Mangrove Monitoring in Google Earth Engine
By Dave Thau, Dominic Andradi-Brown, Luiz Cortinhas, Cesar Diniz, Pedro Souza-Filho, Greg Fiske, Chandra Giri, Liza Goldberg, David Lagomasino, Uday Pimple, Dario Simonetti, Aurelie Shapiro, and Xiangming Xiao
Mangroves are salt-tolerant trees and shrubs that live in tidal areas around the world. They protect shores from storms, filter pollutants, and act as nurseries and habitats for innumerable species, many of which are threatened or endangered.
Given their importance to people and global ecosystems, it’s no surprise that they have received a great deal of research. Earth Engine has been involved in much of this work. The brief case studies below describe a variety of ways researchers have been using Earth Engine to monitor the health of mangroves globally, in China, Mozambique, Thailand, and South America.
Collaborating on Global Mangrove Mapping in Earth Engine
Our first example of global mangrove monitoring is EcoMap, an application created and presented at the 2017 American Geophysical Union by Liza Goldberg, a student at Atholton High School, and David Lagomasino, a scientist at NASA Goddard. EcoMap uses Landsat data to identify areas of past mangrove loss and degradation on the global scale. It then evaluates and aggregates the risks of anthropogenic and natural loss drivers such as urbanization, population growth, agricultural or aquacultural expansion, and erosion to produce total mangrove vulnerability estimates. To account for the varying resilience and adaptive capacity of different mangrove forests, analysts can weight individual drivers in this total risk aggregation. They also can interact with the EcoMap platform through tracking location-based loss drivers, graphing NDVI time series change, and determining the proportion of mangrove area, carbon stocks, and biomass in each mangrove-holding nation under high risk. EcoMap is one of several mangrove mapping initiatives conducted in various study regions around the world by NASA Goddard’s Biospheric Sciences Lab. For more information about these efforts see mangrovescience.org.
Another example of global mangrove monitoring comes from an international group from twenty institutions, led by Jon Sanderman at the Woods Hole Research Center. They recently published a paper showing that globally mangrove forest soils holds about 4.5 times the amount of carbon emitted by the U.S. economy in one year. These results also show that mangrove forest destruction released as much as 122 million tons of soil carbon to the atmosphere between 2000 and 2015.
Earth Engine was used to manage, share, and analyze the soil carbon statistics by country, continent, and region. These analyses were done multiple times as new and more precise country coastal boundaries were developed by project partners. Additionally, during the model development phase, Earth Engine was used to quickly aggregate and distribute the working global map at full resolution to project participants, collaborators, and funding agencies for inspection and review. As new modeled outputs were produced the interactive web map could quickly and easily be updated and accompanied by new statistics in near real-time. A global map showing carbon storage in 2000 for the upper meter of soil can be viewed here: bit.ly/2F1Hsbc.
Similarly, the United States Environmental Protection Agency (US EPA) led by Chandra Giri in collaboration with the United States Geological Survey (USGS), NASA, and Duke University have been mapping and monitoring global mangroves from 1970s to present. Earth Engine is being used to collect training points, classification, editing, and results validation. So far, classification of South Asia, Southeast Asia, Australia/New Zealand, Middle East, Africa, USA, and South America has been completed. The team is also using Earth Engine to develop a methodology to monitor mangrove forests on an annual basis. Once mangrove database is prepared, the team aims to quantify the impacts of mangrove cover changes on carbon stock changes and species extinction risks on an annual basis. They also plan to analyze the effectiveness of the existing protected areas system in avoiding carbon emissions and species extinctions caused by mangrove deforestation and degradation.
Combining Radar and Optical Data to Detect Mangroves in China
Human activities, climate change and sea-level rise have resulted in substantial loss of mangrove forests in China. To start tracking changes to Chinese mangroves, an international team of researchers led by Xiangming Xiao at the University of Oklahoma in collaboration with Chinese university (Fudan University) and research institutes (Chinese Academy of Tropical Agricultural Sciences, Chinese Academy of Sciences) used Google Earth Engine to generate an annual map of mangrove forest in China. They developed a new phenology-based classification algorithm that integrates the biophysical (greenness, canopy coverage, and tidal inundation) and geographical (elevation, slope, and intersection-with-sea) characteristics of mangrove forest with time series optical images from the Landsat 7 and 8 satellites and radar images from the Sentinel 1 satellite. The result was a highly accurate map that is more up to date and complete than previous efforts. For more information, see A mangrove forest map of China in 2015: Analysis of time series Landsat 7/8 and Sentinel-1A imagery in Google Earth Engine cloud computing platform.
