MONITORING OF FOREST DAMAGE CAUSED BY GYPSY MOTH IN HUNGARY USING ENVISAT MERIS DATA ( )

Transcription

1 MONITORING OF FOREST DAMAGE CAUSED BY GYPSY MOTH IN HUNGARY USING ENVISAT DATA ( ) G. Nádor, I. László, Zs. Suba, G. Csornai Remote Sensing Centre, Institute of Geodesy Cartography and Remote Sensing (FÖMI) Bosnyák tér 5., H-1149 Budapest, Hungary, ABSTRACT Remote sensing provides an efficient tool for monitoring forest s caused by gypsy moth. This could be applied at regional level by multi source and multi temporal satellite data evaluation techniques. The objective of this project was to monitor the forest defoliation, the s, caused by gypsy moth in 2005 and The quantitative evaluation of the ENVISAT data, the comparative analysis of the maps, derived from different sensors were all accomplished. This project is a follow on to the monitoring developed in 2004 that applied Landsat TM and IRS LISS data. That basis and the intercalibration of plus FR data made possible the multi year monitoring. As a result, regional remote sensing based forest monitoring technique is available. The ENVISAT FR data provide large scale, timely and reliable information on the development of forest s. 1. INTRODUCTION As a consequence of the climatic change, the gradual deterioration of the forests can be observed worldwide. In Hungary the insect pests (mainly caterpillar of gypsy moth) contributed to this process significantly. The extent of forest caused by gypsy moth was about ha in 2004, about ha in Based on data of gypsy moth eggs spatial assessment made in Autumn, 2005 by the Research Institute of Forestry (RIF) a considerable area was threatened by gypsy moth invasion in Spring, The National Forest Service (NFS) and RIF expected some ha forest that was threatened by gypsy moth. Unlike the previous years, in 2006 the most infected areas were not observed in Veszprém but in Pest, Nógrád and Heves counties. Considerable was expected in Szabolcs-Szatmár-Bereg and Somogy counties, too. The Institute of Geodesy Cartography and Remote Sensing (FÖMI) began a remote sensing based monitoring of s caused by gypsy moth in Cooperation was built with the specialists at RIF. The FÖMI proposal was accepted and supported by the Hungarian Space Office and the Ministry for Informatics and Telecommunication Monitoring the forest caused by gypsy moth in the surroundings of Lake Balaton. The main objective of that project was to monitor the s of forest areas caused by gypsy moth at local and regional scales. Different high and medium resolution and multitemporal satellite data sets were (years 2004 and 2005) used. The assessment of s was done by the quantitative analysis of very high resolution (IKONOS) satellite data in the surroundings of Balatonboglár (local level) by detecting the s in the forests and also in the agricultural lands of their neighbourhood (vineyards, orchards). The regional survey was carried out using medium resolution satellite imagery (2004: ENVISAT, IRS-WiFS, 2005: IRS-P6 ) covering the counties of Veszprém, Somogy and Zala. The comparative analysis of maps delineated by the quantitative interpretation of medium resolution satellite data offered the possibility to monitor and map the procedure of the s and the following processes of regeneration in space and time. Based on the results of the R&D project carried out in 2005 for mapping of the forest caused by gypsy moth it was established that the very high (IKONOS) and medium resolution (mainly IRS-P6 ) satellite data were suitable to identify and monitor forest. The special advantage of remote sensing method in contrast to other methods based on ground survey that we can get quick, objective and unified information from the entire affected area about the extension and spatial-temporal change of the forest both on small area and regional level. This very project is the further developing and spatial and temporal extension of the former R&D project that was carried out by FÖMI in Forest monitoring using ENVISAT and IRS-P6 data presented in this paper was accomplished in two regional study area: Northern-Central Mountain and Veszprém, Somogy counties in Hungary. This project was accepted by ESA in 2006 ( Utilization of ESA Data under Category-1 scheme ESA EO CAT ). It assigned a quota of 50 FR satellite image data (all archived) for the project at reproduction cost. Proc. Envisat Symposium 2007, Montreux, Switzerland April 2007 (ESA SP-636, July 2007)

