The Derivation of Irish Water Quality via Sentinel-2 Imagery

The water quality in Republic of Ireland is regulated under the water framework directive (WFD), which requires all EU countries achieve good ecological and chemical status before 2027. However, the Irish Environmental Protection Agency (EPA) reports in 2021 that just half of the rivers, lakes, estuaries, and coastal waters achieved satisfactory or higher status.

Water quality in Ireland is monitored by traditional methods, which cannot provide timely spatiotemporal information. While remote sensing (RS)-based water quality monitoring work have been carried out in many EU countries in accordance with WFD directive, the use of RS for water quality estimation in Ireland has not been fully explored.

To explore the feasibility of RS for Irish waters, Sentinel-2 surface reflectance has been used to assess several water quality parameters (chlorophyll-a, transparency, turbidity, suspended solids (SS), total nitrogen, total phosphorus, biological oxygen demand (BOD), dissolved oxygen, and chemical oxygen demand (COD)) from March 2017 to July 2022. These were compared with the Sentinel-2 surface reflectance data resulting in a total of 6509 corresponding data points.

Initially, empirical algorithms were used to derive water quality concentrations in rivers, lakes, estuaries, and coastal waters separately. Initial results indicate that the combination of green and blue bands was correlated to coastal waters’ chlorophyll-a (R² = 0.27). For chlorophyll-a in transitional waters, the combination of red and red edge was highly correlated. However, no single band or combination were suitable for deriving chlorophyll-a in lakes. For SS, red and near infrared band are useful in detecting changes in coastal and transitional waters. Whereas, for lakes and rivers, blue and shortwave infrared band were best to derive SS. In addition to empirical algorithms, multiple machine learning methods have been used to derive water quality parameters from Sentinel-2 reflectance, with the aim of exploring if machine learning approaches can improve estimates compared with the empirical approaches.