The importance of input data on landslide susceptibility mapping

Area of Science:

• Geosciences
• Remote Sensing
• Geographic Information Systems (GIS)

Background:

• Landslide detection and susceptibility mapping are vital for risk management and urban planning.
• Advancements in digital elevation models (DEMs) and automatic detection necessitate evaluating input data effects on map accuracy.
• Assessing the influence of data variations is crucial for reliable landslide susceptibility modeling.

Purpose of the Study:

• To evaluate the influence of input data variations on landslide susceptibility mapping accuracy.
• To compare models using manual versus automatic landslide inventories.
• To assess the impact of topographic data resolution, number of causative factors, and sampling techniques.

Main Methods:

• Developed 32 logistic regression models varying landslide inventory type (manual/automatic), topographic data resolution, number of landslide-causing factors, and sampling technique.
• Utilized digital elevation models and various geospatial datasets.
• Employed statistical analysis to compare model performance.

Main Results:

• Models using automatic landslide inventories showed comparable accuracy to those with manual inventories.
• Finer topographic data resolution significantly improved susceptibility model accuracy and precision.
• The number of causative factors impacted lower-resolution data more; high-resolution data yielded accurate maps even with fewer factors.
• Sampling from landslide masses outperformed sampling from mass centers.

Conclusions:

• Most landslide susceptibility models demonstrated reasonable prediction accuracy, highlighting the importance of data quality.
• Automatic landslide detection is a viable alternative to manual methods.
• High-resolution topographic data is key for accurate landslide susceptibility mapping.
• The choice of input data and techniques should align with data availability and study objectives for effective risk assessment.