2 2. DATA USED The medium or medium-high spatial resolution (ENVISAT and IRS-P6 ) satellite images are suitable for identification of forest at regional level. The temporal development of can be assessed using series of vegetation indices derived from satellite images in the actual year, these data were compared the average development of forests in the former years. The no and the degrees of could also be determined. The spatial resolution of IRS-P6 data was intermediate between the Landsat TM, IRS LISS that had been used in 2004 and FR. Moreover IRS- P6 could more easily be correlated to the ground truth The characterization of satellite data By the first evaluation of ENVISAT FR data (spatial resolution 9 ha, spectral resolution 15 narrow bands in the visible and near infrared range) it was obtains that very useful information can be retrieved about the seriousness of the. This is because of the radiation reflected by earth surface is recorded by the sensor of this satellite in 3 bands in the red edge range and 4 bands in the near infrared range. This spectral arrangement provides vital information about the status of the vegetation. Thus spectral changes caused by defoliation due to the caterpillar may be detected particularly. The loss in canopy causes a small increase of reflected radiation depending on the in the red edge range due to the decrease of chlorophyll absorption. The category could be characterized by the reflectance value decrease in the near infrared range. The decrease of the reflectance in this range directly shows the defoliation extent. The bridging of the IRS-P6 satellite with a mediumhigh spatial resolution (0,4 ha) significantly increased the spatial sensitivity of the monitoring compared to ENVISAT data (9 ha). Its spectral resolution was sufficient so that these data provided an excellent linking tool for regional scale forest assessment Reference data collection from d forests Reference data from the study area were collected for the satellite image evaluation and the accuracy assessment of the derived maps. We had used the map of the s caused by gypsy moth from NFS/RIF for Veszprém, Somogy and Nógrád counties generated by their ground survey for the year For the assessment in 2006 we performed reference data collection with the expert of RIF in the Northern-Central Mountain (Fig. 1). Figure 1. Ground reference data collection near Buják, Nógrád county, (21 June 2006) Upper: IRS-P6 LISS III. satellite image (19 June, 2006) The parts of non d forest are brownish, while the parts of forest suffered by defoliation are greenish on the satellite image color composite. The light blue spots (with black identifiers) show the referenced d areas. Below right: ground reference spot (#17) as it reveals the. Below left: photo made on the ground in spot 17. Young oak forest totally defoliated by caterpillar of gypsy moth. 3. METHODS USED The vegetation indices computed from satellite images red and near infrared bands serve as quantitative measurement of defoliation caused by worm. The vegetation index (MGVI: Global Vegetation index) derived by a non linear model from the blue, red and near infrared bands of ENVISAT data [2] so that it described the status of the vegetation as possible. From the red and near infrared bands of IRS- P6 satellite images the well known Normalized Difference Vegetation Index (NDVI) was computed too, that described also the status of the vegetation Preprocessing of ENVISAT and IRS P6 satellite data The radiometric correction procedure of ENVISAT L1b images is described in [1]. This involves the correction due to changing of the solar irradiance in the spectral bands in time to get the reflectances at the top of the atmosphere (ToA). Then the normalisation of the values was applied by using the anisotropic function described in [2]. It corrects the effect of the variation of the sun and sensor geometrical conditions to get normalised ToA reflectance values.

