Identification of Active Deep-seated Landslides in the Central Rif Mountains through PSI, Hotspot and Cluster Analysis
Fonseca, Andre¹; Catalão, João²; Zêzere, José Luís¹; Madeira, José²; El Fellah, Bouchta^3
1RisKam-CEG, University of Lisbon, PORTUGAL; ²LATTEX-IDL (LA), University of Lisbon, PORTUGAL; ³LGC- IS, Univ. Mohamed V, Rabat, MOROCCO

Landslide hazard zonation and the construction of early warning systems for slope instability depend on the existence of complete and regularly updated landslide inventories. These can be achieved through different field and laboratory techniques, normally involving the analysis of aerial photos and satellite imagery, digital elevation models (counter-line based and high-definition such as LIDAR) and time consuming field campaigns. The use of remote sensing data in the study of gravitational processes is becoming increasingly popular as it allows the inspection of large portions of the territory at relatively low costs. Interferometric ground-based InSar surveying, in particular the use of Persistent Scatterer technique, has proven to be a crucial tool in the study of landslide activity and kinematics, allowing the measurement of centimetric to millimetric surface displacements. Over recent years several methodologies have been proposed that enable the interpretation of persistent scatter data in the scope of slope instability analysis.

In this work the velocity and density of persistent scatterers (PS) are analyzed in respect to terrain properties and landslide distribution. Hotspot and Cluster analysis is performed in order to detect PS clusters associated with slow gravitational movement and to refine a regional landslide inventory for large deep-seated slope failures.

The study area is located in the central sector of the Rif Mountains (Northern Morocco) covering an area of approximately 11.000 km2. Reaching over 2000 meters asl this mountain system constitutes a major orographic barrier to the predominant Atlantic air masses traveling from the west. Precipitation is strongly concentrated during winter months with a maximum monthly average of 600mm. The structural setting is characterized by south-verging imbricated thrust faults opposing lithological units with contrasting rheological behavior. Within the hanging-wall of the major thrust faults outcrops the compact and thick fine-grained sandstone and limestone belonging to the Tiziren, Numidian and Dorsal Calcaire units. The footwall blocks are dominated by marls, clay, schists and flysch with affiliation to the Tanger and Ketama Units. The combination of high relief slopes and contrasting lithologies creates the necessary conditions for the occurrence of large deep-seated landslides. Present-day landslide activity is mainly triggered by intense and prolonged precipitation events.

For the purpose of this study a set of 41 SAR images, acquired between May 2004 and September 2010 by ENVISAT satellite along two tracks (51 and 323) were used. The images were supplied by ESA in the scope of ESA Cat-1 (n.10338) Project ''DSLRif - Detection, analysis and kinematics of deep-seated landslides in the central Rif Mountains (Northern Morocco)''. The SAR images were interferometrically processed with one master for each track resulting in two sets of interferograms that were further processed using the persistent scatterer technique.

Spatial clustering of PSs was analyzed using Getis-Ord (Gi) statistics with the objective of identifying sites where deformation is highest and consistent with slow deep-seated slope movement. Getis-Ord statistics is a local spatial statistic method which represents the association of a particular variable (in this case PS) up to a specified distance (d). To ensure that all calculations are performed within the same slope unit, d is normally defined (in the presence of high resolution DEMs) as the average of the shortest distance to channel and ridge. Due to the poor resolution of the available digital terrain data, Sd is here defined as the average over-land flow length from each PS to the ridge and channel (here assumed as proxy to slope length) obtained from an AsterGDEMv2 (pixel size=30.87m; AsterGDEM is a product of METI and NASA). Persistent scatterers with velocities between -1.5&1.5 and those located within intra-mountain basins were not included in the study. To facilitate data interpretation and hotspot identification a smoothed surface was applied to the obtained Gi scores through Kernel density estimation. A landslide inventory was prepared from the analysis of aerial photos at the scale of 1:40 000 and from the interpretation of satellite imagery obtained through free online servers (GoogleEarth and Bing). Furthermore, landslide data was collected from different sources (published papers, thesis and reports). A total of 5340 landslides integrate the final inventory and are primarily slides, flows, earth-flows, and compound or complex movements. Each landslide typology is further sub-divided into deep-seated and superficial movement. For the present study only slow-moving deep-seated landslides with a platform area over 0.1km2 were considered, corresponding to a total of 1530 landslides.

Climatic data for the SAR image timeframe used in this work shows that the winter months between 2004/2005 and 2008/2009 were characterized by an average precipitation bellow 200mm. Nevertheless, the winter period of 2009/2010 presented precipitation values well above average, reaching over 700 and 300mm in December 2009 and January 2010, respectively. Field campaigns allowed the identification of several landslide occurrences and partial reactivations during this last period.

Slope deformation expressed as high velocity hotspots is consistent with sectors previously identified with deep-seated slope deformation. The comparison between the Gi scores and the available satellite images allowed the identification of further 30 sites were slope deformation occurs. Hotpots associated with deep-seated landslides prevail within the less competent lithologies of the Tanger unit (marls, clay and schist). Larger landslides (>4km2) show low deformation Gi scores when compared to smaller landslides. This fact expresses the dependency between landslide size and the ability of a precipitation event to produce a reactivation. In our dataset, while smaller landslides present evidence of reactivation along significant portions of the landslide body, larger landslides tend to suffer partial reactivations, either through the slow downward propagation of a secondary landslide mass or through the degradation (mostly by sliding) of the major landslide deposit along its margins. Nevertheless, this observation is hindered by the fact that PSs are not always evenly distributed within the landslide body, but gathered around particular sites where reflectance is consistently captured by the satellite. Therefore, the nature of partial reactivations of large landslide is difficult to discern when simply referring to Gi hotspot data and further field data collection and detailed geomorphological mapping should be performed in order to avoid misinterpretations of landslide kinematics.