On a multisensor forest inventory system

On a multisensor forest inventory system

F. G. Bercha and D. H. Currie
The Bercha Group
Suite 250, 120 Kensington Road N.W.
Calgary, Alberta, T2N 3P5
Phone : (403) 270-2221;
Telefcx (403) 270-2014; Telex: 03-827666

Abstract
The development and application of an optimal multi-sensor forest inventory system is described. Following review of state-of-the-art approaches to airborne forest inventory techniques, it was concluded that the combination of a multi spectral imaging sensor to provide planar information on forest characteristics with a laser profilometer to provide vertical sectional information constituted an optimal combination. The two sensors have been utilized in unison, coordinated through a digital computer, to provide a unified three-dimensional description of the forest. The description includes principal forest characteristics such as species composition, forest condition, crown density, and secondary characteristics in the horizontal plane, combined with vertical sectional properties including tree height, vertical stratification, and foliage density. These characteristics may then be used to determine economic parameters such a biomass and timber volume. The system described consisting of a multispectral camera and laser profilometer connected to a digital data acquisition system, ahs been developed up to the operational phase through a series of activities including system integration, hardware acquisition and testing, extensive real time and post-acquisition data processing software development, and pilot survey execution. The latter, which included comparison with ground truth data, showed excellent agreement in the primary geometric properly measurements of the forest and good agreement in the areas of secondary property evaluation. The system has been found to provide an economical and efficient technique for obtaining and forest change monitoring for a range of conditions in both temperate and tropical climates.

Introduction
Te purpose of the work described herein was to develop an airborne data collection and analysis facility which would be useful for forest assessment, either as a stand-alone system or in combination with other airborne and ground based data acquisition systems or in combination with other airborne and ground based data acquisition systems. Current technologies used for assessments of large areas include analysis of airphotos and satellite imagery (Ahern 1987). These approaches all depend heavily upon ground truth data for volume estimation since tree heights can only be estimated. Large-scale photography (Spencer 1987) has been used to obtain tree heights from airborne platforms however the technique is costly and time consuming due to the manual analysis required for each photo pair.

Previous investigations by the authors and others have demonstrated the utility of the laser rangefinder for direct measurements of tree heights in an airborne configuration (Aldred 1985, Bercha 1987, Nelson 1984).by pairing this unique sensor with a second one, capable of obtaining images of the forest canopy, an analyzing the acquired data as an integrated set, a system which could generate a three dimensional representation of the forest canopy in a digital format was predicted.

Prior to this work, the laser systems were often operated in conjunction with a video camera in order to recover the aircraft flight line in post mission processing. These two sensors are complementary in the sense that their operational envelopes overlap (flying height, speed etc). additionally, they offer two very different data sets since the laser measures in the vertical plane, while the video camera produces a horizontal image. Unfortunately, conventional vide imagery is not easily interpreted using the analytical produced by these systems. By using a multispectral video camera, the same image data may be separated into several discrete images representing the scene reflectance at a variety of wavelengths. These images can be captured using a video digitizer and analyzed in a manner analogous to the digital analysis of Landsat imagery.

Airborne System Description
The forest inventory system is constructed fro low-cost components using a central data acquisition computer to ensure system integrity and to provide timing control. The main sensors are a laser rangefinder and a multispectral vide camera. Facilities for film cameras and other alternative sensors are included in the data acquisition package. Figure 1 shows the system installed in a small aircraft.

The laser rangefinder is a gallium-arsenide (GaAs) diode laser capable of pulsing at rates of upto 4000 times per second. The laser pulses are in the near infrared at a wavelength of 902mm. Reflection of the narrow pulses are captured by a sensitive photo detector and the two way travel time is converted to distance using the constant speed of light. Special circuitry is provided to allow discrimination between multiple targets and allow the laser system to report either the shortest or longest range as selected by a data acquisition computer. In an airborne operation, the longest range will represent the ground, whereas the shorter ranges will represent the tree tops.

While range of the treetops is reliably obtained, the range to the ground is not always available due to obstruction of the laser beam by heavy foliage. To maximize the frequency of ground ranges, the laser is pulsed at a rate, which provides a 90% overlap between pulses. The data is filtered in real time so as to give a concise data ser consisting of ground ranges and tree heights. A statistical summary of the number of ranges interpreted by the foliage is recorded to allow estimation of crown closure along with profile.

The multispectral video camera is based on a conventional charge coupled device (CCD) sensor, which is scanned at the North American standard 60 fields per second (NTSC). The sensor resolution is 384 pixels by 491 lines spatially, while the spectral response is from 0.4 to 1.1 mm (Frost, 1985). The sensor is located behind a rotating, shutter, which contains six, user selectable, narrow band spectral filters. The shutter speed is synchronized with the scan frequency of the video circuitry in such a way that each field of video imagery is acquired through a different filter. This provides continuous six band coverage at a rate of ten image sets per second.

