Coastal Environments
Areas where the land masses meet the seas
Interface regions where land and sea meet
- Directly in contact with the sea
- 25km away from the sea
- 50km away from the sea
- French Coast are densely populated:
- 4% of the French territory
- 10% of the French population
- Globally:
Source: Cosby et al. (2024) , Reimann et al., (2023)
1 billion people (15%) within 10km (4%)
~3 billion by 2100
Human activities
Hotspots of Economic Growth
Seaport
Dredging
Aquaculture
Energy Production
Artificialisation
- Fishing activities
- Tourism
- Industries
- …
Environmental Impacts
The mark of human activity on nature
Oil spills
Erosion
Alien Species Introduction
Climate change
Habitat destruction
- Fishing activities
- Pollutions
- …
Intertidal habitats
Living on the edge of land and sea
Areas between high and low tide
Saltmarshes
Mangroves
Polychaete reefs
Rocky reefs
Tidal flats
Oyster reefs
A rich variety of intertidal habitats
Soft-bottom substrates
Guadalquivir River, Spain
- A - Magnoliopsida
- B - Bacillariophyceae
- C - Phaeophyceae
- D - Florideophyceae
- E - Chlorophyceae
Five Taxonomic Classes
of Vegetation
Hard-bottom substrates
Vigo, Spain
Saja estuary, Spain
Ecosystem Services
- Protection against Erosion
- Carbon fixation
- Nursery & Shelter
- Eutrophication mitigation
~ $30 trillion per year
Protect these ecosystems:
- Habitat Directive (1992)
- Water framework Directive (2001)
- Marine Strategy Framework Directive (2008)
- Birds Directive (2009)
- Nature Restoration Law (2024)
Good knowledge and monitoring to inform policies
Remote Sensing
A tool to map them all !
Traditional sampling methods:
- Expensive
- Time consuming
- Low extent and temporal resolution
- Hard to access
Remote Sensing:
- Cost effective
- Good coverage/Time ratio
- Synchronous broad-scale view
- Simplifies the field work
From the sky to the sea
The science of obtaining information about objects or areas from a distance
Applied to Earth Observation:
Remote Sensing
Fieldwork remains essential to make sense of what satellites see
Radiometric calibration
Aven, France
Ground truthing
Noirmoutier, France
Features georeferencing
Tainaron, Greece
Sampling
Cadiz, Spain
Challenges to map intertidal vegetation
Introduction to Spectroradiometry
\[R(\lambda) = \frac{L_{\text{u}}(\lambda)}{L_{\text{d}}(\lambda)}\]
- \(\lambda\) is the Wavelength
- \(L_{\text{u}}\) is the upwelling radiance
- \(L_{\text{d}}\) is the downwelling radiance
\[R_i^*(\lambda) = \frac{R_i(\lambda) - \min(R_i)}{\max(R_i) - \min(R_i)}\]
- \(R_i(\lambda)\) is the reflectance the the wavelength \(\lambda\) of the spectrum \(R_i\)
- \(min(R_i)\) and \(max(R_i)\) are the minimum and maximum reflectance of the spectrum \(R_i\)
- Each spectrum is between 0 and 1
ASD FieldSpec Handheld 2
Hyperspectral Sensor
A lot of Narrow Spectral Bands
- Is it possible to discriminate green macrophytes using remote sensing techniques ?
- What is the impact of the spectral resolution on the discrimination accuracy ?
