Suchergebnisse

The current paper estimated the physico-chemical properties of water in the Danube Delta (Romania), based on Sentinel 2 remote sensing data. Eleven sites from the Danube Delta were sampled in spring and autumn for three years (2018-2020) and 21 water physico-chemical parameters were measured in laboratory. Several families of machine learning algorithms, translated into hundreds of models with different parameterizations for each machine learning algorithm, based on remote sensing data input from Sentinel 2 spectral bands, were employed to find the best models that predicted the values measured in laboratory. This was a novel approach, reflected in the types of selected models that minimised the values of performance metrics for the tested parameters. For alkalinity, calcium, chloride, carbon dioxide, hardness, potassium, sodium, ammonium, dissolved oxygen, sulphates, and suspended matter the results were promising, with an overall percentage bias of the estimates of +/- 10% from the observed values. For copper, magnesium, nitrites, nitrates, turbidity and zinc the estimates were fairly accurate, with percentage biases in the interval +/- 10% - 20%, whereas for detergents, led, and phosphates the percentage bias was higher than 20%. Overall, the results of the current study showed fairly good estimates between remote sensing based estimates and laboratory measured values for most water physico-chemical parameters.