An analysis of Alberta Geological Survey`s Map605 – Relative Landslide Susceptibility Model of the Alberta Plains and Shield Regions
The long-term and stable returns provided by the infrastructure assets are under increasing pressure. As the global economy adapts to physical changes, infrastructure investments face new levels of disruption and loss. Physical risks regarding climate change are becoming an essential risk category for infrastructure operators and owners. Natural disasters are a significant cause of infrastructure disruptions, and climate change is expected to intensify these disruptions.
Landslides are a significant geohazard. Socioeconomic impacts of landslides are always underestimated as these are mostly taken as a consequence of other triggering processes.
The term ‘landslide’ is defined as:
“The mass movement of debris, rock, or earth down a slope. It is identified by the movement itself and the resultant landforms. However, collapsed structures and skin holes are not considered landslides.”
The term ‘susceptibility’ refers to:
“the state or quality of being susceptible. It’s the lack of ability to resist some extraneous agent.”
A ‘landslide susceptibility map’ refers to a geographic representation of areas susceptible to landslides classified by intensity, from low susceptibility to high susceptibility. It takes into account where the landslides occur and the causes behind them.
Landslides in Alberta are a common phenomenon causing significant damage to property and people. Therefore, having an awareness of, and general understanding of the spatial distribution landslide susceptible areas, is a vital piece of information needed to be added to an effective infrastructure asset management plan. Landslide Risk Management is a series of events leading to landslide risk reduction. It includes landslide monitoring, engineering works, landslide forecast, insurance, slope strengthening, and more.
Risk Management refers to:
“The problems of landslide and geological risk management are seen as a series of events that lead to risk reduction, including risk assessment, risk analysis, vulnerability evaluation, risk mapping, the concept of acceptable risks, monitoring organizations, and more.”
To minimize the risk and damage due to landslides, it’s necessary to assess the factors responsible for the landslides. This article will highlight the importance of managing risks and assets. We will review and assess Map 605 [1], a relative landslide susceptibility model of the Alberta Plains and Shield Regions, and how the Bantu Khaya Group extracted landslide susceptibility geospatial data
Map 605 Alberta Geological Survey_ Relative Landslide Susceptibility Model of the Alberta Plains and Shield Regions
Alberta Geological Survey made a Map 605 based on the relative landslide susceptibility model of the Alberta plains and shield region. This map represented a predictive statistical model of the landslide susceptibility of the Alberta portion of Interior Plains and Canadian Shield. The model predicts the degree to which terrain can be impacted by landslides depending on a statistical procedure. It helped establish a relationship between the spatial distribution of recognized landslides and predisposing topographic, geological, and climate factors.
From the documented abstract of the publication found at https://ags.aer.ca/publication/map-605 The model results portray the spatial distribution of landslide susceptibility as a relative ranking from low to high. It is intended to be used for educational purposes and regional-scale planning initiatives. Importantly, it does not evaluate the probability of landslide occurrence over any specific period of time, nor does it evaluate the magnitude or impact of any potential landslide activity. Consequently, the model results should not be interpreted for the purpose of site-specific landslide identification, landslide activity assessment, or landslide hazard appraisal. [1]
The analysis of this survey was intended to predict the landslide susceptibility on natural sloped rather than engineered slopes and they go on to note that the evaluation methodology used for this model does not assess the effect of significantly folded and faulted bedrock structure, and therefore does not include the mountains and foothills physiographic regions, where bedrock structure is the main geologic control on landslide susceptibility. While bedrock structure within the Shield is complex, it is not considered a significant geologic control on landslide susceptibility due to the high strength of bedrock in that region; therefore, the Shield region has been included in the model.
The survey took examples, the Alberta plains’ landslides, which typically develop within the landslide-susceptible terrain of the Alberta Plains and are recognized in this model. You can see these examples with proper graphs and images displayed from a hill-shaded bare-earth Light Detection and Ranging (LiDAR) digital evaluation model.
Methodology
The landslide susceptibility model used a point-based sampling of landslide and non-landslide terrain. Information related to landslide distribution was merged from previously published surficial geology maps, university thesis, and reports. These data were augmented by sampling new landslide features mapped from various aerial imagery data sources, including LiDAR.
Landslides Predisposing Factors
Landslide susceptibility modeling was performed by evaluating the spatial likelihood of landslide occurrence on a cell-by-cell basis. A grid-cell resolution of 90m was found to optimize the model performance. In a descriptive analysis, a wide range of landslide predisposing factors were assessed for their capacity to predictively model the landslide distribution in the inventory data. Here are the predisposing factors extracting the most substantial influence on the landslide susceptibility.
- Local terrain morphology
- Regional terrain morphology
- Topographic wetness
- Physiography and climate
- Surficial geology
- Bedrock geology
Modelling Procedure
A predictive modeling method known as Stochastic Gradient Boosting was used to evaluate landslide susceptibility. This method leverages a decision-tree structure to map the occurrence of landslides related to the threshold in the predisposing factors using the hierarchy of branches and splits. Data from the landslide inventory represent the landslide cells used to train the model.
