return LIDAR data along with tree heights (red points) within a single-story plot,
c. Photograph typical of single-story plots encountered in the field,
d. 3D perspective of relative LIDAR-derived tree heights within a multi-story plot,
e. 3D perspective of interpolated 1st return LIDAR data along with tree heights (red points) within a multi-story plot, and
f. Photograph of typical multi-story plots encountered in the field.



Figure 4.0 Mean differences between tree height variances within single-story plots and multi-story plots.



Spatial Analysis

Mapping the distribution of vertical and horizontal structure was performed using two different methods as mentioned above.For vertical structure a decision rule was employed to separate the two structure classes from each other based on the median value between the minimum tree height variance observed in the multi-story plots (2.75 m) and the maximum tree height variance observed in the single-story (1.21 m) plots which resulted in a cutoff value 1.54 m.This was used to classify each 30.0 m cell in the tree height variance data set into a single-story (< 1.54 m) or a multi-story (> 1.54 m) class (Figure 5.0).

Figure 5.0 Vertical structure classification for SW research block. Light green represents single-story class (tree height variance < 1.54 m per 30.0 m cell) and dark green represents multi-story class (tree height variance > 1.54 m per 30.0 m cell); image rendered using bilinear interpolation.



For the vertical structure classification, an accuracy assessment was performed at the landscape level, which revealed a significant agreement between referenced and classified plots of single-story and multi-story conditions respectively. Producer’s accuracy (a measure of how consistent the reference plots compare to the classification plots) was very high for both single-story and multi-story classes as was users accuracy (a measure of how likely a pixel was correctly classified with regard to the same class type on the ground). The overall classification accuracy was>95% and the overall kappa statistic () was 0.89. The results suggest that LIDAR-derived measures of vertical structure can be used to map vertical structure accurately at landscape scales.

To represent the continuous nature of both vertical and horizontal forest structure, unclassified statistical surfaces were generated and rendered. Such datasets should provide more accurate bases from which to invoke decision-support queries to augment the state of knowledge in assumed species-environment interactions. The vertical structure variable was derived using the coefficient of variation (CV)statistic based on LIDAR derived tree heights encompassing each research block. The CV was chosen as it represents a more normalized value of tree height variability around the mean for a given sample area allowing for a more stable spatial representation of the statistical surface (Figure 6.0). Horizontal structure (canopy closure) was determined through the use of interpolated LIDAR data. Any value greater than 6.0 m was assumed to be tree canopy and was assigned a value of 1or otherwise 0. The resulting binary image was then used to calculate the percent canopy closure in 30m x 30m blocks(Figure 7.0) as it is a well-established resolution for landscape level spatial analyses.

Figure 6.0 Horizontal structure (canopy closure) for the SW research block.
The upper left image shows the resulting 1m clumping of the interpolated LIDAR above 6.0 ft (1.827) based on the average height of the field workers.The upper right and larger image were the resulting 30 x 30m percent canopy closure data sets.



Figure 7.0 Vertical structure or coefficient of variation (CV) for the SW research block. The upper left image shows the resulting 30 x 30m representation of tree height CV with red and green showing high and low respectively.The upper right and graph shows a comparison between vertical structures in each of the different ecological land units (ELU) encountered in the field.



The trends in Figure 7.0 indicate clear relationships between vertical structure class and CV. In each case, there were distinct differences between plots of small trees verses plots of medium trees (as defined by the EDM) from lower to higher tree height CV’s respectively. This suggests that plots with larger trees generally have more complex vertical structures, and tree height variance can be used to describe vertical structure. Overlaying both the vertical structure and horizontal structure clearly shows the continuous nature of these forest variables (Figure 8.0). The accurate portrayal of forest conditions in this manner has important implications in designing and managing forests while considering all ecologic, economic, and social consequences of various management regimes.

Figure 8.0 Vertical (single-story = no-blue, multi-story = blue) superimposed on horizontal structure (green = low, red = high %canopy closure) provide color combinations that indicate structural diversity in the SW research block. Call out boxes point to four conditions ranging from single-storied/low canopy closure to multi-storied/high canopy closure conditions.



Literature Cited

Brokaw, N.V.L., and Lent, R.A.1999. Vertical structure. Pp. 373-399in Malcolm L. Hunter (ed.) Maintaining Biodiversity in Forest Ecosystems. Cambridge University Press. Cambridge, United Kingdom

Haufler, J.B., Mehl, C.A., and Roloff, G.J. 1999. Conserving biological diversity using a coarse-filter approach with a species assessment. Pp. 107-125 In R.K. Baydack, H. Campa III, and J.B. Haufler (eds.) Practical Approaches to the Conservation of Biological Diversity. Island Press. Washington, D.C.

Lippke, B. R., and Bishop, J. T. 1999. The economic perspective. Pp. 597-638 in Malcolm L.Hunter (ed) Maintaining Biodiversity in Forest Ecosystems. Cambridge University Press. Cambridge, UK.

Means, J.E., Acker, S.A.,Harding, D.J., Blair, J.B., Lefsky, M.A., Cohen, W.B., Harmon, M.E., and McKee,W.A. 1999. Use of large-footprint scanning airborne LIDAR to estimate forest stand characteristics in the Western Cascades of Oregon. Remote Sensing of Environment 67:298-308.

Mehl, C.A., Steele, R., Warren, S.,Holt, B., Haufler, J.B., and Roloff, G.J. 1998. The ecosystem diversity matrix for the Idaho Southern BatholithLandscape: A users manual. BoiseCascade Corporation. Boise, Idaho.

Ritchie, J., Evans, D., Jacobs, D., Everitt, J., and Weltz,M. 1993. Measuring canopy structure with an airborne laser altimeter. Transactions of the American Society of Agricultural Engineers. V.36(4): I235-I238 – July-August 1993.