Due to rapid technological developments in the area of sensory technology, as well as in computer hardware and software, new perspectives have opened for forestry research over the past several years. Within the framework of this research, which has been supported through the „Bavarian High-Tech Offensive“, therefore, our work has been to find out to what degree these new technologies can be employed for automatic surveys of forest structure. As a prerequiste for this study, it was first necessary to establish a reference area for the testing of remote sensing equipment. Data collection was achieved in four test areas encompassing a total of 3,000 ha in the Bavarian Forest National Park. The test areas include all of the characteristic forest communities (high elevation spruce forest, mixed mountain forest, floodplain spruce forest) and respresent both the natural and development zones. The following remote sensing data was collected: laser scanner (summer and winter flights), digital photography, line scanner, aerial photography (colour and infrared), and interferometric radar (X, L and P bands). In addition, detailed ground reference data was collected in order to be able verify the remote sensing results. The data was taken at 45 precisely (± 5 cm) measured reference areas (0.1 – 0.4 ha), 34 reference stands (0,4 – 6,1 ha), 686 inventory points from the permanent spot testing inventory, and an area-wide survey of forest development stages. With laser scanning, it is possible to determine the ground surface with great accuracy and even better than with conventional forestry practices. The ground model obtained by radar data does not approach this level of accuracy. In stands with tree heights of 20 m to 30 m, it does not represent the actual profile of the terrain. For the measurment of a stand’s upper surface, laser scanning yields a degree of resolution and a level of accuracy which are also superior to the results produced by other methods. Surface models calculated through digital image correlation using digital aerial photographs are also promising. In deciduous forest areas, the accuracy of this method is similar to that of laser systems. With conifers however, the peaks of the trees and the intermittent areas are cut off by the filter algorithm. The resolution and accuracy of the surface models from InSAR data are significantly poorer (5 m). In contrast to laser scanning, radar delivers an averaged value for the stand height. Radar data, therefore, are less appropriate for the measurement of precise local information and can only be used for the deduction of average stand heights. Furthermore, the penetration of radar waves is also dependent on such characteristics of the investigated stands as density and species composition. Three different methods were tested for determining the automated recognition of individual trees and the derivation of tree parameter. While neither template matching nor the fractal net evolution algorithm were able to produce satisfactory values, watershed algorithms were found to yield good results. Employing this mehtod, it was possible to identify 86% of the terrestrial timber volume of the stands in the reference areas. Values, which had been calculated using laser scanning, such as individual tree height (RMS error 3.3 %), crown diameter (RMS error 10.3 %), breast height diameter (RMS error 12.5 %) and individual trunk volume (RMS error 28.2 %) could also be determined. Deciduous and coniferous trees could be distinguished with an accruacy of 90 %. Further investigations were performed to determine to what degree it might be possible to estimate forestry stand parameters derived from laser scanning data using a stastistical approach rather than depending on the delineation of individual trees. For 20 x 20 m sample quadrants, the calculated RMS errors were 5 % for tree height, 10% for diameter breast hight, and between 10 % and 20 % for the volume and various other stand density indices. Stratification of the sample areas in deciduous, coniferous and mixed stands helped to yield more accurate results. Furthermore, a method for the automated recognition of dead wood areas using aerail photographs was also developed. The accuracy of classification exceeded 90 %, making it possilbe to adopt this method for operational use. Less satisfactory were the results for the automated survey of forest development stages. The accuracy of classification reached a value of only 70 %. It could be demonstrated, however, that inexact surveys performed during the forest inspection also had an influence the accuracy of classification. Data collected in the terrestrial surveys attained an accuracy of only 74 %. As a result of these studies, it is possible to conclude that a combination of laser scanning data and digital photographs is well suited for application in forest inventories down to the indivual tree level. Now, that the costs for the collection of this data have gone down considerably in the past few years, it may be assumed that automated methods for remote sening will be playing an important role in forest inventories in the near future.
«
Due to rapid technological developments in the area of sensory technology, as well as in computer hardware and software, new perspectives have opened for forestry research over the past several years. Within the framework of this research, which has been supported through the „Bavarian High-Tech Offensive“, therefore, our work has been to find out to what degree these new technologies can be employed for automatic surveys of forest structure. As a prerequiste for this study, it was first necessa...
»