In the summer term 2023, one student group worked on the observation of algal blooms and the associated fish mortality in the Oder River on the German-Polish border in the summer of 2022. Based on raster images, the course of the river in the study area was first extracted and then various indices were applied to detect the algal bloom. Using multispectral images from Planet Labs PBC's SuperDove satellites with eight bands in the visible and near-infrared range with a resolution of 3x3m and daily recording intervals, the group investigated whether more precise results can be derived about the course and origin of the algal bloom.
Extraction of the water body from the raster data was necessary for more detailed investigations of the process. For this purpose, three different methods were used to classify the land use on the basis of the multispectral images. These classifications were carried out using the existing Corine Landcover data, a threshold for the Normalized Difference Water Index (NDWI) and the Random Forest machine learning algorithm. As the machine learning method provided more precise results, particularly in urban areas, these results were selected for further investigation.
The vectorized polygon of the water body facilitated the specific investigation in a time series between 02.05.2022 and 30.12.2022. Using the Normalized Difference Vegetation Index (NDVI), which is based on high reflectance values in the near-infrared range and absorption of waves in the red range, water areas with a high proportion of algal were to be identified and correlated with the chlorophyll-a concentrations with punctual in-site measurements. Alternatively, the Saltwater Algae Bloom Index (SABI), which specializes in marine algae in coastal regions, was also tested.
The temporal progress of the algal bloom could be objectively mapped well on cloud-free days using the indexNDVI. Improved prior homogenization of the data sets could further optimize the method. In order to establish monitoring of algal blooms in rivers using satellite monitoring, the data should also be verified with more on-site sampling.
Authors: Maximilian Rödel, Ricarda Bay, Bastian Habbel
