Landsifier 2.1.0

A library for predicting the classes of landslides based on their topography

Landsifier

A library for predicting the classes of landslides based on their topography using topological data analysis.

Installation

Windows (Recommended)

For Windows users, we recommend using Conda to avoid compilation issues with GDAL:

conda create -n landsifier python=3.11
conda activate landsifier
conda install -c conda-forge geopandas
pip install landsifier

Linux/macOS

Standard pip installation works on Linux and macOS:

pip install landsifier

Verification

After installation, verify the package works:

import landsifier
print(landsifier.__version__)

System Requirements

• GDAL: Required for geospatial operations. On Windows, install via conda install -c conda-forge gdal. On Linux/macOS, pip will install it automatically through geopandas.
• Python 3.8-3.11

Quick Start

Landsifier provides three main functions for the complete workflow:

1. Extract Features

from landsifier import extract_features

# Extract topological features from a landslide inventory
features = extract_features(
    shp_path="path/to/inventory.shp",
    dem_location="path/to/dem.tif",
    use_existing_dem=True,
    include_interior_points=False,
    label_column="Type"
)

Parameters:

• shp_path: Path to the landslide inventory shapefile
• dem_location: Path to the Digital Elevation Model (DEM) file
• use_existing_dem: If True, use the provided DEM. If False, download a new DEM
• include_interior_points: If True, include interior polygon points in feature extraction
• label_column: Optional column name for landslide type labels
• output_dir: Optional directory to save feature files
• min_area_threshold: Minimum polygon area in m² (default: 500.0)

2. Train Classifier

from landsifier import train_classifier

# Train a Random Forest classifier on extracted features
trained_models = train_classifier(
    path="path/to/directory/with/feature_files",
    top_k=8  # Number of top features to use
)

Parameters:

• path: Directory containing .npy feature files (one per class)
• top_k: Number of top features to select (default: 8)

Returns: A dictionary containing trained models and evaluation metrics.

3. Predict Inventory

from landsifier import predict_inventory

# Predict landslide types for new inventory
predicted_gdf = predict_inventory(
    model_path="path/to/trained_model.pkl",
    shp_path="path/to/new_inventory.shp",
    dem_location="path/to/dem.tif",
    use_existing_dem=True,
    include_interior_points=False
)

Parameters:

• model_path: Path to the pickled trained model file
• shp_path: Path to the landslide inventory shapefile to predict
• dem_location: Path to the Digital Elevation Model (DEM) file
• use_existing_dem: If True, use the provided DEM. If False, download a new DEM
• include_interior_points: If True, include interior polygon points
• optimized_features_path: Optional path to CSV with selected feature indices
• min_area_threshold: Minimum polygon area in m² (default: 500.0)

Returns: A GeoDataFrame with predicted landslide type labels.

Complete Workflow Example

from landsifier import extract_features, train_classifier, predict_inventory

# 1. Extract features from labeled training data
extract_features(
    shp_path="training.shp",
    dem_location="dem.tif",
    use_existing_dem=True,
    include_interior_points=False,
    label_column="Type",
    output_dir="features"
)

# 2. Train classifier
train_classifier(path="features")

# 3. Predict on new data
predictions = predict_inventory(
    model_path="features/Model_All_Features.pkl",
    shp_path="new_data.shp",
    dem_location="new_dem.tif",
    use_existing_dem=True,
    include_interior_points=False
)