Select Page

# Science Review

Apr 15, 2016

Aimed at monitoring the coral reef decline worldwide in agreement with the following scientific approach, the ESA Sen2Coral project of health observatory exploits the Sentinel 2 satellites’ multispectral performances, global coverage, high resolution, frequent revisits time and public access policy.

Authors: John Laporte summarising Dr John Hedley,  in collaboration with the Sen2Coral Project Scientific Team (Claudia Giardino, Tiit Kutser; Antoine Collin; Vittorio Brando; Chris Roelfsema; Stuart Phinn and Peter Mumby).

# Scientific approach: Review of objectives and methodologies

Coral reef monitoring programs based on in situ field surveys have several caveats that restrict its use and the relevance of its outcomes. These are:

• cost-related: detailed, continuous monitoring of coral reefs by field survey is expensive and substantial reef areas are located in developing countries with limited resources;
• scale-related: reefs are highly heterogeneous systems therefore even with sufficient resources monitoring programs provide scattered information in time and space, with some areas being more intensively sampled than others and less easily accessible areas are typically under-sampled.

Using satellite remote sensing within a monitoring program potentially addresses both of these issues. Remote sensing should be seen as a complementary technique to in-situ field surveys. The two approaches produce different types of data which together can be used to build a picture of reef ecology across a wide range of spatial and temporal scales.

### Coral Reef Applications & Methodologies

All published methods appropriate for coral reef applications and methodologies have been reviewed. These include

• surface glint removal
• water column correction
• depth invariant indices
• habitat mapping
• physics based inversion outputs such as satellite derived bathymetry (SDB).

Radiometric changes and bleaching detection have also been reviewed extensively.

##### Water Column Correction

The most practical water column correction techniques are those based on band combinations, which in many cases are entirely image-based. Many of these algorithms are quite similar and typically involve the assumption that the attenuation of light with depth is a negative exponential, i.e. over a constant bottom reflectance,

$R_{i} \propto \exp\left({-k_{i}z}\right)$Where $$R_{i}$$ is the atmospherically corrected above-water reflectance in band $$i$$, $$k_{i}$$ is the water attenuation coefficient for that band and $$z$$ is the depth. The attenuation $$k_{i}$$ depends on constituents of the water and is general unknown, but by combining bands the problem can be reduced to determining relative parameters that can be estimated from the image itself.

##### Habitat Mapping

Habitat classification is the most common objective for coral reef mapping from satellite data. Only standard methods of per-pixel classification will be considered. however due to the submerged nature of the reef special pre-processing techniques for removal of surface glint and the effect of variable depth will be the focus within Sen2Coral.

Benthic habitat classes used for high spatial resolution bottom type classification. Class 8 (reef slope) is used only in the Yucatán site map (shown)

##### Sun Glint Removal

Deglint is a pre-processing algorithm aimed at removing the direct reflection of the sun from the air-water interface or ‘sun glint’

##### Satellite Derived Bathymetry (SDB)

IHO-compliant nautical chart of Yucatán generated from SDB

### Algorithm Implementation

Sen2Coral intends to implement a selection of practical tools based on visual interpretation of time series and pixel based differences. A processing chain comprising ten distinct algorithm modules will be tested and validated to address the Sen2Coral objectives. Access to Implementation details of the selected algorithms available in the Sentinel 2 Toolbox will be granted, including but not limited to:

• spatial registration
• classification
• atmospheric corrections

Comprehensive information on proprietary applications not yet included in the Toolbox will be provided. These will address:

• de-glint
• depth invariant indices
• satellite derived bathymetry
• physics based model inversion
• change and bleaching detection.

Proposed processing chain modules to be used, either pre-existing or to be implemented in the Sentinel 2 Toolbox.

##### Spatial Registration

Change detection requires that image pairs or sequences are spatially aligned.

##### Land, cloud and white cap mask

For many of the processing steps in the proposed modular system it is advantageous to exclude pixels of land, clouds, cloud shadows and breaking waves (white caps).

##### Depth Invariant Indices

Calculation of depth invariant indices is a water column correction step that is frequently a useful pre-processing step for benthic classification. Attenuation of the reflectance is approximately inverse exponential with water depth, therefore the transform,

$X_{i} = \ln\left(R_{i}{-R_{i}^{deep}}\right)$approximately linearises the effect of depth on reflectance. $$R_{i}$$ is the pixel reflectance in band $$i$$, and $$R_{i}^{deep}$$, is the deep water reflectance in that band.
nbsp;

##### Atmospheric Corrections

The potential of the Sentinel-2 for coral reef applications increases with the possibility to have data corrected for the atmospheric effects.

##### Change & Bleaching Detection

Image registration and multi-temporal radiometric corrections are perhaps the most important and indispensable steps in change-detection methods. For change detection, we adopt the common approach used within the ‘land’ remote sensing domain.

##### Sun Glint Removal (De-glint)

The purpose of the glint correction or ‘deglint’ is to remove the contribution of the direct reflectance from the top of the air-water interface from the imagery.

##### Physics Based Model Inversion

The main recent development in coral reef remote sensing algorithms is that of physics-based model inversion approaches. These methods work by establishing a forward model for the spectral water-leaving remote sensing reflectance,$$R_{rs}(\lambda)$$, based on the depth, optical properties of the water and bottom type. The majority of methods are based on a set of equations derived by Lee et al (1998; 1999) which can be represented by a function,

$R_{rs}(\lambda) \approx f\left({P,G,X,H,R(\lambda),\lambda}\right)$Where $$P$$,$$G$$,$$X$$ represent the amount of phytoplankton, dissolved organic matter (CDOM) and particulate backscatter, $$H$$ is the depth and $$R(\lambda)$$ is the bottom type reflectance – which can be taken from a spectral library or linear mixtures thereof.

### Test Areas

Five test areas have been identified by the science team, each characterised by specific properties and covering the Indo Pacific tropical belt most affected by the El Nino phenomenon, extending from the Mozambique Strait, Great Barrier Reef, Andaman Sea and Micronesia all the way to the Marquesas. Validation approaches will rely on comparison of statistic quantities (e.g. water reflectance) and products (e.g categorical habitat mapping) against observed field data and satellite images gathered from Landsat 8 and Sentinel 2 sensors.

### Validation Approaches

Objective Validation Data Considerations for interpretation Accuracy presentation
Categorical habitat mapping Transect or point data of bottom cover Geolocation accuracy (assessed visually near habitat boundaries) Confusion matrices. Overall accuracy
Bathymetry Echo sounder point data Tide corrections. Plumes or other anomalies (assess visually). Geolocation accuracy (as above Correlation plots of estimated vs. sonar bathymetry
Change and bleaching detection Bleaching incidence data (if available). Visual interpretation Plumes or other anomalies (assess visually) Confusion matrices. Time series plots
Water optical properties Data on known IOP ranges None Tabulated results