Miloš Marjanović

Landslide Susceptibility Assessment with Machine Learning Algorithms

Case study addresses NW slopes of Fruška GoraMountain, Serbia. Landslide activity is quite notorious in this region, especially along the Danube’s right river bank, and recently intensified seismicity coupled with atmospheric precipitation might be critical for triggering new landslide occurrences. Hence, it is not a moment too soon for serious landslide susceptibility assessment in this region. State-of-the-art approaches had been taken into consideration, cutting down to the Support Vector Machine (SVM) and k-Nearest Neighbor (k-NN) algorithms, trained upon expert based model of landslide susceptibility (a multi-criteria analysis). The latter involved Analytical Hierarchy Process (AHP) for weighting influences of different input parameters. These included elevation, slope angle, aspect, distance from flows, vegetation cover, lithology, and rainfall, to represent the natural factors of the slope stability. Processed in a GIS environment (as discrete or float raster layers) trough AHP, those parameters yielded susceptibility pattern, classified by the entropy model into four classes. Subsequently the susceptibility pattern has been featured as training set in SVM and k-NN algorithms. Detailed fitting involved several cases, among which SVM with Gaussian kernel over geo-dataset (coordinates and input parameters) reached the highest accuracy (88%)outperforming other considered cases by far.

A geotechnical model of the Umka landslide with reference to landslides in weathered Neogene marls in Serbia

This paper describes a characteristic landslide model for landslides typically hosted in Neogene formations in Serbia, especially along the right banks of the Sava and Danube Rivers. It is focussed on the particular landslide Umka near Belgrade, which is a paradigm for numerous landslides in that area. Various field investigations and laboratory tests carried out in several campaigns, including 1979, 1991–1993 and 2005, underpinned the conception of a general model for this typological landslide. Additionally, a new landslide monitoring campaign started in 2010 provided supplementary data support for the model development. Landslide characteristics, sliding mechanism and material properties based on all these data are first summarised and discussed and then featured in a general model. It is assumed that the landslide is hosted in the weathered zone of grey marls and that the main sliding surface typically propagates along the contact between the fresh and weathered marls. Furthermore, the triggering is principally associated with lateral river erosion in the landslide toe, although heavy precipitation and snow melting have been witnessed to be important indirect triggers. Their correlation to the recorded displacements was difficult to determine due to complex hydrogeological relations and an isolated groundwater system, which is another common characteristic of this landslide type. Back analysis on the basis of the adopted model and the determined geotechnical parameters has been performed. The latter analysis is of particular interest because the Umka landslide is currently under consideration for a mitigation and stabilisation plan related to the construction of a new motorway route.

Using multiresolution and multitemporal satellite data for post-disaster landslide inventory in the Republic of Serbia

This paper focuses on a specific event-based landslide inventory compiled after the May 2014 heavy rainfall episode in Serbia as a part of the post-disaster recovery actions. The inventory was completed for a total of 23 affected municipalities, and the municipality of Krupanj was selected as the location for a more detailed study. Three sources of data collection and analysis were used: a visual analysis of the post-event very high and high (VHR-HR) resolution images (Pléiades, WorldView-2 and SPOT 6), semi-automatic landslide recognition in pre- and post-event coarse resolution images (Landsat 8) and a landslide mapping field campaign. The results suggest that the visual and semi-automated analyses significantly contributed to the quality of the final inventory, including the associated planning strategies for conducting future field campaigns (as a final stage of the inventorying process), all the more so because the field-based and image-based inventories were focused on different types of landslides. In the most affected municipalities that had very high resolution satellite image coverage (19.52% of the whole study area), the density of the recognized landslides was approximately three times higher than that in those municipalities without satellite image coverage (where only field data were available). The total number of field-mapped landslides for the 23 municipalities was 1785, while image-based inventories, which were available only for the municipalities with satellite image coverage (77.43% of the study area), showed 1298 landslide records. The semi-automated landslide inventory in the test area (Krupanj municipality), which was based on coarse resolution multitemporal images (Landsat 8), counted 490 landslide instances and was in agreement with the visual analysis of the higher resolution images, with an overlap of approximately 40%. These results justify the use of preliminary inventorying via satellite image analysis and suggest a considerable potential use for preliminary visual and semi-automated landslide inventorying as an important supplement to field mapping.

Stabilization of fine-grained soils with fly ash

Results of laboratory research focusing on soil stabilization, using fly ash without activators, are presented in the paper. Two types of fine-grained soils were tested: low to medium plasticity clay and very expansive, medium to high plasticity clay. Soil-fly ash mixtures were prepared at optimum fly ash contents (15 and 20 %). The effects of fly ash on the soil plasticity, moisture-density relationship, unconfined compressive strength, shear strength parameters, CBR (California Bearing Ratio) values, deformation parameters, and swell potential, were evaluated. Results obtained show that the use uf fly ash can significantly contribute to the improvement of soil properties.

Rockfall Monitoring Based on Surface Models

This research addresses surface analysis based on the Terrestrial Laser Scanning data exampled on rockslope site along M-22 highroad near Ljig in Serbia. The slope is about 100 m wide and over 20 m high. The scanning was performed in three epochs 2011, 2013 and 2014, but only the latter two were involved in the analysis due to insufficient quality of the pilot epoch. Gap between the latter two epochs coincided whit local rockfall events detected in the middle section of the slope. Scanning was performed by Leica ScanStation P20 instrument. Two consecutive point clouds were produced and gridded to 3 cm resolution. Comparative assessment of both point clouds and related surface models revealed the volumes and spatial extents of the anticipated rockfalls. Data were analyzed in two software packages: CloudCompare and GeomagicStudio, that have different modeling approaches based on direct point cloud analysis and mashed surface comparison, respectively. We compared advantages and disadvantages of each approach. Finally, we addressed how LiDAR technology can contribute qualitatively and quantitatively for research and monitoring of rockslopes.