3 Then the rectification of red and infrared bands was done with a help of the blue band to correct the effect of the atmosphere. The normalised ToA reflectances derived from raw (L1b) images and the Toa reflectance values derived from IRS-P6 images were geometrically transformed and rectified into the Hungarian Unified Map Projection System (HUMPS) using polynomial transformation calculated from ground control points (GCP) and nearest neighbour resampling algorithm. The pixel size was set to 250 m * 250 m for FR and to 50 m * 50 m in case of IRS-P6 data. The next step was the derivation of the vegetation index images. MGVI images were derived from the year 2005 and 2006 ENVISAT rectified reflectance values using the appropriate function [2]. NDVI images from the IRS-P6 ToA reflectance data were also derived Intercalibration of ENVISAT and IRS P6 satellite data The most important aspect of this task is to perform an intercalibration procedure between the available ENVISAT FR and IRS-P6 data set that spans over a two year period ( ). Tab. 1 lists the dates of the selected satellite image pairs used in this calibration procedure. The indices from different data sets had to be intercalibrated to one another so that they could be used simultaneously. For the calibration, image pairs from the two types of images (ENVISAT FR and IRS ) were chosen, that had been acquired in the same day or very close to one another. The relationship between the corresponding index values of the same location in the two images is thoroughly studied. The scatter plot with the ENVISAT MGVI and IRS-P6 NDVI values is shown in Fig 2. It can be observed that the correspondence between the two data sets is non linear. Using second order polynomial regression, the correlation coefficient (R 2 ) is very good, more than 90. This was the basis for IRS-P6 based forest monitoring model for 2005 and for This thematic cross-calibration of these two sensors was a link for the time series translation. Measured MGVI values were used to estimate the NDVI values that did not exist in 2003, the reference year. Table 1. FR and images used for intercalibration (May-July, 2005 and June-August, 2006) Acquisition date of Figure 2. The relation between the corresponding pixels of FR MGVI and NDVI images for the dataset acquired for the periods May-July, 2005 and June-August, data were the temporal bridge introduction of data 3.3. Development of forest maps The forest defoliation s may be characterized by the difference of the values of vegetation indices of 2003 (when there was no ) and the appropriate values of the actual year (2005 or 2006). Based on this fact the forest areas examined may be categorized as the follows: Serious : the decrease of the vegetation index is greater than K1. Moderate : the decrease of the vegetation index is between K1 and K2 Slight : the decrease of the vegetation index is between K2 and K3. No : the decrease of the vegetation index is less than K3

4 K1, K2, and K3 are the parameters of the procedure, which can be determined by the comparison of the vegetation index difference with the reference data. The spatial distribution of these categories is shown on the maps. The examinations were restricted only to the forest areas that had been delineated by CORINE Land Cover 50 (CLC50) database [3] and [4]. 4. RESULTS Based on the examinations of the vegetation index curves of the d forests the maximum of both in the study areas of Northern-Central Mountain and occurred in the period of the end of June and the beginning of July of Thus from the available satellite images only those of the 24 June - 4 July period were used for the forest monitoring. For generating the map derived from FR (Fig. 3.a) the pointwise maximum values of MGVI computed from the satellite data (25th June and 4th July, 2005) were compared to the appropriate MGVI values measured on the 1st July, For the map derived from IRS-P6 data acquired on 4 July, 2005 the NDVI values were translated by the second order polynomial (Fig. 2) to provide simulated MGVI values that were used further. Where the image was cloudy it was completed by the image: 24 June, To determine the extent of the, 1 July, 2003 MGVI values were used to describe the non d status of forests (Fig. 3.b). Based on the vegetation index temporal profile of the d forests, the date of the maximum occurred in mid June, Therefore the images acquired in that period were used for the derivation of the map. The pointwise maximum values MGVI computed from the images acquired on June, 2006 were compared to the maximum MGVI values of the 5-days no images (8-12 June, 2003). This comparison reveals and determines the extent of the s in 2006 (Fig. 4.a). For the map (Fig. 4.b) derived from data, the NDVI values computed (19 June, 2006) were transformed to MGVI values in the same way as for 2005 data. Where the satellite image was cloudy, it was completed by data of 28 June, To determine the extent of the, the maximum of MGVI values computed from the the 5-day series (8-12 June, 2003) was used to describe the non d forest status. The d categories areas in the maps are summarized in Tabs. 2 and 3. We made pointwise comparison (Tabs. 5 and 6) between the maps derived from ENVISAT and IRS-P6 data in both of the study areas plus the two years examined. For the comparison the merged category system (Tab. 4) was used. The comparison of Figs. 3.a and 3.b shows that the localization of d areas is very similar on both maps, derived from and in both study areas, the Northern-Central Mountain and. It is very hard to correlate density sliced () categories just because of the continuous index values got to sliced into discrete categories. The 2006 maps are similar even from different satellites (Figs. 4. a, b). Based on the data of Tabs. 2 and 3 the total surface of pixels ranged in different categories match very well for both regional study areas and for both years examined. The pointwise comparison of the maps in Northern Central Mountain, derived from two sensors showed (Tab. 5) that the matching category is very high (above 82), the ratio of the different category is very low (below 10) in both of the years. This is a very good result taking into account the great difference (25 times) in spatial resolution of the two different satellite sensors. In (Tab. 6) the matching are 78 and 87 (2005, 2006) and the different was less than 6 in both years. This is also a very good result. 5. SUMMARY The comprehensive analysis of the maps derived from ENVISAT and IRS-P6 shows that both satellite images are suitable to identify d forest areas. Few qualitative categories of the s were defined. Their spatial match (per pixel) was very good. Percentage of the categories matching and similar in all comparisons were above 90. The areas of the categories (serious, moderate, slight ) also showed very good agreement with each other. The weighed relative difference was below 5. The examination of the vegetation indices and their calibrating process proves that the spectral feature derived from ENVISAT data (MGVI) offers a much more sensitive monitoring and description of the vegetation (forests) status. Finally it can be stated that both new type of medium or medium-high spatial resolution satellite images are very suitable for spatial and temporal monitoring of s caused by gypsy moth. The used methodology is adaptable for monitoring of defoliating s caused by other reasons too. This can be also produced for the entire country when required. Application of an operative system for monitoring forest s by remote sensing should be used to make efficient control on different defoliation reasons. The conditions are far too good during this mild past winter 2007, for the gypsy moth outbreak.