The non-landslide cells were obtained using simple random sampling of the background physiographic, geological, and climate conditions. The landslide susceptibility estimation represents the membership probability in either the landslide on non-landslide classes.
Unlike a single decision tree, the Stochastic Gradient Boosting algorithm optimizes the prediction accuracy based on an additive process. At each iteration, the algorithm determines the gradient that needs to optimize the modeled fit to the data and chooses a specific model in most agreement with the direction. The final model represents a weighted average of the decision-tree ensemble.
Model Uncertainty and Variability
Model accuracy was evaluated using a bootstrapping process with an ensemble of 20 model replications. These models were constructed by randomly sampling the landslide and non-landslide grid cells. For each model replication, 75% of the mapped landslide cells were drawn randomly from the total population and used to train the model. However, the remaining 25% of landslide cells were used to validate the prediction accuracy.
The final susceptibility map represents the mean of 20 replicate models. Regions with the lowest uncertainty, less than 5%, occur in the plains and lowlands of the province or deeply incised valleys. And the regions with the highest uncertainty, up to ~25%, occur in some regional uplands, including the Porcupine hills.
Relative Landslide Susceptibility of the Alberta Plains
The Stochastic Gradient Boosting model indicated the landslide susceptibility across the Alberta Plains is generally associated with areas of higher relief such as flans of plateaus and uplands and valley walls. Lower relief areas such as lowlands and plains, floodplains, and broad river terraces are usually less susceptible.
The walls of major river valleys and their tributaries comprise the most extended contiguous zones of landslide-susceptible terrain across Alberta. These zones are relatively narrow but can extend along one or both valley walls for 10s to 100s kilometers. Wider zones of landslide susceptible terrain occur within the western part at up to 250m.
Widespread zones of the landslide-susceptible terrain occur along steep slopes flanking relatively un-dissected plateaus. These zones are probably 10s of kilometers long, 6km wide, and up to 500m in height. Less contiguous zones of landslide-susceptible terrain occur across highly dissected plateaus or rugged uplands
BKG`s Geospatial data extraction methodology
Bantu Khaya Group`s team of Geomatics Engineering and Geospatial Intelligence experts developed a solution to analyze the metadata and geo-referenced files provided on the Alberta Geological Survey / Alberta Energy Regulator website, as part of the Map605 documentation. After analyzing the map and noticing that the model results portray the spatial distribution of the landslide susceptibility as a relative ranking from low to high, the team came up with a geospatial data processing solution to classify the geo data into spatially distinctive layers identifying the level of “susceptibility” to landslides using a combination of CAD and different GIS softwares.
- We analyzed all files, maps, data and metadata from the Map605 publication package.
- We then subsequently formulated an execution methodology to commence the data extraction, which started off with the uploading of the geo-referenced geo-tiff map to our own system.
- We then ran multiple geo-processing functions and analyzed the landslide susceptibility levels to range from 0.00 to 1.00.
- We then classified the data by creating quartile ranges to separate the landslide susceptibility data and extracted four ranges to represent.
- Medium landslide susceptibility level
- High landslide susceptibility
- Very high landslide susceptibility
- Export the created layers into the CAD environment for editing and hatching.
- Import edited layers into the GIS environment and st coordinate system to NAD83-Alberta-10-TM.
- Export layers as Geo-referenced Shapefiles in the GIS software.
Quality Control & Verification of produced vector layers
The share files created were overlaid on the Map 605 to compare and check.
Where in Alberta?
Map 605’s predictive analysis analyzes the probability and distribution of landslide occurrence using machine learning techniques. It shows the regional pattern of landslide-prone areas across the Alberta plain. Our team at Bantu Khaya Group created a web map showing extracted data and converted it into useful information.
We then created a search functionality for users to enter and find the locations on the Alberta map to see medium, high, and very high risks to see where your points of interest lie.
Enter the address or geographic coordinates on the Bantu Khaya Group Web Map to determine where your assets lie.
Map link https://bantukhaya.group/maps/albertalandslidesusceptibility/
Conclusion
Yes, environmental hazards such as landslides cause significant effects to infrastructure assets .We cannot stress enough the importance of having an awareness of the spatial distribution of high risk areas as this aids in conducting targeted asset monitoring and management activities. We also recommend the incorporation of innovative data collection methodologies to enhance the quality of data and models produced from that dataset, which would be incorporated into any infrastructure Risk Management Matrix to paint a full picture to all stakeholders including owners/decision makers, as well as regulatory agencies
References
[1] Pawley, S.M., Hartman, G.M.D. and Chao, D.K. (2016): Relative landslide susceptibility model of the Alberta Plains and shield regions; Alberta Energy Regulator, AER/AGS Map 605, scale 1:1 000 000
https://bantukhaya.group/maps/albertalandslidesusceptibility/