Key finding
The study produced the first national-scale mangrove species map for China using a novel four-step methodology: high-separability image selection from Sentinel-2 time series, uncertainty-controlled probabilistic classification, regional partitioning by latitude, and explicit exclusion of mixed-community pixels rather than forcing every pixel into a species class. Covering China's entire 10-degree latitudinal range at 10 m resolution, the map achieved 83.8 to 86.4% overall accuracy across nine mangrove species, with results comparable to local UAV-based studies.
The map revealed that exotic species such as Sonneratia apetala and Laguncularia racemosa have been widely planted along China's coast, particularly in mid-latitude regions like Zhejiang and Fujian where they outcompete native Kandelia obovata. The methodology is built entirely on freely available Sentinel-2 data, making it replicable in other coastal nations. At 64 citations, this is the most-cited mangrove remote sensing paper of the mid-2020s.
What the research asks
Mangroves are among the most carbon-dense ecosystems on Earth but are notoriously difficult to map at species level because different species often grow intermixed in narrow coastal bands, and species composition shifts dramatically across latitudinal gradients. China's mangroves span nearly 10 degrees of latitude, from tropical Hainan to temperate Zhejiang, making it one of the most challenging test cases for satellite-based species mapping. This study asked whether freely available Sentinel-2 time-series imagery, combined with a methodology that deliberately excludes uncertain classifications rather than forcing every pixel into a species label, could produce a national-scale mangrove species map with accuracy comparable to local field-based studies.
What it finds
Mangrove species grow in mixed, non-monospecific patches across vast latitudinal gradients where species composition shifts dramatically: from 20+ species in tropical Hainan to a single cold-tolerant species (Kandelia obovata) in Zhejiang at 28 degrees north. Traditional species mapping methods struggle with this diversity, but this study's four-step methodology overcomes it by selecting the most separable Sentinel-2 images for each species pair, using probabilistic classification that explicitly quantifies and excludes high-uncertainty pixels, and partitioning the coast into latitudinal zones with distinct species assemblages.
The resulting 10 m resolution map covers all of coastal China, identifying hotspots of both native mangrove diversity and exotic species invasion. Two exotic species (Sonneratia apetala and Laguncularia racemosa) were found widely planted beyond their native ranges, with S. apetala dominating plantations in mid-latitude regions. The methodology cost-effectively uses only freely available Sentinel-2 data — no commercial imagery required — making it replicable by coastal nations with limited monitoring budgets. At 64 citations, it has become the benchmark for species-level mangrove mapping from space.
Why it matters
This paper fills two significant gaps in the site's coverage simultaneously: it is the first paper from Asia-Pacific (the site currently has zero Asian studies) and the first mangrove or blue carbon paper (the site has kelp but no tropical coastal ecosystems). Mangroves store 3 to 5 times more carbon per hectare than tropical forests, making species-level mapping essential for accurate blue carbon accounting, since different species accumulate biomass at different rates.
For practitioners, the methodology is a template for species-level mangrove mapping that can be replicated in other coastal nations using only Sentinel-2 data. The finding that exotic species have been widely planted has direct implications for mangrove restoration projects under carbon crediting programs: planting the wrong species can undermine both carbon storage and biodiversity objectives. For countries reporting under the Global Biodiversity Framework, species-level maps provide the detail needed to track mangrove ecosystem health beyond simple area estimates.