EVALUATION OF HURRICANE MITCH DAMAGES IN CENTRAL AMERICA

Transcription

1 EVALUATION OF HURRICANE MITCH DAMAGES IN CENTRAL AMERICA 1 Francis YAKAM SIMEN, 1 Edmond NEZRY, 2 Paul ROMEIJN, 1 Iwan SUPIT, 3 Philippe BALLY 1 PRIVATEERS N.V., P.O. Box 190, Great Bay Marina, Philipsburg, Netherlands Antilles @compuserve.com, Internet: 2 TRE , International Forestry Advisors Prins Bernhardlaan 37, BW 6866 Heelsum, The Netherlands. Internet: 3 SPOT IMAGE 5 rue des Satellites, BP 4359, Toulouse Cedex 4, France. Internet: ABSTRACT The states of Honduras, Nicaragua, and El Salvador have been toughly hit by the Mitch hurricane in the first days of November The extent of damages due to this hurricane, as well as their impact upon local economy were exceptional. In this framework, a remote sensing project was scrambled to provide a large scale evaluation of the damages suffered by the three countries. To reach this objective, new remote sensing products called DYNAMIC products have been designed. These products are, based on using, either RADAR satellite images, or Optical/RADAR satellite data fusion. The most efficient techniques have been applied to produce these products, thus enabling change detection at a very fine spatial scale (10x10 meters). Project schedule and operations, as well as the validation of its products are reported in this paper. 1. INTRODUCTION In the first days of November 1998, Central America was devastated by the hurricane Mitch. Although the three states of El Salvador, Honduras and Nicaragua had already often experienced hurricanes in the past, they had never before suffered such total devastation combining wide area floodings, huge forest cover destruction, important landslides, and large scale destruction of housing and infrastructures. To answer the public call for assistance expressed by these three countries, a joint project was put up by PRIVATEERS N.V., Spotimage and the European Space Agency (ESA) in order to carry out a wide area / high spatial resolution evaluation of the damages caused by the hurricane, using optical (SPOT) and radar (ERS) satellite imagery. The aim of the project was to deliver as quickly as possible accurate high spatial resolution digital and printed maps of the damages that could be used as straightforward as possible by local photo-interpreters. 2. TECHNICAL IMPLEMENTATION To reach this goal, the following data processing procedure was applied by PRIVATEERS N.V. 1) The ERS SAR data were used to detect the damages caused by the hurricane. First, calibration [1] of the whole ERS data set has been done to allow further radiometry-based comparison and change detection. In the change detection process, the ERS-1 SAR archive data acquired before Mitch was compared to the ERS-2 SAR data acquired just after Mitch. 2) Combined speckle filtering and super-resolution techniques developed by PRIVATEERS N.V. [2] have been applied to the ERS SAR data, to detect and map existing targets, before and after Mitch, with particular emphasis to housing and infrastructures, at a spatial resolution of 10 x 10 meters (SPOT Panchromatic resolution). 3) Change detection techniques appropriate to SAR images have been applied to detect and map the damages. 4) SPOT panchromatic archive images or speckle filtered ERS Synthetic Aperture Radar (SAR) images were used as mapping background. 5) Finally, two sets of cartographic (Universal Transverse Mercator projection) maps of before/after Mitch changes have been produced, on 10x10 meters (SPOT/ERS) and 20x20 meters (ERS only) grids, respectively. 6) A documentation describing the photo-interpretation keys and procedures for the end-users in Central America was issued and distributed. Since change detection is carried out on the bi- or multitemporal set of RADAR images, the radiometric resolution of the RADAR images used must be of very high quality. Since the radiometric resolution of RADAR images is naturally corrupted by the presence of speckle, speckle filtering (a too seldom mastered technique!) is the most critical issue for the success of the whole operation. The filter that has been used in this project is an adaptive speckle filter for multi-channel detected SAR images, recently developed by PRIVATEERS N.V.: the Distribution-Entropy Maximum A Posteriori (DE-MAP) filter [2]. The superiority of Bayesian speckle filters is mainly due to the introduction of A Priori scene knowledge in the filtering process. Nevertheless, in the presence of very strong texture or of mixed textures, as it is often the case in SAR images of dense tropical forest, and/or in the presence of relief (which is the case in Honduras, Nicaragua and El Salvador), it may be hazardous to make an assumption about the probability density function of the radar reflectivity of the scene. In this context, the A Priori knowledge with regard to

2 the observed scene can hardly be an analytical first order statistical model. However, in the DE-MAP speckle filters, a Maximum Entropy constraint on texture is introduced as A Priori knowledge regarding the imaged scene. The new DE-MAP filter is particularly efficient to reduce speckle noise, while preserving textural properties and spatial resolution, especially in strongly textured SAR images [2]. It adapts to a much larger range of textures than the previous MAP filters [3,4] developed under the assumption of K-distributed SAR intensity [3]. From the theoretical point of view, it is noteworthy that: - this filter presents the very attractive properties of a control system [5], - it allows to super-resolve SAR images [2,5], which is of high particular interest for ERS (PRI: pixel size: 12.5x12.5 meters, spatial resolution 22x25 meters) and SPOT-Panchromatic (pixel size: 10x10 meters) data fusion. Figure 1: The whole ERS and SPOT coverage used in the Mitch project UTM mosaic of DYNAMIC RADAR and DYNAMIC-SPOT products.

