Teaching the Algorithm to See the Forest
A hands-on workshop brought deep learning into the field. Putting satellite image analysis directly in the hands of the conservationists who need it most.
From 770 kilometers above the equator, a Planet Scope satellite captures a scene that looks, at first glance, unremarkable: a mosaic of green, grey-green, and brown patches stitched across a hillside. But embedded within that image - invisible to any single pair of eyes, detectable only at scale - is the signature of deforestation in progress: a logging road pushed a hundred meters further, a patch of primary canopy turned to scrub.
The problem is not data. The world now produces more satellite imagery than conservation teams can possibly review. The problem is analysis, and that is the gap Conservation Mind set out to close.
"The forest doesn’t wait for the analysis to catch up."
The workshop brought together conservation practitioners and GIS professionals from HAkA Team to work through a complete pipeline: from raw satellite imagery to automatically classified land cover, using deep learning models trained and deployed directly inside ArcGIS. No specialist computing infrastructure required. No machine learning PhD prerequisite. Just the tools that land managers already have, and a methodology that puts detection in their hands
Participants learned to label training samples from multispectral imagery, fine-tune convolutional neural networks for custom land-cover classification, and interpret model outputs in the context of real management decisions. The focus throughout was deforestation detection: teaching models to distinguish intact primary forest from degraded land, secondary growth, and recent clearance, the distinctions that matter most for early warning.
Conservation Mind's role in this was as translator: between the world of machine learning research and the world of field conservation, where decisions are made on limited time and limited budgets, but where the cost of missing a deforestation event is measured in irreplaceable canopy.