Multispectral drone survey maps mangrove biomass in northern Vietnam with high accuracy

A study of the mangrove forests of Dong Rui commune in Tien Yen district, Quang Ninh Province, northern Vietnam, has shown that imagery from a multispectral drone can estimate above-ground mangrove biomass with high accuracy, reaching a coefficient of determination (R2) of 0.831 with a root mean square error of 0.040 in the best-performing model. The work, carried out in the tidal flats between the Ba Che River and the Voi Lon River, combined drone flights with direct field measurements in sixteen standard plots of 10 by 10 metres, and produced a biomass map for the surveyed flight area in which values ranged from 20 to 150 megagrams per hectare. Forest biomass, defined as all organic matter in living and dead form above or below ground, is a central quantity in forest science because forests hold roughly 80 percent of the biomass reserves on land.

Carbon in forest ecosystems is concentrated in living vegetation, fallen material, tree roots and forest soil, and because data below ground are difficult to obtain, most studies concentrate on above-ground biomass, or AGB, usually expressed in dry tonnes per hectare. The traditional way to measure it is direct field measurement in predetermined standard plots, an approach that is accurate but time-consuming, laborious, and especially difficult in remote areas or terrain that is hard to work in. It also cannot deliver a spatial picture of biomass distribution over large areas.

Over the past two decades, remote sensing has been used extensively to estimate AGB, because it reduces data processing time, allows rapid classification over large areas, and makes results less dependent on the judgement of the analyst. Regression analysis is the most commonly used modelling approach, with biomass from sample plots as the dependent variable and spectral features or vegetation indices, such as the Enhanced Vegetation Index, as independent variables. In Vietnam, Landsat imagery has previously been used to calculate mangrove biomass in Quang Ninh Province, and Landsat and Sentinel-2 images have been applied in Thai Binh Province and the Kon Ha Nung Plateau.

However, biomass estimates from such satellite-based agro-forestry studies still carry substantial errors, partly because of model differences and the relatively low resolution of satellite imagery. The new study exploited the improving sensor technology on unmanned aerial vehicles, which can deliver ultra-high-resolution images and carry multispectral sensors with red-edge and near-infrared bands. The researchers used a DJI Phantom 4 Multispectral drone, whose camera records five bands: blue at 450 nanometres, green at 560 nanometres, red at 650 nanometres, red edge at 730 nanometres and near-infrared at 840 nanometres.

Flight parameters were set with DJI GPS Pro software: an altitude of 60 metres, 80 percent image coverage, and a flight window between 7:00 and 8:00 in the morning at the lowest tide, on 15 July 2022. From the drone imagery, the team generated a Digital Elevation Model and a Digital Surface Model from point cloud data. Because the dense mangrove canopy hides the ground, they used GPS data from the field together with gaps in the forest canopy visible in the images to identify ground points and improve the accuracy of the terrain model.

Tree height was then calculated by subtracting the DEM value from the DSM value. In parallel, the Normalised Difference Vegetation Index, a widely used indicator of vegetation vigour, was computed from the near-infrared and green bands as NDVI = (Rnir − Rg)/(Rnir + Rg), producing an NDVI map for the flight area. Four regression formulations were tested to link biomass to the drone-derived variables, combining NDVI and tree height in log-log, log-linear, lin-log and lin-lin forms, where AGB is above-ground biomass and TreeH is tree height.

Field calibration came from the sixteen standard plots, in which tree diameters and heights were measured and biomass was calculated with the allometric equation AGB = 0.251ρ × D2.46, where ρ is wood density and D is the diameter at 0.3 metres for Rhizophoraceae species or at breast height for other species, with all values converted to megagrams per hectare. Pearson correlation analysis and linear regression were then used to identify the relationship between the drone variables and field biomass. The resulting tree height map showed mangrove trees in the bay area ranging from 1.0 to 5.3 metres.

Kandelia obovata, with average heights of 1.5 to 2.5 metres, formed a narrow band prevailing in the low-tide zone along the Ba Che River, while the tallest trees, reaching 5.3 metres, occurred in the north bordering the river, where Bruguiera gymnorhiza dominates and is scattered through the mid-tide zone at average heights of 4 to 5 metres. Rhizophora stylosa averaged 2 to 3 metres, and Aegiceras corniculatum averaged only 1 to 1.5 metres. Moving southwards, away from the Ba Che River, mangrove height tends to decrease.

