Managing Mangrove Forests with Optical and Radar Environmental SaTellites (MAMAFOREST)

Start-End 01/12/2015 -  30/11/2018
Programme STEREO 3
Contract SR/00/323
Objective Space, time and operating mode are common considerations when selecting remote sensing data for observing terrestrial and marine environments. Many optical and radar sensors have allowed observations at spatial resolutions < 30 m but it is the long history of acquisitions by these sensors that enables historical changes in extent but also biophysical characteristics to be quantified.   There have also been significant advances in the domain of radar remote sensing and particularly in polarimetric and interferometric processing, which allows interrogation of the three-dimensional structure of surface objects.

For mangrove forests, the development of robust remote sensing methods for mapping and monitoring their biophysical properties is becoming increasingly important as anthropogenic processes (e.g., pollution) and events (e.g., clearance) lead to significant losses across their range.  For maximum benefit however, focus needs to be on the integration of temporal data from both optical and radar sensors as each provides unique but complementary information on mangroves. The main goal of this project therefore is to establish how the use of radar and optical remote sensing data can be optimised for quantifying the changing extent, structure, species composition and above ground biomass (AGB) of mangroves, with specific focus on the expansive mangroves within the Matang Mangrove Forest Reserve (MMFR) in Malaysia.  This area has been silviculturally managed since 1902 and the areal extent of mangroves has therefore changed very little. There are four distinctive management zones: productive forest, protective forest, restrictive protective forest, and unproductive areas. Most of the silviculture takes place in the productive forest, whereas several types of protected forest constitute the protective forest. Therefore both, productive and protective forests are central to our focus. However, considerable internal restructuring of the productive forests has occurred because of logging activities and regrowth. These forests contrast with those that are protective and which support a greater diversity of structures and species. One of the goals therefore is to establish, primarily through remote sensing observations of both productive and protective forests, the sustainability of silviculture and the impacts on biodiversity and carbon values.

The primary objective is to integrate time series of optical and radar remote sensing data to evaluate the viability and sustainability of logging within the Matang Mangrove Forest Reserve (MMFR) by quantifying the yearly evolution of structures, species and AGB. The study has four specific objectives, with these focusing on time-series analysis, validation, biophysical parameter retrieval and characterising change. In this project, comparisons will be made between productive and protective forests.   

Method The project will be undertaken as six work packages (WP1-6), with WP1 dealing with management of the project and WP6 is dealing with its dissemination.

WP2: Temporal changes in mangrove cover in the Matang Mangrove Forest Reserve with historical, current and new satellite sensors (hereafter also shortened to ‘Time-series analysis’).

