Header

UZH-Logo

Maintenance Infos

Automatisierte Erkennung und Kartierung von Lawinenablagerungen mit optischen Fernerkundungsdaten


Bühler, Y. Automatisierte Erkennung und Kartierung von Lawinenablagerungen mit optischen Fernerkundungsdaten. 2009, University of Zurich, Faculty of Science.

Abstract

Snow-avalanches kill more people in Switzerland than any other natural hazard and threaten buildings and traffic infrastructure. Rapidly available and accurate information about the location and extent of avalanche events is important for avalanche forecasting, safety assessments for roads and ski resorts, verification of warning products as well as for hazard mapping and avalanche model calibration/ validation. Today, isolated observations from individual experts in the field provide information with limited coverage. Only a fraction of all avalanche events can be recorded due to restricted accessibility of many alpine terrain sections during winter season. Information on small to medium size avalanche events within remote regions is collected only sporadically. However, these avalanches notably claim most casualties within the raising number of people pursing off-slope activities. Remote sensing instruments are able to acquire wide-area datasets even over poorly accessible regions. Therefore they are promising tools to close the abovementioned information gap. This research systematically investigates the potential of spatially high resolved remote sensing instruments for the detection and mapping of snow-avalanche deposits. Fieldspectroradiometer data of nine avalanche deposits are analysed to identify universally valid and significant spectral offsets between avalanche deposits and the adjacent undisturbed snow cover. Promising absorption features are found in the near infrared region of the electromagnetic spectrum. Nevertheless, the differences are unlikely to be distinct enough for a detection using air- or spaceborne remote sensing instruments. The directional reflection of rough avalanche deposit surfaces is contrary to the directional reflection of smooth undisturbed snow covers. The potential of multiangular remote sensing data for the detection and mapping of avalanche deposits is demonstrated using multiangular data acquired by the airborne scanner ADS40. However, the difference between observation angles (16°) proves to be insufficient for accurate avalanche detection solely on the base of directional properties. Therefore, auxiliary data has to be utilised. The texture of avalanche deposits and undisturbed snow cover can already be distinguished by the naked eye. Using second order statistics, comprising the spatial distribution of the variation in pixel brightness, textural characteristics in digital image data can be quantified. This is a prerequisite for an automated detection of particular textures. Different established texture measures are tested for their discriminating potential of avalanche deposits and undisturbed snow cover using RC30 aerial images of avalanche deposits acquired within the avalanche winter 1999 in Switzerland. The control parameters such as the size of the filter box are systematically varied to find the ideal settings. The texture measure Entropy is identified as the most distinct and stable indicator to distinguish between rough and smooth snow surfaces. But avalanche deposits are not the only rough snow surfaces within the Alpine winter landscape. For example wind modelled snow surfaces or artificially piled snow at the edge of roads and ski slopes show texture characteristics similar to avalanche deposits. Consequently, a classification approach using texture information only is not sufficient for an accurate identification of avalanche deposits. Based on the findings described above, we develop an avalanche detection and mapping processing chain, combining spectral, directional and textural parameters with auxiliary datasets. The processing chain is tested and improved using data acquired by the airborne scanner ADS40 over the region of Davos, Switzerland. The accuracy assessment, based on ground reference data within three test sites, shows that 94% of all existing avalanche deposits are identified. Even small scale deposits (area < 2000 m2) and deposits within shadowed areas are detected correctly. These results demonstrate the big potential of the proposed approach for an automated detection and mapping of avalanche deposits. Yet, cloud cover constrains the availability of appropriate optical remote sensing data after heavy snowfall while wind modelled snow surfaces, artificially piled snow and sparsely vegetated snow surfaces cause sporadic misclassifications. Despite these constraints, the approach developed within this research shows a big potential to fill existing gaps in avalanche information. Especially within alpine areas of developing countries with almost no reliable information on past avalanche events, such an approach may be used to acquire valuable data for hazard mapping and settlement planning.