Material & Methods
Building a Spectral library of intertidal vegetation
Total of 332 Spectra of 5 taxonomic classes
Spectral degradation
ASD
PRISMA
Drone
S2 - 20m
Pléiades
S2 - 10m
250 bands
50 Bands
10 Bands
8 Bands
4 Bands
4 Bands
Spectral comparisons
Compare the Spectra:
- nMDS + ANOSIM for each spectral resolution
Compare the Sensors:
- Supervised Machine Learning Classifiers
- Random Forest
Training of the model:
- 75 % of the dataset
- Maximisation of the AUC-ROC
Validation of the model:
- 25 % of the dataset
- Accuracy metrics
Putting theory into practice
DJI Matrice 200
Micasense RedEdge-MX Dual
Training of the model:
- Flight Height: 12m
- Product resolution: 8mm
Large scale coverage:
- Flight Height: 120m
- Product resolution: 80mm
Results
Hyperspectral library
Hyperspectral library - nMDS
Hyperspectral library - Random Forest Classifier
- Global accuracy: 0.95
- Cohen’s kappa: 0.93
- Sensitivity: 0.93
- Specificity: 0.98
- Global accuracy: 0.83
- Cohen’s kappa: 0.79
- Sensitivity: 0.84
- Specificity: 0.96
Drone imagery - Example of classification
Chlorophyceae
Bacillariophyceae
Magnoliopsida
Florideophyceae
Drone imagery - Validation
Drone imagery - Variable importance
Discussion
Pigment Composition, Spectral Signature and Variable Importance
Similar pigment composition,…
- … but difference in carotenoid to chlorophyll-a ratios (Repolho et al., 2017)
- … difference in the cell structure and the water content
Distinction between green macrophyte possible using multispectral resolutions, …
- … but ~530 nm & ~650 nm are key wavelengths
Green macrophytes often co-occurs in intertidal areas…
- Ultra high spatial resolution (from 80 to 8mm per pixel)
Green macrophytes often co-occurs in intertidal areas…
- Ultra high spatial resolution (from 80 to 8mm per pixel)
- Easy Photo-interpretation of pixels
- More than 1 000 000 training pixels. Over 11 sites of 3 country
- Diverse training dataset
Drone: 0.26 ha ~ 2.5 millions pixels
S2: 25 000 hectares ~ 2.5x Paris
{style=“transform: rotate(45deg); height=”2000”}
Bourgneuf Bay, July 2024
Ecological Context & Significance
History of the aquaculture of the oyster in Europe
Flat Oyster
Portuguese Oyster
Pacific Oyster
A Hidden Passenger
- Originated from Japan
- 10 000 T spat were imported between 1971 and 1973
- Originated from Japan
- Transport of fragment of Gracilaria vermiculophylla
Resilient to:
- Salinity changes
- Desiccation
- Eutrophic conditions
- Can attach to shells, rocks or colonise soft bottom areas
Well adapted to European estuaries
Gracilaria vermiculophylla
Belon Estuary, France, 2024
First observation in Europe in the Belon, Brittany, in 1996
Aveiro, Portugal, 2021
Etel, France, 2024
Auray, France, 2024
Scorff, France, 2024
Saja estuary, Spain, 2024
Source: Mendoza-Segura et al., 2023
source: Buestel et al., 2009
Ecological Impacts of the invasion
Negatives:
- Can affect native Fucoids and Seagrasses
- Alter the sediment composition and structure
- Modify or disrupt trophic interactions
Positives:
- Create new habitats
- Stabilize the Sediment
Monitoring and Managing
Remote Sensing as a tool to follow the invasion
Satellite & Aerial views:
- Follow the invasion over time
- Go back in time
Drone:
- Flexibility to monitor the early stages of the invasion
- Offer an ultra high resolution
Material & Methods
Historical analysis
Sciences et Techniques, Nantes, 1962
Maps and Aerial photographs archives
- 8 images between 1952 and 2012
- 1 Drone flight in 2024
Photo interpretation of images to retrieve the area covered by G. vermiculophylla
Drone Mapping
DJI Matrice 300
4 Drone flight over G. vermiculophylla
Micasense RedEdge-MX Dual
DJI Zenmuse L1
2 Instruments:
- Multispectral camera
- LiDAR
10 Spectral bands between 444 and 840 nm
- NIR LiDAR
- 240 000 points/s
- ~ 3cm accuracy
- High resolution RGB camera
Digital Surface Model:
- Map of the Slope of the Mudflat
Generalised Linear Mixed Model
\[ \begin{align*} \mathrm{Cover}_{ij} &\sim \mathrm{Beta}\bigl(\mu_{ij}\,\phi,\,(1-\mu_{ij})\,\phi\bigr),\\[1em] \mu_{ij} &= \mathrm{logit}(\eta_{ij}), \\[1em] \eta_{ij} &= \underbrace{\alpha_j}_{\substack{\text{intercept for}\\\text{site }j}} + \underbrace{\beta_1\,\mathrm{Bathymetry}_{ij}}_{\text{effect of elevation}} + \underbrace{\beta_2\,\mathrm{Slope}_{ij}}_{\text{effect of slope}}. \end{align*} \]
Results
Historical records in the Belon estuary
Topography of the mudflat
RGB Composition
DISCOV Classification
DSM Color Composition
Slope Categorized
Elevation vs Presence of Algae
- Higher Cover on the Upper Intertidal
- The steeper the lower the cover
Discussion
First map of the spatial distribution of G. vermiculophylla:
- It can create large monospecific meadows…
- … or be mixed with others intertidal vegetations
First map of the spatial distribution of G. vermiculophylla
Drone mapping G. vermiculophylla with machine learning
Saja estuary, Spain
Belon estuary, France
Distribution linked with the topography
- Inhabit the upper intertidal
- Resistant to desiccation, light and salinity variations
Distribution linked with the topography
- Inhabit the upper intertidal
- Resistant to desiccation, light and salinity variations
- Inhabit flat areas…
- …experiencing lower current velocity during tidal exchanges
Invasion phases
Lag Phase
- Very low abundance
- Need for genetic or mutualistic adjustment
Expansion Phase
- Near‑exponential increase in cover
- Control effort and cost rise sharply
Saturation Phase
- Percent cover reaches a plateau, growth limited by space/resources
- Ecosystem impacts stabilise but persist
Short Lag phase
- Large number of fragments/individuals has been introduced repeatedly in the environment
- G. vermiculophylla is well adapted to European estuaries and already suited to local climate
- Not grazed by native species
Remote Sensing can monitor early stages of the invasion…
- Making it a powerful tool for early decision making
- Facilitates timely interventions
Introduction
Browning of seagrasses across Europe
Quiberon, France, September 2021
What’s in the literature ?
