Smart monitoring of water quality in Ha Long Bay using remote sensing and artificial intelligence

Ha Long Bay and Cua Luc are two strategic water bodies in Quang Ninh province, valued not only for their natural landscapes and ecosystems but also for their key roles in the province’s economic and tourism development. Ha Long Bay, a UNESCO World Natural Heritage site, is famed for its thousands of majestic limestone islands, while Cua Luc acts as the gateway linking the mainland with the bay and hosts intensive port, industrial, and urban activities. Rapid development has posed numerous environmental challenges, particularly the decline in water quality — an essential factor for marine ecosystems and local livelihoods.

Traditional methods of monitoring and assessing water quality, such as sampling and onsite analysis, often require high costs, considerable time, and manpower, and they cannot cover wide areas or provide continuous monitoring. As a result, in recent years, remote sensing technology has emerged as an effective solution for extensive, time-continuous water quality monitoring.

To apply a sustainable solution for water environmental monitoring, the research team from the Vietnam National Space Center (under the Vietnam Academy of Science and Technology) collaborated with the Institute of Geophysics of Poland (under the Polish Academy of Sciences) to carry out the project “Remote Sensing to monitor water quality parameters in freshwater systems.” This project is part of the bilateral international cooperation mission “Remote sensing in monitoring water environmental quality indicators” (code: QTPL01.03/23-24). The Vietnamese team, led by Dr Vu Anh Tuan, focused on applying remote sensing technology to monitor water environmental quality indicators.

Sharing about the achievement, Dr Vu Anh Tuan explained that this is the first study in Viet Nam to simultaneously use Sentinel-2 data, advanced machine learning algorithms (such as Random Forest, Gradient Boosting, AdaBoost), and the GEE platform to model and monitor water quality parameters such as sea surface temperature (SST), total suspended solids (TSS), chlorophyll-a (Chl-a), and chemical oxygen demand (COD). The outstanding feature of this approach is its ability to monitor over large areas at high frequency and low cost, something that traditional sampling methods cannot efficiently deliver at scale.

Dr. Vu Anh Tuan and Polish members at the Vietnam National Space Center

Effective solution for water quality monitoring

Optical indicators derived from remote sensing data play a crucial role in assessing water environmental quality. Notably, Chl-a serves as a key index indicating phytoplankton growth, which closely correlates with water pollution and eutrophication levels.

The study used Sentinel-2 satellite data (MSI sensor) from 2019–2023, combined with in-situ measurements from the Quang Ninh Department of Natural Resources and Environment and the US National Oceanic and Atmospheric Administration (NOAA), to forecast water quality in the study area. A total of 78 satellite images were processed and analysed on the Google Earth Engine platform. Machine learning algorithms such as Decision Trees (DT), Random Forest (RF), Gradient Boosting Regression (GBR), and AdaBoost Regression (ABR) were applied to predict water quality indicators.

Some pictures of the research team taking water samples at survey points

The results showed that the Random Forest (RF) model achieved the highest accuracy among the tested models, particularly for SST, with coefficients of determination (R²) ranging from 0.79 to 0.80. Other indicators like TSS and Chl-a were also predicted with relatively high accuracy, although COD showed weaker prediction performance due to high variability and complex environmental influences.

The research also identified key spectral bands from Sentinel-2 images, helping optimise the machine learning models and reduce future data collection costs. Based on the models, the team developed spatial-temporal water quality distribution maps, enabling the monitoring of changes and early warning of pollution risks in Ha Long Bay. These maps can be used in water resource management, environmental protection, and guiding the sustainable development of coastal regions.

Map of best and average modeled seawater quality variables over the study period (N.H. Quang et al)

According to Dr Vu Anh Tuan, the study opens a new avenue for applying remote sensing combined with machine learning to monitor water quality, providing effective support for water resource management in critical coastal areas. Despite limitations in field data, the Random Forest model still delivered promising predictions for indicators such as SST and TSS. Integrating water quality distribution maps on the Google Earth Engine platform has created a visual tool that enables spatial-temporal monitoring of marine environmental changes, serving practical management and sustainable development planning needs.

Looking ahead, the team plans to improve model accuracy for indicators such as COD and Chl-a by supplementing measurement data and testing more advanced modelling methods. Additionally, they aim to expand the application scope to other coastal regions nationwide and integrate data from other satellites and field monitoring systems to enhance model comprehensiveness and reliability. Future research will also explore seasonal factors and the impacts of socio-economic activities to further improve the applicability and effectiveness of environmental monitoring tools in the context of increasingly evident climate change.

Translated by Phuong Huyen
Link to Vietnamese version