Abstract

Snow-avalanches kill more people in Switzerland than any other natural hazard and threaten buildings and traffic infrastructure. Rapidly available and accurate information about the location and extent of avalanche events is important for avalanche forecasting, safety assessments for roads and ski resorts, verification of warning products as well as for hazard mapping and avalanche model calibration/ validation. Today, isolated observations from individual experts in the field provide information with limited coverage. Only a fraction of all avalanche events can be recorded due to restricted accessibility of many alpine terrain sections during winter season. Information on small to medium size avalanche events within remote regions is collected only sporadically. However, these avalanches notably claim most casualties within the raising number of people pursing off-slope activities. Remote sensing instruments are able to acquire wide-area datasets even over poorly accessible regions. Therefore they are promising tools to close the abovementioned information gap. This research systematically investigates the potential of spatially high resolved remote sensing instruments for the detection and mapping of snow-avalanche deposits. Fieldspectroradiometer data of nine avalanche deposits are analysed to identify universally valid and significant spectral offsets between avalanche deposits and the adjacent undisturbed snow cover. Promising absorption features are found in the near infrared region of the electromagnetic spectrum. Nevertheless, the differences are unlikely to be distinct enough for a detection using air- or spaceborne remote sensing instruments. The directional reflection of rough avalanche deposit surfaces is contrary to the directional reflection of smooth undisturbed snow covers. The potential of multiangular remote sensing data for the detection and mapping of avalanche deposits is demonstrated using multiangular data acquired by the airborne scanner ADS40. However, the difference between observation angles (16°) proves to be insufficient for accurate avalanche detection solely on the base of directional properties. Therefore, auxiliary data has to be utilised. The texture of avalanche deposits and undisturbed snow cover can already be distinguished by the naked eye. Using second order statistics, comprising the spatial distribution of the variation in pixel brightness, textural characteristics in digital image data can be quantified. This is a prerequisite for an automated detection of particular textures. Different established texture measures are tested for their discriminating potential of avalanche deposits and undisturbed snow cover using RC30 aerial images of avalanche deposits acquired within the avalanche winter 1999 in Switzerland. The control parameters such as the size of the filter box are systematically varied to find the ideal settings. The texture measure Entropy is identified as the most distinct and stable indicator to distinguish between rough and smooth snow surfaces. But avalanche deposits are not the only rough snow surfaces within the Alpine winter landscape. For example wind modelled snow surfaces or artificially piled snow at the edge of roads and ski slopes show texture characteristics similar to avalanche deposits. Consequently, a classification approach using texture information only is not sufficient for an accurate identification of avalanche deposits. Based on the findings described above, we develop an avalanche detection and mapping processing chain, combining spectral, directional and textural parameters with auxiliary datasets. The processing chain is tested and improved using data acquired by the airborne scanner ADS40 over the region of Davos, Switzerland. The accuracy assessment, based on ground reference data within three test sites, shows that 94% of all existing avalanche deposits are identified. Even small scale deposits (area < 2000 m2) and deposits within shadowed areas are detected correctly. These results demonstrate the big potential of the proposed approach for an automated detection and mapping of avalanche deposits. Yet, cloud cover constrains the availability of appropriate optical remote sensing data after heavy snowfall while wind modelled snow surfaces, artificially piled snow and sparsely vegetated snow surfaces cause sporadic misclassifications. Despite these constraints, the approach developed within this research shows a big potential to fill existing gaps in avalanche information. Especially within alpine areas of developing countries with almost no reliable information on past avalanche events, such an approach may be used to acquire valuable data for hazard mapping and settlement planning.

Statistics

Downloads

259 downloads since deposited on 19 Feb 2010
17 downloads since 12 months
Detailed statistics

Additional indexing

Item Type:Dissertation
Referees:Itten K I, Kellenberger T W, Frei U
Communities & Collections:07 Faculty of Science > Institute of Geography
Dewey Decimal Classification:910 Geography & travel
Language:German
Date:2009
Deposited On:19 Feb 2010 13:45
Last Modified:21 Nov 2017 14:45
Number of Pages:156
Related URLs:http://opac.nebis.ch/F/?local_base=NEBIS&con_lng=GER&func=find-b&find_code=SYS&request=005942817

Download

Download PDF  'Automatisierte Erkennung und Kartierung von Lawinenablagerungen mit optischen Fernerkundungsdaten'.
Preview
Filetype: PDF
Size: 13MB