Publikation
On The Optimisation Of Machine Learning Models For Predicting The Photosynthetically Available Radiation In The Water Column
Frederic Theodor Stahl; Lars Nolle; Martin Maximilian Kumm; Christoph Tholen
In: Marco Scarpa; Salvatore Cavalieri; Salvatore Serrano; Fabrizio De Vita (Hrsg.). Proceedings of the 39th ECMS International Conference on Modelling and Simulation ECMS 2025. International Conference on Modelling and Simulation (ECMS-2025), June 24-27, Catania, Italy, Pages 531-537, Communications of the ECMS, Vol. 39, No. 1, ISBN 978-3-937 436-86-9, ECMS, United Kingdom, 6/2025.
Zusammenfassung
Photosynthetically Available Radiation (PAR) is a crucial parameter in oceanography. This study explores the optimization of machine learning models to predict PAR in the water column using selected wavelengths of downwelling irradiance. By leveraging Genetic Algorithms (GA), optimal wavelength combinations were identified for two machine learning models: Linear Regression (LR) and Regression Trees (RT). The models were trained on data from the HE533 expedition and validated using datasets from multiple ship expeditions across different geolocations. Experimental results indicate that the LR model, with an optimal wavelength combination of Ed(469), Ed(501), and Ed(600), achieved the highest prediction accuracy (R² = 0.9992, MAE = 5.78). The RT model, using Ed(433), Ed(586), and Ed(687), performed slightly worse (R² = 0.9954, MAE = 16.37). While both models generalised well to unseen datasets, significant prediction errors were observed for small PAR values at lower water depths.