Efficient Chlorophyll Prediction and Sampling in the Sea: A Real-Time Approach With UCB-Based Path Planning

Abstract

This study focuses on predicting chlorophyll concentration in the sea, which is a key factor influencing fish populations, oxygen production, and carbon balance in marine ecosystems. Traditional methods for measuring chlorophyll involve time-consuming and costly sample collection and laboratory analysis, making real-time monitoring a challenging task. To address these challenges, the research utilizes real-time measurable parameters, such as temperature and salinity, to predict chlorophyll levels. A feature selection method is employed to identify relevant factors, such as wind speed and conductivity, ensuring accurate predictions with minimal uncertainty. In the second phase of the study, an exhaustive search algorithm is combined with reward functions like Upper Confidence Bound (UCB), entropy, and variance reduction. This combination allows for balancing exploration (sampling across the area) and exploitation (focusing on high-chlorophyll regions). The results show that UCB initially sampled from high-chlorophyll areas but gradually shifted towards broader exploration, achieving a balance between exploration and exploitation. Furthermore, the performance of UCB was found to closely match that of entropy and variance reduction in reducing uncertainty.

Keywords

• Adaptive sampling
• Gaussian regression
• informative path planning