Back in 2018 we reported on some interesting research being done by Daniel Hölbling (Department of Geoinformatics—Z_GIS, University of Salzburg) and his colleagues on landslide detection, monitoring and forecasting. Since then he has been continuing this research and recently published a new paper entitled “Mapping and Analyzing the Evolution of the Butangbunasi Landslide Using Landsat Time Series with Respect to Heavy Rainfall Events during Typhoons”. This new study investigates (1) semi-automatic approaches to map landslide evolution and (2) the potential correlation between changes in landslide area heavy rainfall during typhoon events.
The use of Trimble eCognition’s Object-Based Image Analysis (OBIA) tools play a critical role in his analysis.
Beyond the immediate damage caused by landslides, Hölbling reports that “large landslides can initiate natural hazard cascades by damming rivers and inducing catastrophic flash floods and debris flows…The large amount of mobilized debris that originates from landslides significantly affects the drainage system”. 2009 brought record-breaking rainfall to the island of Taiwan which led to severe flooding and several horrendous landslides, including the Butangbunasi landslide. Hölbling and his colleagues have focused their recent research on studying this event in an effort to gain a “deeper knowledge of the evolution of landslides and their triggering factors” as this “is crucial for hazard mitigation”.
Remote sensing plays a central role in the investigation of landslides – cost effective data is increasingly available over large expanses of area that are otherwise difficult to access. In addition, such data allows researchers to study other natural disasters that and potential impacts of landslides that may not be in the direct are of the initial event.
The combination of Earth observation (EO) data with eCognition OBIA tools “provides a suitable methodological framework for efficient landslide mapping, as well as landslide change analysis” according to Hölbling. He points out that a number of studies have used OBIA methods to examine landslides, but none of them have considered the integration of multitemporal (i.e. time series) images.
In this study, Hölbling studies the evolution of the Butangbunasi landslide based on a rich multitemporal data set spanning the Landsat archive from 1984 to 2018. In total, 20 images were used – “In particular, we selected the first cloud-free image of the area of interest available after a typhoon […] event that shows a noticeable change in the landslide area compared to pre-event images”. Since the first available image for the study area was captured in 1984, the researchers chose this as their T0. In addition to the Landsat data, a slope layer derived from the 2008 ALOS Palsar DEM was used in the analysis.
To begin the analysis, NDVI, MSAVI and brightness layers were calculated as they flowed into the multiresoluiton segmentation to create image objects that would become the basis for the subsequent classification. The specific segmentation parameters used by the authors can be found in Table 4 of their publication. It is important to point out that the authors are utilizing an often overlooked strength of the Trimble eCognition software, namely the ability to look beyond the initial input data. Sometimes the best features for describing an object are contained in derivatives such as NDVI or brightness – simple calculations that can be made directly within eCognition via Layer Artithmetics.
A knowledge-based classification approach is then applied primarily based on the spectral indices. The authors indicate that the “main indication for mapping the landslide area was the absence of vegetation, which leads to a distinctive spectral contrast between the landslide-affected area and its surroundings”. To avoid false positives, the DEM and slope data was used, “for example, debris accumulation areas with low slopes at low elevations in the Laonong river bed.”
After the automated classification, a manual quality assurance was carried out to correct any small miss-classifications. Once finalized, the authors analyzed the change between two successive acquisitions within the time series.
The authors found that the results of the semi-automated methodology are comparable to manual digitizing approaches. In fact the classification of landslide areas was improved through automation. The authors conclude by stating that “semi-automated techniques can limit the subjectivity in landslide mapping and can contribute to improving the reproducibility of landslide maps…This reduces the analysis time and increases the transferability of the approach.”