Coral-cover change – concept
North Point, Lizard Island · aerial drone, 19 Mar 2024 (bleaching)
Concept demonstration

Measuring coral cover change
from drones to satellite

Lizard Island, Great Barrier Reef · North Point & Palfrey reefs

How surface-drone and aerial-drone surveys can quantify coral cover today – and how that might calibrate a Sentinel-2 product to estimate coral cover across an entire reef, refreshing far more often than any field program could.

Tier 1 · on-water
Surface drone
Millimeter resolution orthomosaics, sub hectares per deployment.
Tier 2 · aerial
Aerial drone
Sub-centimetre resolution orthomosaics; hectares per flight.
Tier 3 · satellite
Sentinel-2 · 10 m
Whole reef, every few days, free and ongoing.
Scroll through the concept
1 The problem

No field program can watch a whole reef often enough

Coral cover can change quickly – a bleaching summer can undo a decade of growth in weeks. To catch that at reef scale you need both wide coverage and frequent revisits. A single data source can never give you both.

Wide but coarse & stale

Existing satellite habitat maps cover everything but are categorical, often years out of date, and of limited accuracy for cover.

Precise but tiny

In-water transects and quadrats are accurate, but cover a few hundred square metres – and often lack precise spatially referenced data.

The gap

Managers need a quantitative, current, reef-wide read on coral cover to decide where to look harder and act. That layer doesn't exist yet.

Three ways we can see the same coral

Each step trades detail for reach. The aerial and surface drones see individual colonies; the satellite sees the whole reef in coarse pixels. The concept is to let the fine data teach the coarse.

Surface vessel imagery of coral at North Point
Tier 1 · on-water
Surface drone
Millimeter scale imagery of the reef, captured by an autonomous robot at high tide and accurately georeferenced.
~1 mm/pixel
Aerial drone orthomosaic of coral at North Point
Tier 2 · aerial
Aerial drone
Sub-centimetre orthomosaics covering hectares of exposed reef in a single flight.
~0.4 cm/pixel
Sentinel-2 satellite view of North Point
Tier 3 · satellite
Sentinel-2
Coarse 10 m pixels, but the whole reef at once, free, every few days.
10 m/pixel
2 The data we already have

The drone can already capture coral cover – and it caught a catastrophe

At Lizard Island in 2024 we mapped North Point and Palfrey reefs at low tide, during and after the mass bleaching. From the orthomosaics we digitised live and bleached coral inside twenty 10 × 10 m quadrats.

Measured – Raoult et al. 2025, Coral Reefs
96.9%
of living coral cover was affected by bleaching across the two reefs (±2.0% SD).
92.2%
mean bleaching mortality – among the highest rates ever recorded (±6.3% SD).
7.5%
mean surviving coral cover of quadrat area afterwards (±6.1% SD; max 23.9%).
North Point reef – drone orthomosaics, during vs after bleaching Drag the divider to compare
Left: 19 Mar 2024, during bleaching (bright/white coral). Right: 4 Jun 2024, after – most coral is dead and darkening with algae. Data captured by CSIRO, Macquarie University, University of Newcastle, and GeoNadir.
3 The bridge – from drone to satellite

Teaching a satellite pixel what the drone already knows

Using data we've already captured, here's where we would start. Lay the Sentinel-2 grid – one pixel is 10 × 10 m – over the drone orthomosaic where we measure coral cover. Each pixel now has two things: a coral-cover % from the drone, and a spectral signature from the satellite. Enough calibration points, and the relationship between them lets a satellite estimate cover where no drone has flown.

Concept only, not a result
The relationship we'd build: drone cover % vs satellite signal

Every drone survey adds calibration points to this cloud. The stronger and more transferable the relationship, the more confidently a satellite pixel alone can be read as coral cover – the whole premise of scaling up.

Illustrative fit · R² ≈ –

Read that relationship in reverse and every Sentinel-2 pixel becomes a coral-cover estimate. Here is what it looks like laid back over the reef, one 10 m pixel at a time.

