Google Earth Engine Tutorial: Estimate Tree Cover with Regression

Introduction to Tree Cover Estimation with Regression in Google Earth Engine

Tree cover estimation using regression in Google Earth Engine offers a powerful method to quantify forested areas across large regions. By leveraging remote sensing data and statistical modeling, you can predict tree cover density and analyze spatial patterns. This tutorial guides you through the process using Landsat imagery and a regression-based approach.

Step 1: Load and Filter Satellite Data

Start by importing the Landsat dataset and filtering it to a specific region and date range. For example:

var dataset = ee.ImageCollection('LANDSAT/LC08/C02/T1_L2')
.filterDate('2020-01-01', '2020-12-31')
.filter(ee.Filter.geometry(ee.Geometry.Rectangle([lon1, lat1, lon2, lat2])));

Replace lon1, lat1, lon2, lat2 with your area of interest coordinates.

Step 2: Preprocess the Data

Apply cloud masking and create a composite image. Use the following code:

function maskClouds(image) {
var cloudScore = ee.Algorithms.Landsat.simpleCloudScore(image);
return image.updateMask(cloudScore.select('cloud').lt(50));
}
var filtered = dataset.map(maskClouds);
var composite = filtered.median();

Step 3: Extract Tree Cover Information

Use spectral indices like NDVI (Normalized Difference Vegetation Index) to identify tree cover. Calculate NDVI from the Landsat bands:

var ndvi = composite.normalizedDifference(['SR_B6', 'SR_B5']);
var treeCover = ndvi.multiply(10000).gt(4000); // Adjust threshold based on your region

This creates a binary map where treeCover is true for pixels with high vegetation density.

Step 4: Build a Regression Model

Create training data using known tree cover locations. For example, use a vector dataset with field sample points:

var training = treeCover.sample({
region: ee.Geometry.Rectangle([lon1, lat1, lon2, lat2]),
scale: 30,
numPixels: 1000,
tileScale: 4
});
var model = ee.Model.linearRegression({
dependent: 'NDVI_value',
independent: ['SR_B4', 'SR_B5', 'SR_B6']
});
var trainedModel = model.train(training);

Adjust the independent variables based on the features most correlated with tree cover.

Step 5: Apply the Model to the Image

Use the trained model to estimate tree cover across the entire region:

var prediction = composite.select(['SR_B4', 'SR_B5', 'SR_B6']).evaluate(trainedModel);
var treeCoverPrediction = prediction.select('predicted');

Step 6: Visualize and Export Results

Add the tree cover prediction to the map and export as a GeoTIFF for further analysis:

Map.addLayer(treeCoverPrediction, {min: 0, max: 100}, 'Tree Cover Estimate');
Export.image.toDrive({
image: treeCoverPrediction,
description: 'tree_cover_export',
folder: 'GEE_Exports',
fileNamePrefix: 'tree_cover',
scale: 30,
maxPixels: 1e10
});

Conclusion

This workflow provides a scalable approach to estimate tree cover using regression. You can refine the model by incorporating additional data sources like canopy height measurements or LiDAR data for higher accuracy.