MAPPING OF BIOTIC DAMAGES USING SATELLITE DATA Mathias Schardt, Technical University of Berlin Klaus Martin, SLU Consulting Grafelfing Manfred Keil, German Aerospace Research Establishment ABSTRACT: In 1987 large parts of the "Nurmberger Reichswald" were damaged by a moth infestation (Lymantria monacha). The damaged area and the level of damages were mapped by using satellite data from Landsat Thematic Mapper. The additional information of stand conditions (soil maps) and tree age (forest management plans) was integrated with a geographical information system (Arc/Info). For verification, the results of the classification were compared with the damage inventory of the Bavarian Forest Experiment and Research Station, based on a color-infrared photo-interpretation. The classification result was merged together with the digitized maps in order to provide further information about the damage distribution in relation to different soil and stand parameters. 1. INTRODUCTION In 1987 large parts of the "Nurnberger Reichwald" were damaged by a moth infestation (Lymantria monacha). Very often this moth appears in great numbers and then leads to very serious forest decline. The caterpillar eats the needles and leaves of all tree species, preferably the needles of spruce (Picea abies) and pine (Pinus silvestris). The most serious effect of the outbreak occurred in the pine stands of the forest district "Allersberg", which is located 30 km south of Nurnberg. The forest of this district is characterized by pure pine stands of nearly the same age class growing on dry and sandy soils. The homogeneous age class distribution (mostly old stands) is a result of a very severe moth outbreak in the beginning of the 20th century. Spruce only occurs on soils with sufficient water supply along small rivers and in the understory of pine stands. The homogeneity of the pine stands makes the forest very vulnerable to insect outbreaks. The distribution of the damages in the pine stands was homogeneous too, so that it was possible to separate training areas mainly consisting of one damage class. The aim of this investigation was to demonstrate the applicability of Landsat Thematic Mapper data for mapping the damaged area and the level of damages. For this purpose the additional information of stand conditions was integrated with a geographical information system (GIS). 2. MATERIALS AND METHODS Thematic Mapper data For the investigation, Landsat Tematic Mapper scenes of three different dates (April and October 1987, and April 1988) were used. The scenes were geometrically rectified using 20 control points based on Gauss-Kruger coordinates and a polynomial of second order. After rectification, the difference between the three scenes was less than one pixel. The scenes were also radiometrically adjusted using undamaged stands as reference areas. Color-infrared aerial photographs In September 1987, color-infrared aerial photographs (scale 1:10,000) were taken of the damaged area. A photo-interpretation performed by the Bavarian Forest Experiment and Research Station was used for ground truth and for the verification. In this investigation altogether 24,628 sample plots (three trees per plot) located on a 25x25m grid were interpreted in order to get information about the area, the distribution and the percentage of damage classes. Four damage classes were separated (Table 1). The aerial infrared photographs and the photo-interpretation were used for the selection of training areas. Training areas, especially undamaged stands, were also checked on the ground. Table 1. Damage class definition of the Bavarian Experiment and Research Station. Damage class Loss of needles ------------ --------------- 0 10% 1 10-50% 2 51-90% 3 91-100% Digital soil maps and forest management plans Keil et al. (1988a,b) showed that the reflectance of pine stands is strongly influenced by the effects of soil conditions. To integrate the information on different water and nitrogen supply conditions, soil maps of the forest district Allersberg were digitized. To minimize the registration error between the Thematic Mapper image and the soil map caused by the "edge effect", the total of 30 different soil types were merged into only 5 different main classes by using the geographical information system Arc/Info (Keil et al., 1990): - sands with low water supply - clays with low water supply - sands with sufficient water supply - clays with sufficient water supply - other soil types (mostly wet soils along small rivers) With this procedure the registration error was less than one pixel. In order to integrate further information about tree age and other important stand parameters, the boundaries of the departments of the forest management plan of the district Allersberg were digitized. 3. SIGNATURE ANALYSIS Signature comparison of undamaged pine stands on different soil types In order to examine the influence of different soil-moisture contents on the reflectance of pines, the signature histograms of unaffected pine stands on wet (mostly clay soils) and on dry soils (mostly sandy soils) were compared. This investigation demonstrated increased signatures associated with a decreasing moisture content of the soil in band 5 (middle infrared) and band 7 (far-middle infrared). The causes of the different reflectance can first be seen in the lower water content and the lower water transpiration of the pine needles of stands growing on soils of low water supply. Secondly, the transparent crowns and the slightly reduced crown closure of those pine stands lead to an increasing portion of illuminated ground vegetation consisting of bilberry and grass. In bands 1-4 (blue, green, red and near infrared) no significant signature changes associated with different site conditions could be observed. Signature comparison of differently damaged pine stands (all soil-types) For the signature analysis, damage classes three and four were combined. Figure 1, Figure 2 and Figure 3 show the histograms of stands with damage classes 0, 1 and 2/3 without any prestratification