iSLIDE - Integrated Semi-automated Landslide Delineation, Classification and Evaluation

Landslides constitute a major natural hazard in almost all mountainous regions of the world. Today, the wide range of available Earth Observation (EO) data implies the need for reliable and efficient methods for detecting, analysing and monitoring landslides in order to assist hazard and risk analysis. Hence, it is of high importance to make use of effective techniques in order to gather information about the exact location, extent and type of landslides in a fast and transparent manner. Object-based image analysis (OBIA) provides a great potential for semi-automated landslide detection and classification, since - in comparison to pixel-based approaches - not only spectral, but also spatial, morphometric, textural, as well as contextual properties can be addressed. Through the integration of multiple data sets landslides can be examined in a more efficient way, making use of the most suitable properties of the available information layers. The overall objective of this project is to develop a methodological framework for landslide delineation, classification and evaluation through the integration of optical remote sensing data and digital elevation information, as well as terrain unit layers using innovative OBIA methods. Additionally, the potential of SAR data for object-based landslide mapping will be investigated. If the use of SAR data appears to be beneficial, it will also be considered as input for the integrated analysis. The methodology will be developed and tested on one Austrian and two Taiwanese study areas, which are frequently affected by landslides. An important component of our framework will be the definition of digital signatures of landslide types that will facilitate the transformation of expert knowledge into machine-understandable rules. Such a conceptual foundation will make the approach robust and transferable to other study areas, en route to fully automated landslide analysis. Furthermore, the development of automated object-based change detection methods will enable a fast detection of fresh landslides after landslide events, as well as the monitoring of existing landslides. Classification results will be repeatedly evaluated by applying novel accuracy assessment methods. Thus, the classification scheme will be improved, and subsequently also the accuracy of the final outputs. The proposed framework is designed to contribute to an increased reliability, transferability and automation in object- based landslide detection, classification and change detection, as well as evaluation through developing innovative methods and applying fully integrated workflows. It is expected that this research will break new ground in the field of object-based landslide analysis, especially with respect to conceptual and methodological developments. The project will make an essential contribution towards the development of a methodology that should be I) objective, II) transferable across areas, III) robust against changing input data and resolutions, and IV) automated.

In the project iSLIDE a methodological framework for landslide delineation and classification through the integration of optical remote sensing data, synthetic aperture radar (SAR) data and digital elevation information using innovative object-based image analysis (OBIA) methods was developed. Landslides constitute a major natural hazard in almost all mountainous regions of the world. Today, the wide range and increasing spatial, spectral and temporal resolution of available Earth Observation (EO) opens up new possibilities for mapping landslides. However, this implies the need for innovative, reliable and efficient methods for detecting, analysing and monitoring landslides and the development of effective techniques in order to gather information about the exact location, extent and type of landslides in a fast and transparent manner. OBIA provides a great potential for semi-automated landslide detection and classification, since - in comparison to pixel-based approaches - not only spectral, but also spatial, morphometric, textural, as well as contextual properties can be addressed. In this project the integration of multiple EO-data sets in an object-based framework, i.e. optical satellite images, SAR data, DEMs, was thoroughly examined and the most suitable properties of the available information layers were used for landslide investigation. While several studies have recognized the value of segmentation optimization for increasing the objectivity and transferability of landslide mapping, the optimization of the classification step is lagging behind. Making EO-based semi-automated methods more objective is of high importance for a range of applications, for increasing their acceptability and for potential implementation in operational workflows. Within iSLIDE a landslide mapping system was introduced that is based on expert knowledge models and implemented in OBIA. These expert knowledge models hold the operational knowledge of experts about landslides and digital landslide mapping such as data, classification features, and feature thresholds, and facilitate the transformation of expert knowledge into machine-understandable rules. Moreover, a class-specific object-based change detection method has been developed that enables a fast detection of fresh landslides after landslide events, as well as the monitoring of existing landslides. The project made an essential contribution towards the development of a methodology that is I) objective, II) transferable across areas, III) robust against changing input data and resolutions, and IV) automated. The EO-based classification methods developed in iSLIDE are not limited to the field of landslide research, but can be transferred to other application areas where specific features are mapped.