Algea Bloom Monitoring Based on ML
Harmful algal blooms (HABs) harm marine life and human health. Finding and tracking these events with satellite data is cost-effective and scalable. This study focuses on detecting Alexandrium, a harmful algae species, using data from a small experimental region along the western coast of Italy. All data were provided by Riccardo Bentivogli, a PhD student at the University of Bologna, and the study was conducted in collaboration with him. It looks at two ways to use Convolutional Neural Networks (CNNs) for this task. The first uses a simple CNN to find features in the data for early analysis of cell concentration in regions. The second creates a basic CNN classifier to train and test the data. Before the training, we focused on atmospheric correction and turning images into arrays. The results show that CNNs work well to find patterns in satellite data but have issues like unbalanced data and unclear features.

Atmospheric correction is an important step in satellite image processing. It removes the effects of gases, aerosols, and water vapor in the atmosphere from the captured data. When sunlight interacts with the Earth’s surface, the satellite sensor records both the light reflected from the surface and the light scattered or absorbed by the atmosphere. Atmospheric correction ensures the signal represents the surface accurately.

Satellite sensors typically provide two types of data:
• L1 Data: Measures Top-of-Atmosphere (TOA) radiance [10], including contributions from the Earth’s surface and atmospheric effects. This data is not corrected for atmospheric influences.
• L2 Data: Represents Bottom-of-Atmosphere (BOA) reflectance, where atmospheric effects have been removed, offering accurate surface reflectance.

Sentinel-2 data was used for this research, which can provide L2 products corrected for atmospheric effects using built-in algorithms.