North Point – drone orthomosaic under a 10 m Sentinel-2 grid Hover a pixel to read its coral cover
Coral cover in pixel
0%20%40%+
Each cell is a single Sentinel-2 pixel over the real June 2024 orthomosaic. Cover values shown here are illustrative – the workflow is real, the per-pixel numbers are placeholders until the calibration study is done.

This isn't pie in the sky – we've done it before

We have already used this exact drone-to-satellite calibration approach to map seagrass density at Warrior Reef in the Torres Strait. There it gave a more accurate, quantitative representation of the benthic habitat than categorical mapping – and it significantly improved downstream species-distribution models built on top of it.

Calibrating a fine-scale measurement to satellite bands is also something we do routinely in terrestrial applications such as vegetation density mapping. Coral cover is a new application of an established method, not an unproven idea.

4 The pay-off – the whole reef

Two drone patches, calibrating an entire island reef

Apply that relationship to every Sentinel-2 pixel and the reef complex lights up as a continuous coral-cover surface – quantitative percentage, not habitat classes – that can be re-made every time a fresh, cloud-free satellite pass comes in. Toggle the two dates to see how a reef-wide read on change would look.

Concept mockup – a simulated cover surface, not a satellite retrieval
Lizard Island reef complex – modelled coral cover from Sentinel-2
Modelled coral cover
0%20%40%+
Teal square = North Point calibration site
Basemap: Sentinel-2 cloudless (EOX). The two teal outlines are the actual drone footprints – everything else is estimated from them in this concept. In reality this is exactly where the research question sits: how far a local calibration can be trusted across the reef.
5 Turning cover into decisions

You set the threshold – the system tells you where to look

A reef-wide cover layer only earns its keep if it drives action. Here you choose a coral-cover threshold, and the concept flags the reef areas that fall below it and recommends sending a drone or on-water team to verify or trigger management action.

Management view – reef below your alert threshold
Coral-cover alert threshold
10% cover
0 ha of reef below threshold – flagged per 10 m Sentinel pixel, grouped into priority areas for revisit.
Flagging runs on the 10 m Sentinel-2 grid; cover values are illustrative. The loop this closes is the real point: satellite flags a hotspot → a drone or blueboat is tasked to verify → that new survey re-calibrates the model. Concept only.

Note

  • Real: the drone and on-water surveys, the orthomosaics, and the Lizard Island bleaching & mortality figures (Raoult et al. 2025). We can, and do measure coral cover from this imagery.
  • Unproven research: that a drone-to-Sentinel-2 cover relationship will hold accurately across a whole reef, or transfer between reefs. We recommend conducting this research.
  • Illustrative only: every satellite-derived cover value, the reef-complex surfaces, the zone flags, and the scatter fit on this page. They show what the system could look like, not a result.
  • Nothing here says this is easy or certain. It is a concept to decide whether the research is worth funding.
6 Where this goes

A small, honest first step – with a reef-wide horizon

The ask is deliberately modest: prove the relationship at one well-understood site with substantial data before promising anything larger.

Fund now · MVP

Build & test the relationship at Lizard

  1. Manually label coral cover across the North Point and Palfrey orthomosaics – beyond the published quadrats.
  2. Extract Sentinel-2 signatures for the matching pixels and dates already in hand.
  3. Fit and test the drone-to-satellite cover relationship; quantify how well it holds and where it breaks.
  4. Deliver a first reef-wide cover surface for Lizard, with uncertainty.
If it works · the horizon

Scale the calibration across the network

250+reefs visited each year – every drone & on-water survey becomes a calibration point
  1. Ingest incoming drone and blueboat surveys automatically as they're captured.
  2. Continuously re-calibrate the cover model as coverage and conditions grow.
  3. Train detection (AI segmentation of live coral) to remove the manual labelling step.
  4. Operate a living, quantitative coral-cover layer that tells managers where to act – and tasks the drones that keep it honest.