On subtidal Zostera marina and Cymodocea nodosa:
- Highly vulnerable to elevated sea temperatures in winter and spring, leading to early flowering, high mortality, and reduced biomass.
- Highly vulnerable to elevated sea temperatures in winter and spring, leading to early flowering, high mortality, and reduced biomass.
- Photosynthetic activity rises during HWs but diminishes during recovery, impairing performance and reducing leaf biomass.
- Highly vulnerable to elevated sea temperatures in winter and spring, leading to early flowering, high mortality, and reduced biomass.
- Photosynthetic activity rises during HWs but diminishes during recovery, impairing performance and reducing leaf biomass.
- Responses vary greatly between species…
- Highly vulnerable to elevated sea temperatures in winter and spring, leading to early flowering, high mortality, and reduced biomass.
- Photosynthetic activity rises during HWs but diminishes during recovery, impairing performance and reducing leaf biomass.
- Responses vary greatly between species…
- …and within a single species across latitudes.
What about Zostera noltei ?
Impact on the reflectance ?
Impact of Extreme Atmospheric temperature ?
Extreme Temperature Events = Heatwaves
Defined as periods of three or more consecutive days where the daily maximum temperature (Tmax) exceeds the calendar-day 90th percentile based on at least 30 years
Defined as periods of three or more consecutive days where the daily maximum temperature (Tmax) exceeds the calendar-day 90th percentile based on at least 30 years
Defined as periods of three or more consecutive days where the daily maximum temperature (Tmax) exceeds the calendar-day 90th percentile based on at least 30 years
Defined as periods of three or more consecutive days where the daily maximum temperature (Tmax) exceeds the calendar-day 90th percentile based on at least 30 years
Defined as periods of three or more consecutive days where the daily maximum temperature (Tmax) exceeds the calendar-day 90th percentile based on at least 30 years
Hypothesis & Objectives
Heatwaves alter the spectral reflectance of Zostera noltei seagrass. This change can be detected using remote sensing.
- Evaluate the direct impact of heatwave-induced thermal stress on the reflectance of Zostera noltei through controlled experiments.
- Develop a spectral index for detecting stress-induced changes in seagrass coloration.
- To apply findings from experimental reflectance changes to satellite-based remote sensing, assessing the spatial extent and temporal dynamics of an heatwave event that occurs in September 2021, in Quiberon, on seagrass meadows.
Material & Methods
Experiment in the Lab
Intertidal chambers from ElectricBlue
- Air Temperature : from 18 to 60°C
- Water Temperature : from 8°C to 55°C
- Programmable tides
- Programmable lights
Measure variation of seagrass leaves reflectance over time.