Scaling and Automating Mangrove Mapping in Mozambique
In 2015, WWF Germany published methods for mapping mangrove extent in the Zambezi Delta from Landsat imagery. This approach has been improved, expanded, scaled up and automated using Google Earth Engine. To assess mangrove at the national scale in Mozambique, Earth Engine was used to create cloud-free Landsat image composites over several time periods (1994, 2001, 2008, 2016) which were then classified into mangrove cover, gain and loss using the Random Forests classification algorithm. Additionally, a mangrove map for 2017 from Sentinel-2 imagery at 10m resolution imagery was developed which showed the complexities, advantages and disadvantages of mapping mangroves at finer spatial scales — notably that overall mangrove extent is much lower at higher resolution because mangrove stands which are lower density, or regenerating are not identified as mangrove with Sentinel-2, but more often classified as mud flats. The use of Google Earth Engine enabled quicker, faster processing of repeatable methods at larger scales that show that while mangroves are dynamic in many areas where there are both gains and losses, there is an overall net gain of mangrove extent in Mozambique. These semi-automated methods are flexible in that they can be adapted to any landscape and are currently being used by other WWF offices to assess mangrove cover in India, Pakistan and Indonesia to support planning, conservation and restoration efforts.
Long-Term Analysis of Mangroves in Thailand
A team of researchers (SEA-EU-NET consortium) from the Monitoring and REstoration for Sustainable Coastal Ecosystems (RESCuE) project used Google Earth Engine to create annual composites of Landsat series from 1987–2017 for mapping the magnitude of mangrove and its surrounding changes. The study found an increase of shrimp/fish farm of about 8% while mangrove forests had gradually recovered over the same period of time going from 34.20% (77.43 km2) in 1987 to 36.17% (81.90 km2) in 2017. This trend is indicative of the local community’s awareness for mangrove forest conservation and in the detrimental effect that shrimp farming can have on forest conservation. They attributed some of the improvement in mangrove range to a decision by the Thai government to adopt restrictive policy on the expansion of low-salinity shrimp farming within the freshwater regions of the country. For more information about their work, view their paper titled “Landsat Imagery Analysis for Mapping of Mangrove Forests and Its Surroundings in the Trat Province of Thailand”.
Combining Supervised and Unsupervised Techniques to Map Mangroves in South America
Lastly, Cesar Dintz and his collaborators at Solved, ITV and UFPA have created an Earth Engine managed pipeline to compute a spectral index specifically designed to better discriminate mangrove forest from its surrounding vegetation. They call it the Modular Mangrove Detection Index — MMDI. The index is part of a fully automated methodology that aims to systematically map the mangrove forests of South America, from 2000 to 2017 (see flowchart below).
At first, annual cloud free composites are generated and subsetted to include only areas where mangrove forests are likely to occur. Next, inside a narrowed training region, defined by application of a buffer zone (~50 km) over the global mangrove cover data from Giri et al., 2011, a K-means analysis is run, creating a refined sampling area for each one of their land cover classes. Lastly, a Random Forest classifier is used on data sampled within each class to distinguish between Mangrove, Not-Mangrove, Water, and Not-Vegetation.
The result is a systematic and continuously mapped mangrove cover map for each South American country, ranging from 2000 to 2017. Once published, this data will support a better understanding of mangrove dynamics, updating the entire mangrove cover status for the South American continent.
This post has presented a wide variety of approaches to studying mangroves with Earth Engine. They’ve varied in terms of the sensors used, the time scales and spatial extents studied, and the techniques applied. Each gives a new view of the global health of mangroves and how their extent is changing over time. If you’re interested in following developments in mangrove health, many excellent resources are available at Mongabay, Oceana, Smithsonian Institute, and WWF. And of course, keep your eye on the researchers doing the work described here as they continue their efforts to monitor and restore these critical habitats.