3 3. PROJECT SCHEDULE The SPOT and ERS satellite images were made available by Spotimage and ESA/Eurimage respectively in the course of November and December Altogether, these data allowed to study about km 2 in the most devastated areas of Honduras, Nicaragua and El Salvador. PRIVATEERS N.V. was in charge of project implementation and production of the value added products. The first ERS SAR archive images have been received by PRIVATEERS N.V. on November 17th, Before the end of December 1998, all ERS (19 archive images, and 7 images acquired in November 1998) and SPOT (11 archive images, and 2 images acquired in November 1998) images had been produced and delivered to PRIVATEERS N.V. by Spotimage and ESA. The whole ERS / SPOT coverage used in this project is shown in Figure 1. All ERS-based value-added cartographic products and part of the SPOT/ERS-based value-added cartographic products have been delivered by PRIVATEERS N.V. already during the month of December All the cartographic high spatial resolution value-added products had been delivered in the first days of This efficiency in managing and operating the project enabled rapid delivery of these results to the United Nations for validation planned in January/February 1999 in El Salvador. Validation was also carried out in Nicaragua by Spotimage in January/February 1999 and by PRIVATEERS N.V. in Honduras in March DETECTION OF DAMAGES USING SAR DATA ONLY For every ERS frame, a set of products called "DYNAMIC- RADAR" is produced to enable the detection and the photointerpretation of the changes and damages caused by the Mitch hurricane. The DYNAMIC-RADAR products are cartographic products (Universal Transverse Mercator projection, georeference WGS1984). The sampling rate adopted in the present project is 20 x 20 meters. The DYNAMIC-RADAR products are defined as follows: 1. A multi-date composite Black and White (B&W) image called "CARTO". This image is formed by averaging the calibrated and speckle filtered ERS archive data acquired before the Mitch hurricane. 2. A B&W image file illustrating the change in radar backscatter, and called "RATIO". This image is produced by doing the ratio of amplitudes between the calibrated and speckle filtered ERS image acquired after the Mitch hurricane and the CARTO product. 3. A 3 channels color image file called DYNAMIC- RADAR. A DYNAMIC-RADAR product combines: - the CARTO image (B&W), which is the background RADAR map. - the positive changes in radar backscattering (coded in red) build using the RATIO image. - the negative changes in radar backscattering (coded in blue) build using the RATIO image. The DYNAMIC-RADAR product results from the color composition of the CARTO image and of the two masks of temporal changes, in green/red/blue, respectively. Regarding the photo-interpretation keys, in the DYNAMIC- RADAR product, red tones denote in most cases, either the areas where the vegetation has suffered heavy damages, or destroyed infrastructures. Blue tones denote in most cases flooded areas, or water saturated areas (i.e. Figure 3). 5. DETECTION OF DAMAGES USING SAR AND OPTICAL DATA For every SPOT panchromatic frame (SPOT-PAN or SPOT- M), a set of products called "DYNAMIC-SPOT" is produced to enable the detection and the photointerpretation of the changes and damages caused by the Mitch hurricane. The DYNAMIC-SPOT products are cartographic products (Universal Transverse Mercator projection, georeference WGS1984). Their sampling rate is 10 x 10 meters. The DYNAMIC-SPOT products are as follows: 1. A B&W image called CARTO. This image is a SPOT panchromatic image acquired before (archive data) or after (recent acquisitions) the Mitch hurricane. 2. A B&W image file illustrating the change in radar backscatter, and called "RATIO" (same as for the DYNAMIC-RADAR products). This image is produced by doing the ratio of amplitudes between the calibrated and speckle filtered ERS image acquired after the Mitch hurricane and the CARTO product. 3. a 3 channels image file called DYNAMIC-SPOT. Thus, the DYNAMIC-SPOT product combines: - the SPOT panchromatic image (B&W), which is the background map. - a mask of the positive changes in radar backscattering (coded in red) build using the RATIO. - a mask of the negative changes in radar backscattering (coded in blue) build using the RATIO images. The DYNAMIC-SPOT product results from the addition of the SPOT panchromatic image and of the two masks of temporal changes, in green/red/blue, respectively (i.e. Figure 2). It is therefore the result of a common-sense data fusion process integrating RADAR change detection and SPOT panchromatic data to facilitate photo-interpretation of the final product. Regarding the photo-interpretation keys in DYNAMIC- SPOT products, they are similar to those of the DYNAMIC RADAR products

4 Figure 2: DYNAMIC-RADAR: Provinces of Cortès, Yoro, Atlantida. The main banana plantations area in Honduras, near San Pedro Sula, the second city in Honduras. Situation on November 5, Area represented is 56 x 77 kilometers. In blue, the flooded areas. In red, the damages to vegetation.