The first step is the processing of historic satellite imagery with a spatial resolution that is sufficient to resolve relatively small patches of mangrove forest. The Landsat historic archive is a logical choice to start from because of the availability of data with a high spatial (30 m) and temporal resolution and existing processing methods. 40 years of consecutive Landsat imagery will be processed for the MMFR, with this enriched by mangrove maps digitized manually from historic US military airborne data. Data from the newly launched Sentinel 2 sensor will also be processed to facilitate up-to-date (2015 onwards) mapping and comparison with Landsat-8 data. VITO is developing a Sentinel-2 processor within the Highroc FP7 project. The processor will start from the Level 1C TOA reflectances, which are delivered as tiles in UTM WGS 84 (in SAFE format). The OPERA atmospheric correction, developed by VITO, will then be applied to retrieve Level-2 reflectance products. OPERA includes an image-based Aerosol Optical Thickness (AOT) retrieval and MODTRAN based atmospheric correction. The AOT retrieval is based on the procedure developed by Guanter et al. (2008) and describes the surface by a linear combination of 4 pure vegetation spectra and 1 soil spectrum. The atmospheric correction is based on MODTRAN Look-Up-Tables (LUT) resampled to Sentinel-2 spectral bands and following the formulation given by de Haan et al. (1996; described in detail in Sterckx et al. (2011)).
To classify the extent of mangroves at each time-step, all Landsat and Sentinel-2 channels will be considered with the inclusion of the SWIR infrared channels regarded as essential (Chen et al., 2003). Furthermore, also SAR data (X, C- and L-band) and derived products (e.g., height maps) will be integrated in the classification process. Given that mangroves only occur near the coast, a global digital elevation model (SRTM NASA v3, 2014) will be used to assist with the initial classification of mangrove extent given their occurrence in low lying areas. An upper threshold of 30 m will capture potential areas with mangroves (noting that this is the maximum canopy height expected). An object-based approach to classification of mangroves will be applied, with this considered to be advantageous over pixel-based approaches which suffer from salt and pepper effects. High resolution satellite imagery (WorldView-3), along with data from the WP3 field campaign (GPS tagged pictures), will be used for the validation of the final mangrove maps.  For each year, maps of the extent of mangrove and non-mangrove (including logged areas) will be generated, with reference made also to previous years’ classifications to determine whether the forest is regrowing (hence leading to a regrowth class). Reference will also be made to existing logging coupes as mapped by local forestry departments.
To establish the history of logging in the MMFR over the time-series, maps of mature and regrowth forest and non-forest for each observation point (sub-annual) will be compared to generate maps of forest age, frequencies of clearance (or harvesting) and periods as non-forest (bare ground, cut stumps in the interim period between clear felling and regrowth). These maps will ultimately be used to establish the link between forest age and retrieved biophysical attributes in WP4 and also to determine whether the history of logging and disturbance impacts on the ability of forests to restore tree species diversity, structural integrity and AGB. The maps will also be used to guide the collection of field data in WP3.

WP3 Gather and analyse field data on mangrove (hereafter also shortened to ‘Fieldwork’).

One field campaign will be organized outside the Monsoon season. The fieldwork will be organized in such a way that the stratified sampling units (squared plots of at least 10 m × 10 m, located in homogeneous forest stands of varying age and history) can be used for ground-truthing the imagery and for more fundamental ecological use. Vegetation structure and silvimetric measurements will include the species composition of the sampling plots, tree position (within the plots), diameter and the height of different strata (e.g., young vegetation layer and a canopy tree layer) and an assessment of the juvenile vegetation layer (propagules and seeds). A rapid appraisal of the above-ground root complex as a proxy for the surface in which propagules and seeds can be entangled will also be undertaken. Salinity will also be recorded using a refractometer to establish whether this influences tree architecture and growth.  All signs of cutting in the exploited and supposedly non-exploited zones will be recorded. The collected field data in this WP are essential for validating the mangrove maps (WP2), the calibration and validation of radar derived biophysical parameters (WP4) and further generic interpretation of the results of the mangrove evolution and sustainability analysis (WP5).

WP4 Retrieval of mangrove biophysical parameters based on radar and optical data (hereafter also shortened to ‘Biophysical parameters’).

From the field measurements, height metrics (e.g., maximum height, Lorey’s height, the distribution of plant material in the vertical profile) will be calculated and used to validate equivalent metrics derived from interferometric X- and C-band SAR (including Sentinel-1a). Changes in height metrics that will have occurred between the time of SAR data acquisition (in the case of Tandem-X and the SRTM C-band SAR) and field measurement will be compensated for by referencing growth rates generated from historical forest inventories.  Estimates of the AGB of the different plant components (roots, trunks, branches and foliage) will be generated using available allometric equations, with the estimation of root biomass being particularly relevant given that the productive forests are dominated by R. apiculata and R. mucronata. These estimates will be used to develop relationships between total and component biomass and L-band SAR backscatter, with or without height surfaces derived from X- and C-band interferometric SAR. The L-band SAR backscatter is lower for higher biomass mangroves with prop root systems compared to those without and of equivalent stature. Indeed, the L-band backscatter (particularly at HH polarisation) decreases with increases in AGB (to over 300 Mg ha-1).  This is in contrast to mangroves without prop root systems where the L-band backscatter increases asymptotically, with the saturation level being approximately 60-100 Mg ha-1.  Using the available SAR data acquired at different epochs, height metrics and AGB will be mapped across the MMFR for both productive and protective forests.

