Geospatial machine learning combines spatial data analysis with predictive modeling. By incorporating geographic features such as coordinates, distances, and environmental variables, machine learning models can uncover spatial patterns and make predictions about geographic phenomena.

Applications include:

  • Species distribution modeling
  • Land cover classification
  • Climate prediction
  • Urban growth modeling
  • Environmental risk analysis

This guide introduces workflows for building spatial machine learning models using Python.

Key Libraries

Common Python libraries used for geospatial machine learning:

Library Purpose
geopandas Vector spatial data
rasterio Raster data processing
numpy Numerical operations
pandas Tabular data manipulation
scikit-learn Machine learning models
xarray Multidimensional geospatial data

Install required packages:

pip install geopandas rasterio scikit-learn xarray numpy pandas

Import libraries:

import geopandas as gpd
import pandas as pd
import numpy as np
from sklearn.model_selection import train_test_split
from sklearn.ensemble import RandomForestClassifier

Spatial Features in Machine Learning

Spatial models often use features such as:

  • Latitude and longitude
  • Distance to features
  • Elevation
  • Climate variables
  • Land cover classifications

Example dataset:

latitude longitude elevation temperature species_presence
40.58 -105.08 1500 12.4 1

Preparing Spatial Data

Load spatial dataset:

gdf = gpd.read_file("species_observations.shp")

Extract coordinates:

gdf["lon"] = gdf.geometry.x
gdf["lat"] = gdf.geometry.y

Convert to machine learning table:

df = pd.DataFrame(gdf.drop(columns="geometry"))

Feature Engineering

Create spatial features to improve model performance.

Example: distance to nearest river.

gdf["distance_to_river"] = gdf.geometry.distance(river_geometry)

Example: spatial clustering features.

from sklearn.cluster import KMeans

coords = df[["lat","lon"]]
kmeans = KMeans(n_clusters=5)
df["region_cluster"] = kmeans.fit_predict(coords)

Splitting Data

Split dataset into training and testing sets:

X = df[["lat","lon","elevation","temperature"]]
y = df["species_presence"]

X_train, X_test, y_train, y_test = train_test_split(
    X, y, test_size=0.2, random_state=42
)

Training a Model

Train a Random Forest classifier:

model = RandomForestClassifier(n_estimators=200)

model.fit(X_train, y_train)

Make predictions:

predictions = model.predict(X_test)

Evaluating Model Performance

Measure accuracy:

from sklearn.metrics import accuracy_score

accuracy = accuracy_score(y_test, predictions)
print(accuracy)

Other useful metrics include:

  • Precision
  • Recall
  • F1-score
  • ROC-AUC

Spatial Prediction

Use trained models to predict across geographic space.

Example workflow:

  1. Generate prediction grid
  2. Calculate spatial features
  3. Apply trained model

Example:

prediction_grid["predicted_presence"] = model.predict(grid_features)

These predictions can be visualized as maps.

Mapping Machine Learning Results

Convert predictions to a GeoDataFrame:

results = gpd.GeoDataFrame(prediction_grid, geometry="geometry")

Plot prediction map:

results.plot(column="predicted_presence", legend=True)

Spatial Cross Validation

Standard machine learning validation may produce biased results because nearby observations are similar.

Spatial cross-validation splits data by geographic region instead of random sampling.

Example strategy:

  • Divide study area into spatial blocks
  • Train model on some regions
  • Test on withheld regions

This produces more realistic accuracy estimates.

Working with Raster Features

Machine learning models often use raster datasets such as:

  • Elevation
  • Land cover
  • Climate variables
  • Satellite imagery

Extract raster values at observation locations:

import rasterio

with rasterio.open("elevation.tif") as src:
    gdf["elevation"] = [
        x[0] for x in src.sample(
            [(geom.x, geom.y) for geom in gdf.geometry]
        )
    ]

Example Workflow

A simplified spatial ML workflow:

import geopandas as gpd
import pandas as pd
from sklearn.ensemble import RandomForestClassifier

data = gpd.read_file("observations.shp")

data["lon"] = data.geometry.x
data["lat"] = data.geometry.y

X = data[["lat","lon","elevation"]]
y = data["target"]

model = RandomForestClassifier()
model.fit(X,y)

Summary

Geospatial machine learning combines:

  • Geographic data
  • Environmental variables
  • Machine learning models

These approaches allow researchers to:

  • Predict environmental phenomena
  • Model spatial processes
  • Analyze geographic patterns

Python provides a flexible ecosystem for building advanced geospatial models.

Next Steps

Advanced topics to explore include:

  • Spatial deep learning
  • Remote sensing classification
  • Spatial autocorrelation modeling
  • Geostatistics and kriging
  • Spatiotemporal forecasting
« Previous: Geospatial Data Analysis Intermediate Next: Satellite Remote Sensing »