Search

ⓘ Info

How It Works

The map uses logistic regression to estimate how likely a landslide is to occur at any location in Santa Barbara County. The model was trained on 926 confirmed landslide locations and ~4,600 comparison points, placed at least 200 meters away from any known landslide location. It uses eight terrain, climate, and land condition factors to score each pixel across the county, then groups those scores into five classes, from Very Low to Very High, based on where they fall in the county-wide distribution.

Model Factors

Factor — Data Source — LR Influence

• Slope — USGS 3DEP 10-m elevation data — 27.5%
• Precipitation — NOAA Atlas 14 100-year/24-hour storm intensity — 22.4%
• Land Cover — USGS GAP/LANDFIRE 2011 ecosystem classification — 17.1%
• Terrain Curvature — USGS 3DEP 10-m elevation data — 9.3%
• Burn Severity — CALFIRE perimeters, weighted by recency (3-year decay) — 9.1%
• Lithology — USGS State Geologic Map Compilation — 7.6%
• Soil Erodibility — USDA NRCS gSSURGO soil survey — 6.7%
• Vegetation Density (Normalized Difference Vegetation Index, or NDVI) — Sentinel-2 satellite imagery — <1%

Accuracy

Cross-validation AUC: 0.719 (1.0 = perfect, 0.5 = random). The wide variation across spatial blocks (±0.256) reflects the fact that most landslides are clustered in the Santa Ynez Mountains, making performance uneven across the county.
A random 20% of training points were held out for testing: 49.7% fell in the High or Very High class (n = 185).
For independent validation, 8,323 landslide points from the January 2023 Santa Ynez atmospheric river storm (not used in training) were compared, where 92.8% fell in High or Very High.

Training Data

• USGS National Landslide Inventory v3 — 926 recorded landslide locations with Likely or High confidence ratings

Independent Validation

• USGS 2023 Santa Ynez Mountains Inventory — 8,323 landslide points from the January 9, 2023 storm event. Thomas et al., 2025, USGS data release. Not used in training.

Map Layers

• County Boundary — U.S. Census Bureau TIGER/Line
• Fire Perimeters — CAL FIRE, 2016–present
• Fault Lines — USGS Quaternary Fault and Fold Database
• Geology — USGS State Geologic Map Compilation, supplemented by Macrostrat
• Recorded Historical Landslides — USGS National Landslide Inventory v3 (1,043 deposits, all confidence levels)
• NLI Training Points — The 926 landslide locations used to train this model

Limitations

This is a broad screening tool, not a prediction of specific landslide events. Most training points are concentrated in the Santa Ynez Mountains because that is where landslide mapping has been done — not because other areas are definitively safe. The model reflects long-term terrain conditions and does not update in real time with rainfall, soil moisture, or new fire activity. Do not use this map as a substitute for on-the-ground engineering assessment.

To view the code for this project, visit the GitHub Repository.

© Ryan Green, 2026

Landslide Risk Assessment Tool for Santa Barbara County

This tool models landslide susceptibility across Santa Barbara County using logistic regression trained on recorded landslide locations and eight geospatial factors. Click the Info button for methods and data sources.