5 Northern Central Mountain Northern Central Mountain 3 a. Northern Central Mountain 4 a. Northern Central Mountain 3 b. Figure 3. Forest maps derived from (a) and (b) satellite data in 2005 Northern-Central Mountain and 4 b. Figure 4. Forest maps derived from (a) and (b) satellite data in 2006 in Northern- Central Mountain and Table 2. Sum area of map categories derived from satellite images in 2005 and 2006 in Northern- Central Mountain categories serious moderate slight no Total

6 Table 3. Sum area of map categories derived from satellite images in 2005 and 2006 in Veszprém- Somogy counties categories serious moderate slight no total Table 4. Definition of the merged category system used for the pointwise comparison of maps derived from and data (m: matching, s: similar, d: different) Categories of map derived from slight moderate serious no Categories of map derived from data slight moderate serious no m s s s s m s d s s m d s d d m Table 5. Result of the pointwise comparison of maps derived from and data in Northern-Central Mountain in 2005 and percentage () matching similar different total percentage () matching similar different total Table 6. Result of the pointwise comparison of maps derived from and data in in 2005 and area percentage () Matching Similar Different Total area percentage () Matching Similar different total ACKNOWLEDGEMENTS The project ( ) was carried out by the support of the Hungarian Space Office. The ENVISAT and other datasets were provided by ESA (ESA EO CAT ) and FÖMI. The reference data were collected by the experts of the Research Institute of Forestry and National Forest Service. REFERENCES 1. N. Gobron, B. Pinty, M. Verstraete, Y. Govaerts (1999): The Global Vegetation Index (MGVI): description and preliminary application, International Journal of Remote Sensing, 1999, Vol. 20. No N. Gobron, B. Pinty, M. Verstraete, M. Taberner (2002): An optimized FAPAR Algorithm Theoratical Basis Document, JRC Publication No. Eur, EN 3. G. Büttner, G. Maucha, M. Bíró, B. Kosztra, R. Pataki, O. Petrik (2004): National land cover database at scale 1:50,000 in Hungary. EARSeL eproceedings 3(3), G. Büttner, G. Maucha, M. Bíró, B. Kosztra, R. Pataki, O. Petrik (2005): CORINE Land Cover mapping at scale 1: in Hungary. International Conference Reaching out to the New EU Member States: Cooperation on Applied Earth Observation/GMES, 27th to 29th of September 2005.