Design a landslide susceptibility map: geomorphology, triggers, inventories, and…
- User
Design a landslide susceptibility map: geomorphology, triggers, inventories, and validation.
Assistant
The user wants a design for a landslide susceptibility map. The request is structured around four key components:
- Geomorphology (The terrain factors).
- Triggers (The dynamic factors).
- Inventories (The historical data).
- Validation (Checking accuracy).
I need to structure the response as a comprehensive guide or a methodological framework for creating such a map. It shouldn't just be a definition; it needs to be actionable, explaining how to incorporate these elements.
1. Introduction:
- Define what a landslide susceptibility map is (spatial probability).
- State the objective: Zonation of hazard.
2. Geomorphology (The Static/Preparatory Factors):
- This is the physical setting.
- Slope: Crucial. Gravity needs an angle.
- Aspect: Solar radiation, wind exposure, vegetation differences.
- Curvature: Concave/convex (flow convergence/divergence).
- Elevation/Relief: Energy potential.
- Lithology/Geology: Rock strength, permeability.
- Soil Type/Depth: Shear strength.
- Land Use/Land Cover (LULC): Vegetation stabilizes; roads destabilize.
- Structural features: Faults, bedding planes.
- Hydrology: Distance to drainage, drainage density.
3. Triggers (The Dynamic Factors):
- Susceptibility vs. Hazard. Susceptibility is usually static, but the design must consider what triggers landslides to weight factors correctly.
- Rainfall: Intensity-duration. Antecedent moisture.
- Seismicity: Peak Ground Acceleration (PGA), distance to epicenters.
- Anthropogenic: Slope cutting, loading, deforestation.
- Note: For a standard susceptibility map, these are often implicitly handled through the geomorphological response, but for a hazard map, they are explicit inputs. I must clarify that for susceptibility, we look at the "preconditions" for triggering.
4. Inventories (The Training Data):
- This is the backbone. Heuristic and statistical methods need past events.
- Sources: Satellite imagery (Pre/Post event), historical records, field mapping, LiDAR.
- Types: Scarp, deposition area, or runout path?
- Format: Polygon vs. Point data.
- Importance: The "Truth".
5. Methodology/Modeling (Connecting Inventories and Factors):
- How do we combine them?
- Heuristic/Knowledge-based: Weights of Evidence (expert opinion).
- Statistical/Data-driven:
- Bivariate (Frequency Ratio).
- Multivariate (Logistic Regression).
- Machine Learning: Random Forest, SVM, Neural Networks (modern approach).
- Deterministic/Physics-based: Factor of Safety (Slope stability models). Usually for local scales, not regional.
6. Validation (The Quality Check):
- How do we know it works?
- Confusion Matrix: True Positive, False Positive, etc.
- ROC Curve (Receiver Operating Characteristic): AUC (Area Under Curve). The standard metric.
- Success Rate vs. Prediction Rate:
- Success: Training data.
- Prediction: Testing data (split dataset 70/30 or 80/20).
- Spatial Cross-Validation: Ensure no spatial autocorrelation bias.
7. Structure of the Response:
- I will organize the response logically, following the user's prompt structure.
- I will use a hypothetical workflow or a "Design Framework" approach.
- Visual descriptions (tables/bullet points) will help.
Refining the content:
- Geomorphology: Emphasize DEM (Digital Elevation Model) derivatives.
- Triggers: Distinction between Susceptibility (where) vs. Hazard (when). I will focus on Susceptibility but mention triggers as context.
- Inventories: Mention the "Multi-temporal" aspect if possible.
- Validation: Be specific about the 80/20 split.
Drafting the Plan:
- Title: Design Framework for Landslide Susceptibility Mapping.
- Phase 1: Data Acquisition (Geomorphology & Inventories).
- Phase 2: Trigger Analysis (Context).
- Phase 3: Modeling.
- Phase 4: Validation.
- Phase 5: Output/Zonation.
Let's flesh out the "Triggers" part. Since the user asked for "Triggers" in the design of a susceptibility map, strictly speaking, triggers (temporal) are usually excluded from pure susceptibility maps (which are spatial). However, the proximity to trigger sources (like fault lines or river channels causing erosion) are susceptibility factors. I need to make this distinction clear. Rainfall is the trigger, but High Precipitation Zone can be a susceptibility factor.
Self-Correction during drafting:
- Make sure not to confuse "Hazard" (Probability x Time) with "Susceptibility" (Probability x Space). The user asked for Susceptibility.
- Ensure the distinction between heuristic (expert) and data-driven methods is clear.
Final Polish:
- Use professional terminology (e.g., "True Positive Rate", "Factor of Safety", "Curvature").
- Format with clear headings.
Let's write it out. Designing a landslide susceptibility map is a process of spatial modeling that estimates the probability of landslides occurring in a given area. It relies on the principle that "the past and present are keys to the future"—meaning future landslides will likely occur under the same geological, geomorphological, and hydrological conditions as past ones.
Here is a comprehensive design framework for creating a landslide susceptibility map, broken down into the four requested components.
