AMD Monitoring using multispectral imaging from Worldview-3, Sentinel-2 and drone-based data

The sulfidic sulfur contained in the host rocks and mining waste leads to strong acid mine drainage processes in the mining landscapes of Trepca, Kosovo and Pyhasalmi, Finnland. In the present, the water quality is usually monitored by discrete sampling and analysis of dissolved metal particles and other chemical parameters. Not only is this a cost- and time-consuming process, but the assessment takes place only on discrete locations.

The main aim of this application is to elaborate the suitability of multispectral remote sensing (R/S) data from different sensors for area-wide identification and quantitative mapping of Acid Mine Drainage (AMD) constituents such as dissolved iron concentration (Fe³⁺), pH value etc. in water bodies. The potential for mining waste to be subject to AMD processes is also being investigated through area-wide quantitative mapping of the sulfate content (SO₄^2-) in solid ground.

In this framework, water and solid ground samples were collected to calibrate and validate the supervised machine learning algorithm of Artificial Neural networks (ANN), used for the identification of dependencies between the multispectral R/S data and the ground measurements. The ANNs of multilayer perceptron type (MLP) is implemented in the advangeo® 2D Prediction software from Beak Consultants GmbH (www.advangeo.com). The modelling and prediction software analyses complex non-linear relationships between a wide variety of spatial controlling parameters and natural complex processes or occurrences, by using methods of artificial intelligence within a Geographic Information System (GIS) environment.

In the mining landscapes of Artana 1 & 2 and Kelmend, AKG has allocated and analysed about 20 water samples and 15 soil samples between May – August 2022 in two field campaigns, whereas low pH values (3 – 4), dissolved iron concentrations up to 25 mg/L and sulfate contents up to 28474 mg/kg have been recorded. Because of the small-scale features in the mining landscapes, high-resolution multispectral images from Worldview-3 and time-series of drone-based acquisitions are used as controlling parameters for the modelling process.

In the tailing pond of Pyhasalmi and the surrounding water environment, the Oulu University has allocated and analysed about 60 water samples between June – October 2022 in two field campaigns. Low pH values (3 – 4), dissolved iron concentrations up to 1800 mg/L and sulfate contents up to 2200 mg/l have been recorded. In this case, medium-resolution multispectral images from Sentinel-2 (Level-1C TOA and Level-2A BOA products) and high-resolution images from Worldview-3 are used as controlling parameters for the modelling process.

In all scenarios, the imagery was acquired during similar time frames as the sampling, to ensure that the measured water / soil grounds parameters correspond to the surface reflectance information.

In the study, advantages and limitations of different multispectral imaging sensors are elaborated. The newly established dependencies from the ANN models can be used to perform area-wide monitoring of AMD processes in time-series, drastically reducing the need for terrestrial measurements in the future.

The paper has been prepared in the frame of the Horizon 2020 co-funded project GOLDENEYE, which has received funds through the Grant Agreement 869398.