due to different classes of soil moisture for the Thematic Mapper bands 3 (red), 4 (near infrared) and 5 (middle infrared). The histograms of band 5 (middle infrared) show a significant increase of grey values associated with an increasing percentage of needle loss ( Figiure 3). By using the reflectance difference in this band, it is possible to separate damage classes 0, 1, and 2/3. From this result it can be deduced that a decreasing water supply and an increasing needle loss of pine trees leads to similar signature characteristics. Thus a stratification of the area into areas of lower and higher water supply categories is necessary to increase the separability of different classes of needle loss. In band 3 (visible red) the grey-values also increase with loss of needles, but not as significantly as in band 5 ( Figure 1). In band 4 (near infrared) a decrease of the reflectance goes hand in hand with an increase of needle loss. The difference of reflectance between stands of damage class 0 and 1 is not sufficient to separate them clearly. A decisive reason for the signature overlapping can be seen in the immense influence of ground vegetation of damaged pine stands, because grass and bilberry cause high reflection in band 4. Because of the negative correlation of TM band 4 (near infrared) and 5 (middle infrared) according to needle loss symptoms, the separability of the damage classes can be increased by computing the difference band TM 4 minus TM 5 (Forster, 1989). Figure 4 shows the signature of pine stands without consideration of different soil conditions. Signature comparison of differently damaged pine stands (stratified by soil) Finally after stratification of the training areas into dry and wet soils, a more satisfactory separation of all three damage classes (0, 1 and 2/3) was possible ( Figure 5 and Figure 6 ). From this histogram comparison it can be seen quite clearly that the most significant improvement of separability could be reached for the damage classes S0 and S1. 4. CLASSIFICATION The visual interpretation of the Thematic Mapper color composite already shows different damage classes. For classification, the simple method of thresholding was used. Each damage class was fixed in the difference band TM 4 minus TM 5 by separate thresholds for areas of dry and wet soils (Table 2). The stratification of the Thematic Mapper image into areas of dry and wet soils was performed digitally using the digitized soil map. In comparison to the maximum likelihood method, the simple threshold method offered a very fast and distinct method for this application. Table 2. Thresholds (grey values) for separating damage classes in the difference band TM 4 minus TM 5. Damage class Soil conditions 2/3 1 0 --------------- ----- ----- ----- Dry soils <73 73-82 >83 Wet soils <75 75-84 >84 Combining the classification result with the digitized maps Combining the classification results with the soil map and the forest management plan can provide additional information about damage characteristics in the form of tables (Arnold et al., 1989). Pine stands on dry and sandy soils were more damaged than stands on wet and clay soils (Table 3). Younger stands were less damaged than older ones (Table 4). Table 3. Damage class distribution on different soil types. Percentage of damage class Soil conditions 0 1 2/3 --------------- ----- ----- ------ Dry soils 46.5% 40.7% 12.8% Wet soils 78.6% 19.5% 1.9% Sands 58.3% 34.9% 7.8% Clays 85.5% 13.5% 1.9% Table 4. Damage class distribution in relation to stand age. Percentage of damage class Stand age 0 1 2/3 --------- ----- ----- ------ <20 57.5% 36.9% 5.6% 20-60 33.3% 56.7% 10.0% >60 34.0% 57.5% 8.5% 5. VERIFICATION OF THE CLASSIFICATION RESULT To verify the results, the differences between the damage distribution of the TM-classification and the aerial infrared photo-interpretation were computed for each of the 34 compartments. To calculate the average classification error, these differences were weighted by the size of each compartment. Table 5 shows a satisfactory correspondence (more than 80%) between the satellite classification and the photo-interpretation. Table 5. Verification of the classification results. Damage class % correspondence ------------ ---------------- 0 80.7% 1 80.3% 2/3 95.7% 6. CONCLUSIONS Thematic Mapper data is qualified for mapping biotic damage such as moth infestation. A prerequisite for a successful classification is a homogeneous distribution of the damages so that one pixel can be defined as one damage class. The reflectance of pine stands depands very much on the water supply of the trees arising from different soil types. The consequences of this phenomenon are that poor water supply and needle loss symptoms have the same effect on the signature. By integrating additonal information on soil types into the process of classification using digitized soil maps provided by a geographical information system, it is possible to improve the classification accuracy. 7. REFERENCES Arnold, F., H.-U. Hartmann and M. Keil. 1989. Fernerkundungsdaten als Input in das raumliche Informationssystem LANIS am Beispiel der Waldkarte Regensburg. Fernerkundung, Daten und Anwendungen, Leitfaden 1. Beitrage der Interessengemeinschaft Fernerkundung, Wichmann Verlag, Karlsruhe. Keil, M., M. Schardt, A. Schurek and R. Winter. 1988a. Forest mapping with satellite imagery in Bavaria. Proceeding of the Willi Nordberg Symposium, Graz. Keil, M., J. Janot, I. Roth, M. Schardt, A. Schurek and R. Winter. 1988b. Waldkartierung mit Satellitendaten in Bayern. 2. DLR-Statusseminar "Untersuchung und Kartierung von Waldschaden mit Methoden der Fernerkundung", Oberpfaffenhofen. Keil, M., M. Schardt, A. Schurek and R. Winter. 1990. Untersuchung und Kartierung von Waldschaden mit Methoden der Fernerkundung. Auswertung von Satellitenbilddaten. DLR Abschulssdokumentation Teil B7, pp. 71-135. Forster, B. 1989. Untersuchung der Verwendbarkeit van Satellitenbilddaten zur Kartierung von Waldschaden. DLR Forschungsbericht (DFVLR-89-06, Oberpfaffenhofen).