STORM DAMAGE ASSESSMENT BY HIGH RESOLUTION SATELLITE REMOTE SENSING: A CASE STUDY FROM THE NORTHERN BLACK FOREST, GERMANY Steffen Kuntz Abteilung Luftbildmessung und Fernerkundung, Universitat Freiburg, Werderring 6, D-79098 Freiburg, Germany & Christoph Kleinn Abteilung Forstliche Biometrie, Universitat Freiburg, Werderring 6, D-79098 Freiburg, Germany ABSTRACT: At the beginning of 1990, two hurricanes caused the largest storm damages ever known in the forests of Middle Europe with a total of about 73 million cubic meters of broken wood. The survey of the damage situation was mainly done by field survey and frequently in conjunction with aerial photography utilizing local experience. At the same time, a pilot project was carried out to investigate the possibilites of using satellite remote sensing, Landsat TM (Thematic Mapper), to assess the area of storm damages. Some results of this study are given here. Due to the heterogeneity of the storm damage area class and the properties of Landsat TM with respect to spatial and spectral resolution and due to organizational aspects, there are still many problems that inhibit a standard application of this technique at present. STUDY AREA AND OBJECTIVES At the beginning of 1990, two hurricances caused the largest storm damages ever known in the forests of Middle Europe. The forest services had to solve problems of a magnitude not known until then. In the western states of Germany the damage was about 73 million cubic meters of broken wood, which is almost twice as much as the regular annual allowable cut under a regime of sustainable yield. The damages were not spread evenly over the entire region but were concentrated in only some areas, depending on topography, stand structure and other local factors. It was one of the most important issues for the forest administration to get a timely overview of the situation, i.e., of the extent and distribution of the damages. Obtaining reliable information was necessary to plan and organize the logging and wood market activities. Two different levels of information are required then: for forest policy considerations overall statistics may be sufficient, but for the planning of the processing activities spatial information is required. Figure 1 shows the storm damage pattern in a window of the study region. The study region has a size of about 25 x 25 km and is located in the northern Black Forest. One sees that the damaged areas have considerable variation in size and spread more or less over the entire area, while the overall percentage of damaged areas in the entire region is relatively small, less than 1% in this case. The variable of damaged area, therefore, can be regarded as a 'rare event' - at least from the sampling point of view. Sampling procedures for area assessment are critical under these circumstances as highly fragmented areas with a low percent have relatively large sampling errors and therefore require considerable efforts to obtain satisfactory results (Kleinn, 1990). On the other hand, maps are required for planning activities that can hardly be produced by sampling methods. Remote sensing can provide large area information in a short time and give a complete coverage of the region of interest that can in many cases be used to produce maps. Most states in Germany used aerial photographs for an operational storm damage assessment. In conjunction with these activities, interviews and field survey activities were evaluated using local knowledge of the foresters. The general demand of the state forest service was to obtain an overview of the situation within 3 weeks period to be able to adjust management planning. Another idea was to utilize satellite imagery which is available at even lower cost for large areas. No experience in the state forest service was available on the application of Landsat TM for storm damage assessment. Therefore the government of the state of Baden-Wurttemberg supported a pilot study to test the possibilites and to find out the limitations of this technique. The objectives of this study thus are strongly oriented towards an operational use of satellite imagery for storm damage area assessment. 'Storm damage area' is a wide land cover category covering the transition between few single broken trees per hectare and completely broken stands. To get an operational definition for satellite remote sensing of what is to be considered 'storm damage area', in this study a storm damage area was defined as one which is completely or almost completely broken, which means that the crown cover is less than 40%. As in the study area only closed forests exist, reduction of crown coverage means either human logging activities or any catastrophic event. MATERIALS AND METHODS The study area is situated in the northern Black Forest. It corresponds to one map sheet of the topographic map at 1:50,000 scale and has an extent of about 25 x 25 km (map sheet L7316, Wildbad). It was known previously that this area was affected severely by the two hurricanes. Earlier studies (Rahlmann, 1991) showed that mainly due to the heterogeneity of the storm damage area class only a multitemporal approach allows reliable identification of storm damage areas. Three Landsat 5 TM images, dated July 1987, March 1990 and July 1990, were available for the study. No automatic classification procedure was adopted, as Fritz (1991) showed that due to the limited geometric and radiometric resolution and again due to the heterogeneity of the storm damage area class an automatic classification cannot be operational. Even within storm damage areas, the variation of the pixel values is considerable and it seems that 7 spectral bands are not sufficient to properly recognize the target areas by standard automatic classification procedures. The areas were therefore delineated visually at the monitor; which of course requires well trained staff. The data processing and map producing procedure consisted of several steps as listed in the following. The map production used, as sources of information, the multi- temporal Landsat TM images and some known damage areas that served as training areas. #1 Geometrical correction and transformation to the geodetic coordinate system (Gauss-Kruger). #2 Selection of optimal combination of spectral bands to represent damage areas in the image processor (some known damage areas were used as training areas for this purpose). #3 Contrast stretching and edge enhancement for optimization of image contrast. #4 Visual delineation of storm damaged areas. #5 Map production from satellite image interpretation and ancillary information (layers of the digitized topographic map, available from the state geodetic survey). This procedure was adopted for the entire Landsat scene within a 4 weeks period after having received the satellite images, and produced a complete coverage of the inventory