Remote sensing and artificial intelligence applied to monitoring functional traits of mangrove vegetation in Vietnam
According to Dr Nguyen An Binh, quantitative monitoring of mangrove forest health will support management agencies in detecting early signs of degradation, enabling the development of appropriate conservation strategies. Moreover, this method can be scaled up to monitor coastal mangrove forest ecosystems nationwide, contributing to environmental protection and climate change adaptation programmes.
Dr Nguyen An Binh conducting a field survey with the Dat Mui Protective Forest Management Board
Building on the theory of spectral interactions between vegetation and light, the research team developed an advanced methodology combining radiative transfer modelling with machine learning algorithms to accurately estimate key physiological and ecological traits of mangrove forests. Data processing and time-series analysis techniques were also optimized to accommodate the tropical monsoon climate characteristic of Vietnam, enhancing the reliability of analytical results and laying the groundwork for long-term monitoring and sustainable conservation strategies.
Workflow diagram for estimating functional traits of mangrove forests
As a crucial coastal ecological buffer, mangrove forests not only protect shorelines from erosion, storm surges and rising sea levels, but also regulate the climate, absorb carbon, and sustain biodiversity. However, mangrove forests in Vietnam and globally have been rapidly declining in recent years due to climate change and human activity. Rising sea levels, extreme weather events, land conversion for aquaculture, logging, and pollution have contributed to a global mangrove loss of 20–35% over the past few decades, resulting in serious environmental and socio-economic consequences for coastal communities. Furthermore, this sensitive ecosystem has received relatively little scientific attention due to its limited distribution and inaccessibility.
A view of mangroves at Ca Mau Cape National Park
In this context, accurate and continuous monitoring of mangrove health has become increasingly urgent. Traditionally, mangrove monitoring has relied on field surveys, which, while accurate, are costly, time-consuming, and difficult to scale. The advancement of remote sensing and cloud computing technologies has opened new avenues for monitoring the physiological and ecological characteristics of mangroves with broad spatial coverage and high temporal resolution.
As part of the research project entitled “Quantifying the functional diversity index of mangrove ecosystems using Earth observation satellite data: a case study in Ngoc Hien district, Ca Mau province” (project code: VAST05.03/23-24), the team, comprising Dr Nguyen An Binh, Assoc. Prof. Dr Pham Viet Hoa, and MSc Giang Thi Phuong Thao, collaborated with scientists from the University of Zurich (Switzerland) and the University of Valencia (Spain). The study integrated Sentinel-2 satellite imagery with physical spectral simulation models and machine learning algorithms to establish a robust estimation workflow, embedded within the GEE platform for large-scale and continuous monitoring of mangrove functional traits.
Study area overview: (A) Geographic location, (B) Ngoc Hien district, Ca Mau province, and (C) Sentinel-2 true-colour composites of: (1) newly planted mangroves, (2) natural mangroves, and (3) aquaculture-integrated mangrove plantations
The researchers applied a combined approach using the PROSAIL leaf-canopy radiative transfer model and Gaussian Processes Regression (GPR) algorithms. This state-of-the-art methodology enables the accurate estimation of key mangrove functional traits, including Leaf Area Index (LAI), chlorophyll and water content, and dry leaf biomass. The GPR model was trained and optimised on PROSAIL-simulated data, and enhanced using active learning sample selection techniques to improve accuracy and reduce the need for redundant training data. When validated against field measurements, the models achieved high accuracy, with a Normalised Root Mean Square Error (NRMSE) below 17%.
The study area included a range of mangrove landscapes, such as natural ecosystems (e.g. Ca Mau Cape National Park) and aquaculture-integrated mangrove systems (e.g. shrimp-forest models) in Ngoc Hien district, Ca Mau province. The estimation procedure was tested on Sentinel-2 imagery spanning the last five years. The team also addressed the challenges posed by the tropical monsoon climate by developing data reconstruction techniques to mitigate cloud cover limitations in optical remote sensing and applying seasonal trend-aware time-series analyses.
One of the study’s highlights is the interpretation of the GPR machine learning algorithm and its integration into the GEE platform, enabling automated processing of Sentinel-2 data from 2019 to 2023. Additionally, the researchers applied Whittaker-based reconstruction techniques to fill cloud-related data gaps. As a result, they produced continuous, high-resolution functional trait maps and pixel-level uncertainty estimates, a critical step towards practical applications, particularly as the global scientific community increasingly prioritises open and reliable remote sensing workflows that ensure spatial and temporal data integrity.
The developed workflow is currently available open access via GitHub (https://github.com/thangbomhn87/GEE_Mangrove) and through the PyEOGPR library, hosted on the European Space Agency (ESA)’s Copernicus Data Space ecosystem (https://dataspace.copernicus.eu/analyse/openeo).
Looking ahead, the research team intends to further develop dynamic models of mangrove ecosystems and test the integration of data from multiple sensor sources to improve forecasting accuracy. This work not only holds significant scientific and practical value, but also affirms the importance of applying remote sensing and artificial intelligence to the management and conservation of natural resources in Vietnam, led by scientists from Vietnam Academy of Science and Technology.
The study’s findings were published in the ISPRS Journal of Photogrammetry and Remote Sensing (Q1, IF 10.6, CiteScore 21.0), Issue 214, pp. 135–152, on 15 June 2024 – https://doi.org/10.1016/j.isprsjprs.2024.06.007.
Translated by Tuyet Nhung
Link to Vietnamese version
