Data Objective

Object-oriented and cognitive detection and characterisation of landslides

Tapas R. Martha1,2 and Norman Kerle1

1

Faculty of Geoinformation Science and Earth Observation (ITC), Department of Earth Systems Analysis

(ESA), Twente University, P.O. Box 6, 7500AA Enschede, the Netherlands, *****@***.**

2

National Remote Sensing Centre (NRSC), Indian Space Research Organisation (ISRO), Hyderabad

500625, India - ******@***.**

Abstract

Geoinformatics tools and methods have proved to be of great value in disaster risk management,

with remote sensing allowing comprehensive, accurate and timely acquisition of information on

hazard processes, elements at risk or consequences of a hazardous event, which can be analysed

or integrated with auxiliary information. Landslides are one of those phenomena where

geoinformatics developments have opened up new ways to monitor potential or ongoing slides, but

also to build inventories of previous mass movements as the basis for hazard and thus risk

assessment. This constitutes a valuable tool for a hazard type that led to more than 400 fatal

disasters worldwide in 2008 that in total killed over 32,000 people, and often is the only method

that allows rapid and timely landslide mapping in mountainous areas. Several recent disaster

events, such as the earthquakes in China (2008) or Haiti (2010), led to hundreds or even

thousands of slides, traces of which often start to disappear before field-based mapping is

possible, leaving remote sensing data as the only suitable mapping basis.

In the past such mapping was primarily done by visual analysis of aerial photos or,

increasingly, satellite imagery, and a number of automatic methods have been developed.

However, until recently only pixel-based methods, primarily employing different classification or

change detection techniques, were developed (e.g. Mantovani et al., 1996, Nichol and Wong,

2005). Those are beginning to be replaced with approaches based on objects or segments.

Object-oriented analysis (OOA) is inherently more suitable, as it can address the phenomena

studied, landslides in this case, as what they are objects, not pixels that have spectral, spatial

and contextual characteristics (Dragut and Blaschke, 2006). They thus allow limitations of pixel-

based methods, which are largely restricted to using spectral and texture information, to be

overcome. Past landslide characterisations have identified a number of different landslide types

and defined them, for example in terms of source material type, run-out length, failure plane

curvature or crown shape (Moine et al., 2009, Martha et al., 2010). Potentially any of these

characteristics can be employed in OOA, provided suitable data needed to calculate those

parameters are available (McDermid and Franklin, 1994).

Using OOA efficiently also raises several problems. The actual analysis is reliant on proper

image segmentation, the subjectivity and trial-and-error nature of which has been the subject of

years of research (Martha et al., 2010). Hence here we address how image information itself,

rather than visual fine-tuning, can be used for an objective segmentation. We show an approach to

select objectively the parameters of a region-growing segmentation technique to outline landslides

as individual segments. The multiple scale parameters needed to represent the size-variable

features were determined using a plateau objective function derived from the spatial

autocorrelation and intrasegment variance analysis of Moran s I. We used a high resolution (5.8 m)

Resourcesat-1 LISS-IV multispectral image of an area in the High Himalayas in India as

segmentation basis for the initial landslide detection, and obtain terrain curvature from a Cartosat-1

(2.5m) derived digital elevation model for subsequent landslide type identification. Here optimal

segments were used in a knowledge-based classification approach, with the thresholds of

diagnostic parameters derived from K-means cluster analysis. This allowed the detection of five

different types of landslides, with an overall detection accuracy of 76.9%. The robustness and

transferability of the approach were assessed by applying the ruleset in a geomorphologically very

different area in Darjeeling (Martha et al., in review).

While multi-spectral data are an ideal source to distinguish landslides from a range of false

positives, often only pan-chromatic data exist shortly after a disaster event. This raises the

question to what extent such data can be used in a similar OOA approach. Hence we adapted our

ruleset for IRS-1D panchromatic data (5.8 m), focusing specifically on the use of GLCM texture

measures in conjunction with derivatives from the previously used Cartosat-based DEM. This

procedure, too, is largely automated, and showed that panchromatic images alone can be used in

OOA-based landslide detection and characterisation, with results comparable to the approach

based on multispectral data.

Finally, we also discuss how the OOA approach presented here can be extended to include

also the mapping of other parameters needed in risk assessment, such as elements at risk, to

establish the basis for comprehensive risk assessment.

References

Dragut, L. and Blaschke, T., 2006. Automated classification of landform elements using object-

based image analysis. Geomorphology, 81(3-4): 330-344.

Mantovani, F., Soeters, R. and Van Westen, C.J., 1996. Remote sensing techniques for landslide

studies and hazard zonation in Europe. Geomorphology, 15(3-4): 213-225.

Martha, T.R., Kerle, N., Jetten, V., van Westen, C.J. and Vinod Kumar, K., 2009. Characterising

spectral, spatial and morphometric properties of landslides for automatic detection using object-

oriented methods. Geomorphology, doi:10.1016/j.geomorph.2009.10.004.

Martha, T.R., Kerle, N., Jetten, V., van Westen, C.J. and Vinod Kumar, K. Segment optimisation

and data-driven thresholding for knowledge-based landslide detection by object-oriented image

analysis, in review for IEEE Transactions on Geoscience and Remote Sensing.

McDermid, G.J. and Franklin, S.E., 1994. Spectral, spatial, and geomorphometric variables for the

remote- sensing of slope processes. Remote Sensing of Environment, 49(1): 57-71.

Moine, M., Puissant, A. and Malet, J.-P., 2009. Detection of landslides from aerial and satellite

images with a semi-automatic method. Application to the Barcelonnette basin (Alpes-de-Haute-

Provence, France). In: J.-P. Malet, A. Remaitre and T.A. Boogard (Editors), International

Conference 'Landslide Processes: from geomorpholgic mapping to dynamic modelling'. CERG

Editions, Strasbourg, pp. 63-68.

Nichol, J. and Wong, M.S., 2005. Satellite remote sensing for detailed landslide inventories using

change detection and image fusion. International Journal of Remote Sensing, 26(9): 1913-1926.

Varnes, D.J., 1978. Slope movements types and processes. In: R.L. Schuster and R.L. Krizek

(Editors), Landslides: Analysis and Control. Special Report 176. Transportation Research Board,

National Academy of Sciences, Washington D.C., pp. 11-33.

Contact this candidate