Seagrasses inside of a chamber
Hyperspectral measurement every minute in each tank
Control
Treatment
Both Experimental and Satellite Mapping
- Well established radiometric Indices:
\[ f''(\lambda_i) \approx \frac{f(\lambda_{i+1}) - 2f(\lambda_i) + f(\lambda_{i-1})}{(\Delta \lambda)^2} \]
\[NDVI = \frac{R(NIR)-R(Red)}{R(NIR)+R(Red)}\]
\[GLI = \frac{[R(Green)-R(Red)]+[R(Green)-R(Blue)]}{(2 \times R(Green) )+ R(Red) + R(Blue) }\]
Results
Spectral signatures:
SHSI design
Heatwave experiment
Spectral metrics:
\(R''_{665 \, \text{nm}}\) drops by 68 %
\(NDVI\) drops by 31 %
\(GLI\) drops by 54 %
\(SHSI\) increases by 420 %
Material & Methods
Satellite Mapping of the impact of Heatwaves on seagrasses
Atmospheric heatwave between the 4th of September 2021 and the 7th of September 2021 in Quiberon
3 Sentinel-2 images, level L2A, Low Tide:
- Before: 1st of September 2021
3 Sentinel-2 images, level L2A, Low Tide:
Before: 1st of September 2021
During: 6th of September 2021
3 Sentinel-2 images, level L2A, Low Tide:
Before: 1st of September 2021
During: 6th of September 2021
After: 8th of October 2021
Results
in situ Satellite Mapping
Before
During
Before
During
After
Discussion
Mapping Impacted meadows
Seagrasses impacted by heatwave have a distinct spectral signature
- Reflectance drops at 560 and 740nm
Thermal stress = Oxydative stress
- Degradation of Chlorophyll-a
- Shift in pigment ratios = Carotenoids becoming more prominent
Oxydative stress = Membrane damages
Possible to detect seagrass thermal stress using satellite remote sensing, using SHSI
- Most Sensitive early-metric to flag damage over meadows
- Designed to be used by most space missions (Sentinel-2, Pléiades, Wordview-3, SkySat, GeoSat-2…)
- but also by future missions (Sentinel-2 Next Generation, LandSat Next)
Satellite mapping reveals tide-modulated, patchy impacts
- Upper intertidal areas are more impacted
- 95% of darkened patches are above 3.9 m
- Darkening occurs in seagrasses exposed more than 13 h per day
Rapid global escalation of HW frequency, intensity and duration
- How does increasing heatwave pressure affect recovery capacity of Z. noltii?
- Can seagrasses maintain resilience under more frequent aerial exposure?
- Are upper-intertidal meadows approaching a tipping point?
Macrophyte Discrimination: Successes & Challenges
Key Success: Multispectral RS + ML effectively differentiate intertidal vegetation, even with similar pigments (e.g., seagrass vs. green macroalgae).
Core Mechanism: Relies on detecting subtle spectral variations (pigment proportions & concentrations).
DISCOV Algorithm – A Key Output:
- Dynamic, adaptable, and open-source (GitHub).
- Challenge: Initial underperformance with Florideophyceae due to limited training data.
- Solution: Expanded training dataset significantly improved model accuracy & robustness.
Ongoing Challenge: Natural variability in pigment content (phenology, stress, intraspecies variation).
Future Direction: Continue expanding the training dataset (broader geographic/temporal range) to enhance algorithm generalisability.
Drone and Satellite Interactions: A Powerful Synergy
Complementary Strengths:
- UAVs: Unmatched spatial resolution, Flexible, ground-truth validation.
- Satellites: Broad spatial coverage, consistent temporal monitoring for large-scale trends.
Example: The ICE CREAMS Model (Davies et al., 2024a,b):
Methodology: DISCOV (drone-derived) classifications used to train and validate ICE CREAMS (Sentinel-2 satellite) model.
Achievements:
- Mapped intertidal seagrass across diverse European sites.
- Revealed latitudinal gradients in seagrass phenology.
- Tracked inter-annual trends in seagrass extent.
Key Benefit: This synergistic approach balances local accuracy with regional/global scalability, optimising monitoring efforts.
RS: A Powerful Lens on Coastal Changes
The DAPSI(W)R(M) framework
DAPSI(W)R(M), a framework to connect human activities to environmental impacts
👥 Driver
The growing population necessitates more sites for oyster production.
- 🛰 Study Sea Surface Temperature, Turbidity, Chlorophyll-a
- 🛸 Provide ultra-high-resolution topographic mapping of intertidal sediments.
⚙️ Activity
Expansion of oyster aquaculture in coastal and estuarine areas.
- 🛰 Study Sea Surface Temperature, Turbidity, Chlorophyll-a… associated is oyster production models
- 🛸 Provide ultra-high-resolution mapping of farm structures on demand.
⚠️ Pressure
Waste, sediment resuspension, and nutrient input from farming operations.
- 🛰 Track turbidity and chlorophyll anomalies.
- 🛸 Mapping the spread of alien species
🌿 State & Impact
Altered water quality and degraded seagrass beds near farming sites.
- 🛰 Detect habitat trends over time and across regions.
- 🛸 Assess detailed plant cover, diversity and benthic healths.
🛠 Responses (Measures)
Environmental regulations, site management, and restoration programs.
- 🛸 Provide metrics for restoration efforts and track their success.