Global Mangrove Watch (GMW)

Background

The Global Mangrove Watch (GMW) dataset (v2.0) depicts the global extent of mangrove forests for the years 1996, 2007, 2008, 2009, 2010 (baseline dataset), 2015 and 2016, derived using a combination of L-band Synthetic Aperture Radar (SAR) and optical satellite data 1.

The Global mangrove Watch was established in 2011 under the Japan Aerospace Exploration Agency’s (JAXA) Kyoto & Carbon Initiative by Aberystwyth University, solo Earth Observation and the International Water Management Institute, with the aim to provide open access geospatial information about mangrove extent and changes to the Ramsar Convention on Wetlands. In collaboration with Wetlands International and with support from DOB Ecology, the first GMW baseline maps were released in 2018 at the Ramsar COP13. In 2019 the GMW teamed up with The Nature Conservancy, Wetlands International, NASA, and a host of partners to develop the Global Mangrove Watch Platform. The effort is supported by the Oak Foundation, COmON Foundation, DOB Ecology and the Dutch Postcode Lottery.

The GMW maps also constitute the official mangrove datasets used by UNEP for reporting on Sustainable Development Goal 6.6.1 (change in the extent of water-related ecosystems over time).

Methodology

The GMW mangrove maps were derived in two steps:

  • generation of a baseline map of global mangrove extent for the year 2010, and,
  • detection of changes (both gains and losses) between the 2010 baseline and each of the other six annual epochs, respectively.

(1) Generation of 2010 global mangrove extent baseline map:

The 2010 baseline map was derived by Random Forest Classification of a combination of radar (ALOS PALSAR) and optical (Landsat-5, -7) satellite data. Approximately 15,000 Landsat scenes and 1,500 ALOS PALSAR (1 x 1 degree) mosaic tiles were used to create optical and radar image composites covering the coastlines along the tropical and sub-tropical coastlines in the Americas, Africa, Asia and Oceania.

The classification was confined using a mangrove habitat mask, which defined regions where mangrove ecosystems can be expected to exist. The mangrove habitat definition was generated based on geographical parameters such as latitude, elevation and distance from ocean water. Training for the habitat mask and classification of the 2010 mangrove mask was based on randomly sampling some 38 million points using historical mangrove maps for the year 2000 (Giri et al., 2010; Spalding et al., 2010), water occurrence maps (Pekel et al, 2017), and Digital Elevation Model data (SRTM-30).

(2) Generation of 1996, 2007, 2008, 2009, 2015 and 2016 maps of mangrove extent and changes:

The maps for the other six epochs were derived by detection and classification of mangrove losses (defined as a decrease in radar backscatter intensity) and mangrove gains (defined and a backscatter increase) between the 2010 ALOS PALSAR data on one hand, and JERS-1 SAR (1996), ALOS PALSAR (2007, 2008 & 2009) and ALOS-2 PALSAR-2 (2015 & 2016) data on the other. The change pixels for each epoch in question were then added or removed from the 2010 baseline raster mask (buffered to allow detection of mangrove gains also immediately outside of the mask) to produce the new yearly extent maps.

Accuracy assessment

Classification accuracy of the 2010 baseline dataset was assessed with approximately 53,800 randomly sampled points across 20 randomly selected regions. The overall accuracy was estimated to 95.25 %, while User’s (commission error) and Producer’s (omission error) accuracies for the mangrove class were estimated at 97.5% and 94.0%, respectively.

Classification accuracies of the changes were assessed with over 45,000 points, with an overall accuracy of 75.0 %. The User’s accuracies for the loss, gain and no-change classes respectively were estimated at 66.5%, 73.1% and 83.5%. The corresponding Producer’s accuracies for the three classes were estimated as 87.5%, 73.0% and 69.0%, respectively.

NOTE: The accuracy assessments above were undertaken using “traditional” accuracy assessment methods (where confusion matrices are used to assess overall accuracy and omission/commission errors). A more comprehensive accuracy assessment employing the “Good practices for estimating area and assessing accuracy of land change” as developed by Olofsson et al. (2014) is planned for 2020.

Notes and cautions when using the GMW maps

  • Users should be aware that the Global Mangrove Watch map is a global dataset, and as such, it should not be expected to achieve the same high level of accuracy everywhere as a local scale map derived through ground surveys or the use of very high spatial resolution geospatial data. A global area mapping exercise using consistent data and methods – although supplemented with ground-based data for calibration and validation – for logistical reasons generally requires a trade-off in terms of local scale accuracy. Nonetheless, global maps can be improved locally (or nationally) by adding improved information (in-situ data and aerial or drone data) for training and re-classification.
  • Several different factors can affect the classification accuracy, including satellite data availability, mangrove species composition and level of degradation.
  • While the original pixel spacing of the satellite data used for the mapping is 25-30 metres, a minimum mapping unit of approximately 1 ha is recommended due to the classification uncertainty of a single pixel. The classification errors (in particular omission errors) typically increase in regions of disturbance and fragmentation such as aquaculture ponds, as well as along riverine or coastal reef mangroves that form narrow shoreline fringes of a few pixels.
  • In general, the mangrove seaward border is more accurately defined than the landward side where distinction between mangrove and certain wetland or terrestrial vegetation species can be unclear.
  • Striping artefacts due to Landsat-7 scanline error are present in some areas, particularly West African regions due to lack of Landsat-5 data and persistent cloud cover.
  • Known data gaps in GMW v2.0 (to be addressed during 2020 revision):
    • Aldabra island group (Seychelles)
    • Andaman and Nicobar Islands (India)
    • Bermuda (U.K.)
    • Chagos Islands
    • Europa Island (France)
    • Fiji (part east of Antemeridian)
    • Guam and Saipan (U.S.)
    • Kiribati
    • Maldives
    • Marshall Islands
    • Peru (south of latitude S4°)
    • Wallis and Futuna Islands (France)

Next GMW Data Release

Release of a revised dataset, with maps from 2017 and 2018 to be added, known data gaps (see above) to be filled, is scheduled for release in Q2/2021

GMW Data Access

The UNEP World Conservation Monitoring Centre (UNEP-WCMC) serves as the central data hub for access to the GMW datasets:

https://data.unep-wcmc.org/datasets/45

GMW Online Vieweing

  1. Bunting, P., A. Rosenqvist, R. M. Lucas, L.‐M. Rebelo, L. Hilarides, N. Thomas, et al. 2018. The global mangrove watch—a new 2010 global baseline of mangrove extent. Remote Sens. 10, 1669. https://doi.org/10.3390/rs10101669
data.unep-wcmc.org