The Autonomous Monitoring Box (AMB) will be equipped with sensors for the assessment of nutrients, phytoplankton biomass (Chl a), functional groups, CO₂, and CH₄. With Planktoscopes, modular open-source instruments for quantitative, high-resolution imaging, the biogeochemical data will be linked to biodiversity.

The AMB allows for high-frequency data collection both extensively across the water surface and depthwise. In the development phase, the results of the AMB will be evaluated using laboratory measurements.

Development takes place in collaboration with 4H-Jena to create a transferable product. bbe Moldaenke will supply sensorics for the measurement of phytoplankton biomass, functional groups such as cyanobacteria, and for the validation of satellite data from Product 3. 

A CytoSub will generate up to 1,500 in situ images per minute, which can also capture small phytoplankton species, cell abundance with an in situ flow cytometer and will be compared to the cost-effective alternative of the PlanktoScopes.

The CAU will develop an energy-efficient neural network for image analysis (AI product 2) using CytoSub and AMB data. AI-Product 2 will be suitable for deployment on small computers, such as Raspberry Pis, as well as on smartphones, tablets, and laptops directly on site. It will enable in situ data analysis with battery-operated equipment in the field.

Why is a phytoplankton image recognition tool necessary? The traditional method of identifying phytoplankton through microscopy is time-consuming and requires extensive taxonomic knowledge. Therefore, an automatic AI analysis of images could provide stable results over long periods and in different laboratories while saving time. For underwater imaging to function, it is crucial that the AI has low power consumption.

The FAIR image recognition tool is based on previous work in zooplankton analysis and follows deep learning approaches that can benefit from existing databases, such as Ecotaxa. The algorithms will be FAIR, meaning they will be freely available. Due to the lack of research on high-resolution phytoplankton images, we will investigate which organism characteristics are most relevant for automated classification.

For the KIMMCO project, we will collect phytoplankton images ourselves using a Cytosub and a PlanktoScope to train the AI image recognition tool.

Current Situation: Measurements of phytoplankton are usually taken point-in-time and with low temporal resolution, leading to uncertainties in estimating phytoplankton biodiversity and its CO₂ uptake potential.

• NEW in KIMMCO: AI-driven and satellite-based monitoring can enable continuous spatial and temporal monitoring of phytoplankton groups.
• Limitations are: the footprint of satellite measurements is over 300 meters and the depth integral in the water column only goes up to the Secchi depth. The Secchi depth is approximately the lower limit of the euphotic zone, where enough light is present for phytoplankton growth.
• Possible improvements by AI: calibration could lead to improved vertical distribution and spatial resolution.

The project will link satellite data to in situ phytoplankton presence data to improve the calibration of satellite data in the shallow coastal waters of the Baltic Sea. This will allow for the high-resolution, large-scale determination of microalgae and their productivity. Synchronizing shipborne measurements with the overflight times of the correct satellites is necessary for this process. An AI-driven climate visualization tool will clearly illustrate the status of the near-coastal Baltic Sea ecosystem.

To improve the evaluation of satellite data in shallow coastal waters, in situ measurements will be carried out using an above/below-water spectral radiometer (Ramses). Ramses measures incoming radiation from the sun just above and just below the water surface and upwelling radiance, i.e. light coming from the water column below the sensor toward the water surface. These additional light data will allow algorithms to be adapted specifically to the conditions in the Baltic Sea. These data will also be used for CO₂ fixation calculations in AI Product 4.

The visualization tool is intended to present the complexity of the project with the numerous different data-generating devices.

• The monitoring authorities will have access to project data and methods through this tool. This will enhance the transferability of the AI products developed in stages 1-3.
• The public can be informed of and participate in the results through this tool.
• For visualization, proven methods such as, for example, the BELUGA Navigator, ArcGIS StoryMaps, or Digital Earth Viewer are used, which are linked to FAIR repositories to create components for an operational "digital twin".