In order to relate the video imagery with the laser profiles in post-mission processing, a digital time code is inserted into the video signal prior to recording. These time codes are reliable by a computer controlled video playback unit.

Figure.1 Multisensor System

Laser rangefinder data
Reduction of the laser data is a two-step process. The first requirement is to determine the range fro the aircraft to the ground. The involves filtering out the ranges which are intercepted by the foliage so that the remaining range measurements indicate the ground surface. These ground ranges are considered an intermediate result for scaling the video images.

Once the ground trace is defined, the second step is to subtract the short ranges to give the tree heights. A considerable amount of research has been devoted to determining how well the laser detects the top of the tree crown (Aldred 1985). The factors which come into play here are the sensitivity of the laser to the small branches at the top of deciduous tress ad the difficulty in determining whether or not the laser detected the highest point on the crown. Both of these effects will cause a shorter tree height than is actually the case of be measured. Studies with the laser used in the present application have shown that sensitivity to small objects is not a problem as fairly wide beam dispersion is used and the ranging electronics provide excellent control over the target selection. Thus, for deciduous trees with large, flat-topped crowns, the height measured will be fairly accurate. Another consideration is that the laser is not maintained n a constant vertical orientation and will therefore produce height measurements which are slightly longer that the actual free size. The net effect has been found to be that the raw range data is within 0.5 meters of the control data when averaged over small segments of the flight line.

An important aspect of the laser cross-section is its display of the vertical foliage distribution along the profile as shown in figure 2. This output is useful for discriminating between tree species during image interpretation and can also be used to locate defoliated crowns, which may be due to disease or insect attack.

Figure.2 Laser Range Data

Multispectral video imagery
The video imagery is captured using a video "frame grabber" board interfaced to a digital image analysis system. This system utilizes high-speed analog to digital converters ADC's) to convert the incoming video signal to a digital raster. Consecutive fields are captured and registered to form a multiband image. The time codes stored in the video signal are also captured and used to correlate the image is scaled and then radiometrically enhanced so a to equalize the relative brightness of each band. The images are then registered to a master image to a obtain a high-resolution image strip of the flight line.

The goal of the analysis phase is to obtain a general classification of the imagery by delineating stand boundaries. The laser data is used to differentiate between stands with similar appearances but different heights. Recognition and identification individual tress is possible manually but ha not been implemented as an automatic process. In mixed stands, separation of deciduous species is unreliable. One of the video sensor, identification of the individual species is unreliable. One of the main factors affecting the image quality is the amount of solar illumination.

General classifications are first attempted using an unsupervised clustering techniques. If this approach, a supervised approach is used. At this point the laser data can be used to advantage. Since different tree species present different vertical profiles in terms of leaf and branch distributions, the analyst can identify the various types by comparing the laser data to the video image. This generally leads to a successful classification.

Forest assessments
The multi-sensor data set obtained from the laser and video data processing is most easily interpreted in relation to existing maps or small-scale imagery. The delineated stand boundaries are mapped from the master image and their areal extent is determined. The species composition, average age, and height for each stand is then tabulated. Using existing height-volume growth curves, the merchantable and total timber volumes for each stand is then calculated. If reliable growth curve information is unavailable, ground truth programs in the survey area my be required to ensure the accuracy of the assessment.

Due to the highly accurate height measurement obtained and the low cost of surveying large areas, the laser/video system is ideal fro-monitoring yearly growth rates o specific stands.

Conclusions
The development and application of an integrated system comprising an airborne laser rangefinder and a multispectral video camera was described. The sensor package described has the advantages of relatively low cost while providing a direct measurement of tree height and thematic inventory data which can be easily processed by semi-automatic means.

References

Ahern, F.J.; Leckie, D.G. 1987. Digital Remote Sensing for Forestry : Requirements and Capabilities, Today and Tomorrow. Geocarto International 2:3 pp 43.52.
Aldred, A.H; Bonnor G.M, 1985 Application of airborne lasers to forest surveys. Information Report PI-X-51, Can. For. Serv. Petawawa, Ont.
Bercha et al, 1987. Airborne Laser Mapping Pilot Project Final Report, Central Kalimanath, Indonesia.
Forest, P.A. 1985. A Multispectral Video Imagign and analysis System Xybion Elc.Systems Corp. Cedar Knolls, Nj
MacLean, G.A: Krabill, W.B 1986. Gross Merchandable timber Volume estimation using an airborne Lidar System. Can.J.of Rem. Sens. 12:1 pp 7-18.
Nelson, R.; Krabill, W.; MacLean, G. 1984. Determining Forest Canopy characteristics using airborne using airborne laser data. Rem. Sens. Of Env. 15: 201-212.
Spencer, R.D. 1987. Large Scale Aerial Photographic Systems for Forest Sampling in Canada. The Canadian surveyor 21:1 pp 9-12.