Besides AGB and height also maps of the dominant species type will be established. Therefore, the L-band SAR data will be combined with optical (Landsat and Sentinel-2) data to differentiate mangroves dominated by different genera or species. Classification will be based on spectral responses of forests over time using the fielddata, and a classified VHR image as reference. The results of this WP will serve as a basis to assess the temporal and spatial variation of the species composition, forest biomass production and losses (carbon fluxes) and evaluate the sustainability of the management practices in WP5.

WP5 Spatio-temporal analysis on mangrove biomass evolution in relation to the productive vs protective zones of the forest (hereafter also shortened to ‘Change detection’).

The maps of forest age, frequencies of harvesting and periods as non-forest prior to regeneration will be associated with different height and AGB categories to establish the influence of management on the capacity of forests to recover tree species diversity and structural integrity but also to quantify forest degradation. Insights into forest biomass production and losses and gains in carbon (derived from AGB) and tree species diversity will also be quantified to determine the benefits (or otherwise) of managing these forests, with comparisons made against the protective forests.

Landscape metrics will also be used to determine the size and shape of cleared and non-cleared (productive and protective), with this information then used to establish whether the geometric sizes, shapes and configurations are sufficient to support tree species and floral diversity.  These metrics will also be used to establish whether mangroves are sufficiently well distributed to a) sustain the required production of propagules for regeneration and b) ensure that the functional diversity of non-exploited species is sufficient to fulfil the functioning of the mangrove forest as a whole. Key metrics to be considered are among others i) class area, ii) number of patches (increases are associated with fragmentation, iii) mean patch size (increase indicates fragmentation increase), iv) Shannon Evenness index (identifies the distribution of patches) and v) Shannon Diversity index (diversity of patches in an area).

•    Assessment of the sustainability of the logging activity in Matang Mangrove Forest Reserve based on a time-series analysis of Landsat and Sentinel-2 sensor data and radar (ALOS PALSAR, Sentinel-1 SAR)
•    Integration of optical and radar data to classify mangrove dominant species
•    Comparison of time-series of JERS-1 SAR, ALOS PALSAR and ALOS-2 PALSAR-2 to assess trends in AGB as a function of L-band SAR backscatter
•    Generation of height surfaces using the Shuttle Radar Topographic Mission (SRTM) and both the Tandem-X and Sentinel-1 data (linked to ICESAT GLAS) for specific points in time


•    Classifications and derived products of the MMFR (based on historic Landsat, US military airborne imagery, Sentinel 2 and radar imagery)
•    Database with associated data collected on species composition, structure and AGB.
•    Level 3 processed products of biophysical parameters (height metrics and AGB) for the MMFR based on historical SAR data, Sentinel-1a SAR and ALOS-2 PALSAR-2 data
•    Maps of dominant species composition based on L-band SAR, Sentinel 2 and Landsat
•    Allometric equations for selected mangrove species
•    Spatio-Temporal analysis of the Mangrove forest biomass evolution in relation to the productive vs protective zones of the forest
•    Peer-reviewed publications
•    Report for the Forestry Department of Perak with key findings

Website link
Team Member: LOKMAN BIN HUSAIN Mohd Universiti Malaysia Terengganu (UMT)
Coordinator: DAHDOUH-GUEBAS Farid ULB (Université Libre de Bruxelles)
Team Member: VAN DE KERCKHOVE Ruben VITO (Vlaamse Instelling voor Technologisch Onderzoek)
Team Member: LUCAS Richard University of New South Wales - School of Biological, Earth and Environmental Sciences
Sensors used
Related presentations BEODay 2016 - 15. MAMA-FOREST: Mapping mangrove forests with optical and radar environmental satellites
Related publications
Datasets used