1. Geomorphology: The Preparatory Factors (Static Variables)
Geomorphology deals with the form of the landscape. In a GIS (Geographic Information System) environment, these are usually raster layers derived primarily from a Digital Elevation Model (DEM).
- Slope Gradient: The most critical factor. Steeper slopes have greater shear stress.
- Design Input: Calculated in degrees.
- Aspect (Direction): Influences exposure to sunlight (drying effects), wind, and rainfall.
- Design Input: Categorized into 8 directions (N, NE, E, etc.) or flat.
- Curvature: Describes the shape of the slope.
- Plan Curvature: Convergence/divergence of water.
- Profile Curvature: Acceleration/deceleration of flow.
- Significance: Concave slopes tend to accumulate water and debris; convex slopes are drier but steeper.
- Relative Relief: The difference in elevation within a specific neighborhood.
- Significance: Indicates tectonic activity or erosion potential.
- Drainage Proximity: Distance to nearest stream channel.
- Significance: Riverbanks are prone to undercutting and saturation (toe erosion).
- Lithology (Geology): The physical properties of the rock and soil.
- Design Input: A geological map reclassified based on shear strength and permeability (e.g., hard granite vs. weak shale).
2. Triggers: The Dynamic Factors (Causative Mechanisms)
For a standard susceptibility map, triggers are often used to define the context or weighting of the static factors. However, if you are building a dynamic (near-real-time) hazard map, triggers become active inputs.
- Seismicity:
- Proxy: Distance to active fault lines.
- Weighting: Zones closer to faults are weighted higher in the model.
- Hydrology (Rainfall):
- Proxy: The Topographic Wetness Index (TWI). This is a static geomorphological derivation that predicts where water will accumulate during a rain event.
- Dynamic Input: If mapping specific "Hazard" rather than "Susceptibility," you overlay Isohyetal lines (rainfall intensity contours).
- Anthropogenic Factors:
- Land Use/Land Cover (LULC): Deforested areas or barren land are highly susceptible compared to dense forests.
- Road Proximity: Cut slopes created by road construction often fail due to over-steepening. A buffer zone (e.g., 50–100m) is usually created around roads.
3. Inventories: The Training Data
A susceptibility map is only as good as its inventory. This is the database of known past landslides used to "train" the model to recognize dangerous conditions.
- Data Sources:
- Remote Sensing: Analysis of pre- and post-event satellite imagery (Sentinel-2, Landsat) or high-resolution drone imagery.
- Historical Records: Government reports, newspaper archives, and maintenance logs.
- Field Mapping: Ground truthing to verify scarps and depositional areas.
- Database Structure:
- Type: Slide, Flow, Fall, Topple.
- Activity: Active, Dormant, or Reactivated.
- Geometry: Polygon (for large complex slides) or Point (for source location).
- The Golden Rule: For statistical modeling, you typically split the inventory: 70% of landslides are used to train the model, and 30% are withheld for validation.
4. Modeling & Validation: The Synthesis
Once you have the layers (Geomorphology/Triggers) and the points (Inventory), you must combine them.
A. The Modeling Approach
There are three main tiers of modeling complexity:
- Heuristic (Knowledge-based): An expert assigns weights to layers (e.g., "Slope is worth 40%, Lithology 30%"). Simple, but subjective.
- Statistical (Bivariate/Multivariate):
- Frequency Ratio (FR): Calculates the probability of a landslide occurring in a specific class (e.g., slopes >30°) based on the density of past landslides.
- Logistic Regression: Calculates the probability of a binary outcome (landslide / no landslide) based on multiple variables.
- Machine Learning (Advanced):
- Random Forest / SVM: Algorithms learn complex non-linear relationships between the inventory and the environmental layers.
B. Validation
Validation determines if the map is trustworthy. The standard metric is the ROC (Receiver Operating Characteristic) Curve.
- The AUC (Area Under Curve):
- You test the map against the 30% of the inventory that was not used to train the model.
- 0.5 AUC: The map is a random guess (useless).
- 0.7 - 0.8 AUC: Acceptable/Fair.
- 0.9 - 1.0 AUC: Excellent/Outstanding.
- Success Rate vs. Prediction Rate:
- Success Rate: How well the map classifies the training data (can be prone to overfitting).
- Prediction Rate: How well the map classifies the testing data (the true test of future prediction).
Summary Workflow
Step Action Tools Needed 1. Data Prep Acquire DEM, Geological maps, Satellite Imagery. ArcGIS, QGIS, Google Earth Engine 2. Inventory Map past landslide polygons/points. High-res imagery, Field GPS 3. Factor Prep Calculate Slope, Aspect, TWI, Distance to faults/roads. ArcGIS (Spatial Analyst), SAGA GIS 4. Analysis Apply Frequency Ratio or Machine Learning model. R Studio, Python (Scikit-learn), GIS 5. Validation Generate ROC Curve and Confusion Matrix. R, SPSS, Python 6. Output Reclassify continuous probability into zones (Low, Moderate, High, Very High). GIS Final Output: A color-coded map where Red represents High Susceptibility (High probability based on past conditions) and Green represents Low Susceptibility.