Skip to main content

How to use drone data to predict feature density across broader areas

Use Predict Density to turn detailed drone observations and classifications into landscape-scale coverage estimates with Sentinel-2 data

Written by Karen Joyce

Who can use this feature

Editors of a project within a Professional or Pro + workspace.

Selection requirements for tool to work

A drone data classification OR

A polygon layer


AND

A Sentinel-2 image over the same area

TEMPORARILY UNAVAILABLE

Predict Density helps you quickly understand how much of a feature (e.g. woody vegetation, bare ground) exists across a much larger area than your drone map alone can cover. By combining the fine detail from your drone classification (or any polygon layer) with broader-scale Sentinel-2 satellite imagery, the tool lets you “scale up” what you see locally to the surrounding landscape.

To make sure the satellite data are clean and reliable, the tool automatically removes any Sentinel-2 pixels affected by clouds, shadows, snow, or other poor-quality conditions. This means the predictions are made only from good-quality satellite observations, giving you more trustworthy results.

Once both datasets are aligned, the tool measures how much of your chosen feature sits inside each area equivalent to a Sentinel-2 pixel. These real measurements are used to train a Random Forest regression model (a type of machine-learning algorithm) to learn the relationship between your high-resolution drone data and the coarser satellite data.

The end result is a map predicting the density or percent coverage of your feature across the entire Sentinel-2 scene. This allows you to extend your detailed drone insights to much larger regions, saving time, reducing field effort, and giving you a scalable way to monitor change, support reporting, or guide management decisions.

  1. Create a classification with your drone data, or use the polygon tool to digitise features of interest (e.g. woody vegetation)

  2. Stream a Sentinel-2 image covering the same area into your project

  3. Select a category within your drone classification OR select your polygon layer

  4. Use CTRL or CMD to multi-select your Sentinel-2 scene

  5. From the top menu bar, click the Predict density tool

Top tips

  • Make sure that you choose a cloud-free Sentinel-2 image

  • Choose a Sentinel-2 image as close in time to the drone data as possible

  • Spend time getting your input classification or polygon layer as accurate as possible. The prediction can only be as accurate as the input, and errors will be amplified.

Known errors

  • Input extent is too large - please keep the area covered by your input polygons <300ha

  • Insufficient overlap between input training data and Sentinel-2 image - make sure that these cover the same area

  • If the overlapping area in the Sentinel-2 image is covered in cloud or cloud shadow, it will not return a result. Make sure that the satellite image you choose is cloud free.

Did this answer your question?