5 Figure 3: DYNAMIC-SPOT (Area represented is 51 x 52 kilometers): In Blue: Floodings of Rios Choluteca and Rio Negro, Gulf of Fonseca, Honduras. In Red: damages to pisciculture and agriculture. 6. VALIDATION OF RESULTS Nicaragua The mission of the Spotimage team in Nicaragua in January/February 1999 was the first opportunity to confront the DYNAMIC products to ground reality. After Spotimage team members of this validation campaign, the DYNAMIC products were reflecting correctly the situation, and were suitable to assess the material damages caused by the Mitch hurricane. In particular, the damages caused by the deadly mudflow descending from the Casitas vulcano had been correctly detected, as well as the damages in and around the city of Chinandega. Since then, it seems that the DYNAMIC products concerning Nicaragua were intensively used there. El Salvador As mentioned earlier, the results of this project have been transmitted to the commissioned United Nations body. Original intention was to validate and use these results in El Salvador. However, no feedback has been received. Honduras During a visit in Honduras in March 1999, PRIVATEERS N.V. team members went to compare the DYNAMIC products to ground reality in the area around Tegucigalpa, and along the road from Tegucigalpa to Juticalpa.

6 - In Tegucigalpa, missing houses could be identified, and destroyed houses could be found relying on the corresponding DYNAMIC-RADAR product. The outstanding performances of the new PRIVATEERS N.V. processing techniques preserving the spatial resolution of the SAR could be appreciated there. - Along the road Tegucigalpa-Juticalpa, destroyed bridges could be accurately identified on both the DYNAMIC- RADAR and DYNAMIC-SPOT products. We relied on these products to plan in advance overcoming these difficulties when traveling in this area. - Huge areas (mainly north of Tegucigalpa, easily identifiable in red on Figure 1) where the forest cover has been completely wiped out by the hurricane have also been identified as change areas detected (in red tones) by the DYNAMIC products. In particular, the changes observed on the slopes, originally thought to be artifacts, reflect with a good accuracy the disappearance of forest cover on the top and on the slopes of hills and mountains. A striking example of these changes has been observed on the hills south of Tegucigalpa, where both the trees and the housing were completely destroyed, as detected in the DYNAMIC-RADAR product. 7. TRANSFER OF INFORMATION TO LOCAL END-USERS Whenever possible, and as much as possible, the cartographic products were communicated by the project partners to the concerned bodies and institutions. Transmission of the results in Nicaragua has been done quickly, already in February 1999, due to good prior knowledge of the country. In Honduras, results have been transmitted after a rather long delay. Due to our limited knowledge of the country, the difficulty was to identify the bodies and institutions who could use remote sensing products. Now, Honduran end users (mainly national development agencies) are using them for the reconstruction. The DYNAMIC cartographic products have also been used to organize reconstruction missions send from Europe to Honduras by NGO s (i.e. Association France - Amérique Latine ) in CONCLUSION The Mitch project proved that, in order to produce useful results very shortly after a natural disaster such as a huge hurricane, the building of a professionally serious team, good project partners coordination and the use of appropriate and well mastered techniques lead to rapid project implementation and execution, even if the concerned area is very wide, and the requirements in terms of spatial resolution are very strict (In this project, km 2 were covered at the resolution of 20x20 meters, and km 2 were covered at the resolution of 10x10 meters). Regarding the products themselves, the DYNAMIC products proved to be reliable photo-interpretation tools to locate and identify the changes and the damages caused by the hurricane: flooding (San Pedro Sula, Gulf of Fonseca, etc.), destruction of housing (Tegucigalpa, Chinandega, Leon, etc.) and infrastructures (bridges), forest damages and brutal deforestation (especially in Honduras). However, in future projects of the same nature, improvements must be made with regard to the dissemination of results and the procurement of appropriate technical support to the concerned end-users. This applies in particular to some developing countries where remote sensing is still a little known technology. 9. REFERENCES [1] H. Laur, P. Bally, P. Meadows, J. Sanchez, B. Schaettler, E. Lopinto and D. Esteban, 1998: "ERS SAR calibration: derivation of the backscattering coefficient in ESA ERS SAR PRI products", ESA document nr.es-tn- RS-PM-HL09, Issue 2, Rev.5b, 47 p., 7 September [2] E. Nezry and F. Yakam-Simen, 1999: "Five new distribution-entropy MAP speckle filters for polarimetric SAR data, and for Single or Multi-Channel Detected and Complex SAR Images", Proceedings of SPIE, Vol.3869, pp , September [3] A. Lopes, E. Nezry, R. Touzi and H. Laur, 1993: "Structure detection and statistical adaptive speckle filtering in SAR images", International Journal of Remote Sensing, Vol.14, nr.9, pp , June [4] E. Nezry, F. Zagolski, A. Lopes and F. Yakam-Simen, 1996: "Bayesian filtering of multi-channel SAR images for detection of thin details and SAR data fusion", Proceedings of SPIE, Vol.2958, pp , September [5] E. Nezry, F. Yakam-Simen, F. Zagolski and I. Supit, 1997: "Control systems principles applied to speckle filtering and geophysical information extraction in multichannel SAR images", Proceedings of SPIE, Vol.3217, pp.48-57, September 1997.