Green Innovation Challenge 2026 · AI for Environmental Monitoring & Awareness

Ain Marsad عين مرصد

Are the mangroves we planted still alive? We check every creek on the UAE coast from space, year by year, for free.

"Ain Marsad" means "the watchful eye." It turns free Sentinel-2 satellite photos into a map that shows where mangroves grew, where they were lost, and where planting didn't survive.

Sentinel-2 · 10 m pixels Archive from 2017 Random forest classifier Runs on a laptop
Illustration of how the output looks. Not real satellite data. Each square stands for one 10 m × 10 m pixel.
The problem

A promise with two numbers, and we only measure one

The UAE has pledged to plant 100 million mangroves by 2030. Planting is easy to record. Someone is there, a photo is taken, and the number goes into a report. What happens to that seedling six months later usually isn't written down anywhere.

How many were planted?
Recorded today

Logged at the moment of planting, by the people doing it.

How many are still alive?
Not measured across the coast

Checked only in small sample plots. A creek that failed can go unnoticed if nobody surveyed it.

100 million

mangroves the UAE pledged to plant by 2030, announced at COP26.

Source 1
≈17,600 ha

of planted and mature mangroves in Abu Dhabi Emirate, storing roughly 3 million tonnes of carbon (EAD estimate).

Source 2
+29%

UAE mangrove area, 7,081 ha in 2017 to 9,142 ha in 2024, measured from Sentinel-2 in a published study.

Source 3
Why nobody just counts them

Mangroves grow in the hardest place on the coast to survey

They sit in soft mud between high and low tide, cut by channels that fill and empty twice a day. You can reach the edge on foot and a little more by boat. The middle of a large forest is slow going. Doing that from Abu Dhabi to Ras Al Khaimah every year isn't realistic, so surveys sample a few plots and estimate the rest.

MethodCoverageHow oftenRunning costCan look back in time?
Field surveySample plotsOccasionalTeams, boats, timeNo
Ain MarsadEvery creek on the coastAny year, in minutesFree public data and toolsYes, back to 2017

Ain Marsad doesn't replace field teams. It tells them which creek to visit.

How it works

Three steps from satellite photo to answer

Step 1 · Collect

Take a year of photos

Sentinel-2 photographs the whole UAE coast about every five days. In Google Earth Engine we stack one year of images and keep the middle value for each pixel. That strips out clouds, haze and some of the tide effect.

Step 2 · Classify

Label every pixel

A random forest model, trained on known sites such as Jubail Island, the Eastern Mangroves and Khor Kalba, decides whether each 10 m pixel is mangrove or not. We test it on labelled examples it never saw.

Step 3 · Compare

Put two years side by side

Comparing an early map with a recent one splits the coast into gained, lost, and planted-but-not-surviving areas, with hectares for each.

The clue the model learns

Mangrove is the only thing on our coast that's green and wet at once

Healthy plants reflect lots of infrared light, which we measure with NDVI. Water absorbs infrared, which we measure with NDWI. Desert shrubs are green but dry. The open sea is wet but not green. Mud and sand are neither. Mangrove scores high on both.

Wet mud, shallow water and algae can still look alike in one photo. That's why we use a trained model and a full year of images instead of one snapshot.

Desert plants green, dry Mangrove green + wet Sand, mud, built neither Open sea wet, not green Wetness (NDWI) · low → high Greenness (NDVI) · low → high

Concept diagram. Positions are qualitative, not measured values.

What the map shows

Compare two years and the coast splits into three colours

New mangrove

Growing where there was none before. Successful planting or natural spread. This is what the 2030 target is really asking about.

Lost mangrove

There before, gone now. Construction, dredging or die-off. Loss never shows up in a planting report, and it can quietly cancel the gains.

Planted, didn't survive

A known planting site where the model sees no mangrove two or more years later. This tells a team exactly which creek to replant.

The third colour needs the location and year of each planting. Where we have that record, we can flag failures. Where we don't, we report gain and loss only.

Live demonstration

What you'll see us run

Everything runs in Google Earth Engine from a laptop. No hardware, nothing to set up on the day. Results appear here once we've measured them.

The coast, before and after

An early year beside a recent year, mangroves highlighted on both. You can see the difference from across the room.

Shown live

The change map

Gain in green, loss in red, with hectares gained, hectares lost and the net difference.

Hectare totals: shown live

One creek, up close

We zoom into a single site and step through it year by year. A national statistic becomes one real place with one clear answer.

Site: shown live

How often we're right

Accuracy on labelled examples the model never saw during training. We'll report our own number, whatever it is.

Accuracy: reported live
Why it matters

Cheap to run, complete, and able to look backwards

No running cost

Satellite images are free, and Google Earth Engine is free for research and non-commercial use. No sensors to install or repair after salt water gets to them. Running cost is usually why monitoring projects stop.

The whole coast

Every creek, including the ones nobody would think to visit.

History on day one

The archive goes back to 2017, so we can measure change that has already happened instead of waiting years for a baseline.

Independent check

The 2030 target is a public promise. This measures it without relying on planters to report on their own work.

What it can't do

We'd rather tell you the limits ourselves

It can't see seedlings

A 10 m pixel is far bigger than a young plant. Sites usually appear after two to three years of growth. So it answers "did the planting survive?", not "did it happen?"

Tides add error

A forest looks different at high and low tide. Yearly median images reduce this but don't remove it. It's probably our biggest source of error.

Mangrove or not, that's all

It doesn't separate species. Almost all UAE mangrove is one species, so this matters little here.

It doesn't replace field work

It can't measure tree health, height or wildlife. It tells a team where to go, not what they'll find.

Our accuracy will be lower than published research

Researchers used bigger datasets and weeks of training. We're students building this in days, and we'll report our own results.

The method isn't new

Researchers have already mapped UAE mangroves with Sentinel-2. Our contribution is a year-by-year change map anyone can read.

What could come next

A yearly mangrove report for every emirate, produced for free

Area per emirate, change since last year, and a list of places where planting seems to have failed. After that, the same method could be tested on seagrass, and on any other country's mangrove pledge.

The information to answer this question already exists. It's been free for years. What was missing was someone willing to look.

Team & sources

Applied Technology School, Al Ain

Hamdan Nabil Butaih Abdulla Al Shamsi · Mayed Khalid Abdulla Alnaqbi
Supervisor: Rahma Al-Najjar

  1. Tolley, G. "Cop26: UAE pledges to plant 100 million mangroves by 2030." The National, 9 Nov 2021. thenationalnews.com
  2. Environment Agency – Abu Dhabi, via Abu Dhabi Mangrove Initiative. "Mangroves: Important Nature-Based Solution." 15 Mar 2024. admangroves.ae
  3. "Artificial Intelligence Mangrove Monitoring System Based on Deep Learning and Sentinel-2 Satellite Data in the UAE (2017–2024)." arXiv:2411.11918, Nov 2024 (preprint). arxiv.org
  4. European Space Agency. Sentinel-2 mission. sentiwiki.copernicus.eu
  5. Google Earth Engine. earthengine.google.com