Among the four candidate models, the Log-Log regression performed best, with the highest R2 of 0.831 and the lowest RMSE of 0.040, and was selected to build the biomass map. Comparison with the sixteen field plots showed a mean deviation of −0.001: four plots matched the field values exactly, six differed by 0.001, and the remaining six deviated by between 0.002 and 0.006. Overall, the model's mean error was 0.002 and its mean absolute error was 0.002, corresponding to 28 percent, while the modelling efficiency index reached 0.91.

The biomass map showed values from 20 to 150 megagrams per hectare, with 52 to 90 megagrams per hectare predominant. The highest values were concentrated in the north, where Rhizophora stylosa and Bruguiera gymnorhiza predominate with average heights of 3 to 5 metres, while low values clustered in the south and southeast, where the short Aegiceras corniculatum stands grow. The accuracy achieved is comparable to or better than satellite-based studies elsewhere.

The Log-Log result matches the XGBoost model applied to mangrove biomass in the Beibu Gulf, China (R2 = 0.8319, RMSE = 22.7638 megagrams per hectare) and exceeds the Catboost model used for invasive mangroves in subtropical Chinese estuaries (R2 = 0.7644, RMSE = 11.1725 megagrams per hectare). It also outperforms satellite-based models such as the XGBR-GA model built on Sentinel-2, Sentinel-1 and ALOS-2 PALSAR-2 data for the Red River Delta Biosphere Reserve in Vietnam (R2 = 0.683, RMSE = 25.08 megagrams per hectare) and the support vector regression model used for the Hyrcanian forests of Iran (R2 = 0.73, RMSE = 38.68 megagrams per hectare). The Dong Rui biomass values themselves sit within the range reported for other mangrove regions: lower than the Sundarbans in Bangladesh, where plot-level AGB ranges from 111.36 to 299.48 megagrams per hectare, but higher than the Magallanes area in Agusan del Norte, the Philippines, where values range from 1.66 to 39.52 megagrams per hectare.

Within Vietnam, Dong Rui is broadly similar to Thai Binh Province (22.57 to 37.74 megagrams per hectare) and Hai Phong City (39 to 100 megagrams per hectare by species), while southern Ca Mau Province is higher, averaging 191.1 megagrams per hectare with a range of 49.6 to 357.4. The authors note limitations and future directions. LiDAR was not yet combined with the UAV in this study, although drone-LiDAR integration is considered one of the most effective tools for determining tree height and building AGB maps for mangroves, with particular advantages in capturing the vertical, three-dimensional structure of the forest.

Ultra-high-resolution imagery can also support species-level classification, allowing biomass estimation by individual species and potentially greater accuracy. Drones themselves have constraints in range, wind resistance and battery life. Even so, the team proposes that UAV imagery can identify mangrove biomass with accuracy high enough to complement or replace traditional field-based standard plots, and that drone-derived reference points can be combined with medium-resolution satellite imagery such as Sentinel or WorldView-2 to map biomass across much larger mangrove areas.

Because mangroves are carbon sinks that also buffer coasts against erosion, storms and waves, and support fisheries and water quality, reliable biomass maps of this kind provide a practical basis for managers to calculate and synchronise carbon service payments, with benefits for local livelihoods in coastal Vietnam.

Subject of Research: Estimating mangrove above-ground biomass in northern Vietnam using multispectral UAV imagery and field calibration plots

Article Title: Application of multispectral UAV to estimate mangrove biomass in Vietnam: A case study in Dong Rui commune, Quang Ninh Province

Article References: Ngo, D., Nguyen, H., Nguyen, K., Dang, C., Nguyen, H., Dang, N., & Pham, T. (2023). Application of multispectral UAV to estimate mangrove biomass in Vietnam: A case study in Dong Rui commune, Quang Ninh Province. One Ecosystem, 8, Article e103760. https://doi.org/10.3897/oneeco.8.e103760

Image Credits: AI Generated

DOI: 10.3897/oneeco.8.e103760

Keywords: mangroves, UAV, remote sensing, biomass, NDVI, carbon, Vietnam, Quang Ninh, multispectral, regression model, forest monitoring, Application