region. The most time consuming jobs are the correction procedures in step 1 and the final map production in step 5. The delineation of the storm affected areas itself in the 25 x 25 km window (study area) took the experienced interpreter about two work days. These time figures, however, are experiences from the study region only. Identification and delineation depends very much on the local situation. RESULTS Map accuracy with respect to the target information is one of the most important aspects in map production. The accuracy assessment of products that have been created with non-statistical techniques can only be carried out knowing the true data. To obtain the truth about the storm damaged areas, 64 orthophotos of scale 1:10,000 of the study region dating from summer 1990 were used to delineate the storm damaged areas. The damages were very well visible in these images. There were 241 areas found and digitized, with area sizes between 0.1 ha and 66.5 ha. Several field checks and checks with local foresters, refering to size and location of the damage areas, determined that there were only negligible deviations between the aerial photography delineation and the field situation. The map window shown in Figure 1 is based on the orthophotos, and the corresponding distribution of area sizes is given in Figure 2. It can be seen that the major part of the storm damaged areas have small extent. With respect to this situation, one has to keep in mind that the spatial resolution of Landsat is about 30 x 30 meters and therefore recognition of very small areas poses technical problems. Table 1 compares the total number of areas and total area for the evaluation of the satellite imagery and the aerial photography. The number of areas is very close, but the size of the damaged areas is overestimated by about 20% through the satellite image interpretation, though in the satellite image the small areas delineated in the orthophotos can not be identified. The overestimation can partly be explained by the properties of the sensors. There is a high percentage of mixed pixels, as they generally occur in highly fragmented area structures; and as the damaged areas normally appear markedly lighter than the surrounding closed forests, the mixed pixels tend to be classified as damaged area (Kuntz, 1991). Table 1. Summary of results of the interpretation of aerial photographs and the satellite imagery. Method Number of damaged areas Size of damaged areas ------ ----------------------- --------------------- Satellite 243 580 ha Airphotos 241 483 ha There are several categories of undamaged areas wrongly classified as damaged, such as rocks and less dense old stands with mostly high quality timber (Table 2). Of the damaged stands, 81% were identified and classified correctly (390 ha) in the satellite image. Table 2. List of categories interpreted in the satellite imagery, number of areas and (in parentheses) the corresponding size of the areas. =========================================================================== Recognized as 243 Not recognized 54 damaged areas damaged areas (29 ha) --------------------------------------------------------------------------- Damaged areas 187 (390 ha) Rocks on ground 6 ( 5 ha) Low stocked areas 13 ( 16 ha) Road intersections 8 ( 6 ha) Very young stands 17 ( 23 ha) Low stocked 12 ( 14 ha) old stands =========================================================================== DISCUSSION AND CONCLUSIONS There are several problems that generally exist with the application of remote sensing surveys after catastrophes: - The situation after the catastrophe changes very fast, especially in regions with intensively managed forests. The goal of the forest services is here to process the broken wood as fast as possible, and it can happen that a remotely sensed survey after perhaps two months will give a highly biased figure relative to the true extent of the catastrophe. - When there are several catastrophic events in the same area, possibly with different causes like snow damage and storm damage, it can be very difficult to make a proper distinction if no additional information is available. - With respect to satellite remote sensing, the availability of the images may be a bottleneck. It is not guaranteed that suitable imagery will be available in the time immediately after a catastrophe. Cloud cover is a severe problem as storms normally are connected with a cloud covered sky. - A general restriction to the use of Landsat TM is that the overflight is at 9:30 hrs Central European Time. Because of the shadow problem, the application of these data in mountainous areas is limited to the summertime. However, storm damages occur mostly in winter in Central Europe. - In the study area there is a pattern of storm damaged areas with a large number of small areas. The geometrical resolution of Landsat TM of 30 x 30 meters means that only areas of about 1 hectare and larger can be clearly recognized. As many of the damaged areas in the study are smaller than 1 hectare, application of Landsat imagery is marginal. - Also critical is the misinterpretation of old hardwood stands, forest roads, and rocky ground as damaged. The spectral reflectance of areas where the ground is visible through the canopy layer is similar to that of damaged areas. This problem can only be solved by processing additional information as could reside in geographic information systems. As a general conclusion, it can be stated that, at present, it is not possible to fully make use of the advantages of satellite imagery in the context of storm damage assessments under the situation in the study area. LITERATURE CITED Fritz, R. 1991. Digitale Waldkarte und Kartierung der Sturmschaden 1990 mit Landsat-TM Daten fur das Forstamt Bebenhausen. Unpublished thesis, Forestry Faculty, University of Freiburg, Germany. Kleinn, C. 1990. Estimating the areas of irregular-shaped objects by dot counts. In: Proceedings of the International IUFRO S.4.02 and S.6.04 Symposium on Forest Inventories in Europe with Special Reference to Statistical Methods, May 14-16, 1990, Birmensdorf, Switzerland. Kuntz, S. 1991. Anwendung von Satellitenbildern zur Erfassung und Kartierung von Sturmschaden. In: Oesten, G., S. Kuntz und C. P. Gross, Eds. Fernerkundung in der Forstwirtschaft - Stand und Entwicklungen. Wichmann-Verlag, Karlsruhe, Germany. Kuntz, S. & C. Zimmermann. 1993. Erfassung von Sturmschaden im nordlichen Schwarzwald mit Satellitendaten. AFJZ, accepted, in press. Rahlmann, H. 1991. Visuelle Interpretation van Sturmschaden anhand von Landsat-TM Daten am Beispiel des Schonbuchs der Forstdirektion Tubingen. Unpublished thesis, Forestry Faculty, University of Freiburg, Germany. Zimmermann, C. 1992. Kartierung von Sturmschaden im mittleren Schwarzwald mit Landsat-TM Daten. Unpublished thesis, Forestry Faculty, University of Freiburg, Germany.