IPL Project 181: Study of Slow Moving Landslide Umka Near Belgrade, Serbia

Serbia is well known for numerous landslide phenomena. Landslides are particularly notable for the valley walls of the rivers Sava and Danube and their respective tributaries. They have in common the fact that they all originated in complexes of Neogene sediments made of different lithological elements, and most often clays, sands, and marls with pronounced zones of weathering up to approx. 20 m deep. Landslides on the right banks of the Sava and Danube have deep sliding surfaces, formed on the contact of the weathered zone and unaltered clay and marl sediments. The basic trigger of the processes, apart from the precipitation, is prolonged erosion of the right banks of the Sava and Danube rivers. Most of the landslides are active or suspended, where periods between reactivation phases could be several years long. The IPL-181 Project started in November 2012. The study area is located on the right bank of Sava River, 25 km southwest of Belgrade, Serbia. The project focused on review and analysis of previous detailed site investigations and field instrumentation, analysis of aerial photo and orthophoto images, and analysis of monitoring results. Project beneficiaries will be the local community, and local and regional authorities. Here we present results of the 1st year of proposed project targets—a review and analysis of previous field investigations performed by Project participants.

Machine Learning and Landslide Assessment in a GIS Environment

This chapter introduces theoretical and practical aspects for applying GIS and geocomputation methods in landslide assessment problems. Machine Learning techniques in combination with GIS are proven useful for computation and building of complex non-linear spatial models, which is why they have been chosen in our work. Modeling principles that include basic Machine Learning techniques (Artificial Neural Networks, Decision trees, Support Vector Machines) and additional useful procedures are described to show how they can be applied to address a complex problem such as landslide assessment. Two types of models are proposed in the work herein that are useful for describing landslide susceptibility and landslide prediction. The region of Halenkovice in Czech Republic is presented as a case study to illustrate and bring closer the practical aspects of landslide assessment. These aspects consider data preparation and preprocessing, scale effects, model optimization, and evaluation. The results show that Support Vector Machines and similar Machine Learning (ML) techniques can be successfully applied to address the zoning of landslide susceptibility, which might be an important breakthrough for potential applications in regional planning and decision-making.

Relative Landslide Risk Assessment for the City of Valjevo

This paper represents a relative landslide risk assessment of the City of Valjevo in Western Serbia. After the extreme rainfall during the May 2014, many new landslides were triggered, and Valjevo was one of the most affected areas in Serbia. The modeling was preceded by the data selection, and included ranging and preprocessing of the conditioning factors. The following eight factors were chosen as representative: stream distance, slope, lithology, elevation, distance from hydrogeological borders, land use, erodibility and aspect. Landslide susceptibility analysis was completed using the Analytical Hierarchy Process (AHP) multi-criteria method. Validation was performed by cross-referencing with an existing landslide inventory, which was made by field mapping and interpretation of satellite images. Finally, the relative risk was determined for the City of Valjevo by using a realistic population distribution model as a source for elements at risk. The results show the distribution of risk and suggest that 20% of the inhabited area falls into the high risk class, but this encompasses less than 5% of the total population.

Project BEWARE—Landslide Post-disaster Relief Activities for Local Communities in Serbia

The project—on harmonization of landslide data and training of municipalities for its monitoring, nicknamed BEWARE (BEyond landslide aWAREness) was implemented by the Geological Survey of Serbia, and the University of Belgrade Faculty of Mining and Geology. The Project partners were UNDP Office in Serbia, Ministry of Mining and Energy and Government Office for Reconstruction and Flood Relief of the Republic of Serbia. Project was funded by People of Japan. Overall aim of BEWARE project was to standardize post-event landslide database and closely involve local community of 27 municipalities affected by May 2014 flooding and landslides episode in Serbia, and prepare them to cope with catastrophic events in the future. In this paper we are presenting main BEWARE project activities and results implemented on local communities in Serbia after May 2014 event.

Massive Landsliding in Serbia Following Cyclone Tamara in May 2014 (IPL-210)Open image in new window

The IPL project No 210, titled “Massive landsliding in Serbia following Cyclone Tamara in May 2014”, started in March 2016. The study area is located in the Western and Central part of the Republic of Serbia territory affected by Cyclone Tamara in May 2014. The project aims to summarize and analyse all collected relevant data, including historic and current rainfall, landslide records, aftermath reports, and environmental features datasets from the May 2014 sequence. Objectives of the proposed project include: collecting all available and acquired landslide data, analysing the trigger/landslide relation in a feasible time span and in the May 2014 event, relating the landslide mechanisms and magnitudes versus the trigger, identifying spatial patterns and relationships between landslides and geological and environmental controls, proposing an overview susceptibility map of the event and numerical modelling of the site-specific location and landslide mechanisms. The Project will be organized by University of Belgrade, Faculty of Mining and Geology and Faculty of Civil Engineering. Project beneficiaries are local community and local and regional authorities. In this paper we will present preliminary results of the proposed project targets performed by project participants.