A Bayesian Optimization Approach for Water Resources Monitoring Through an Autonomous Surface Vehicle: The Ypacarai Lake Case Study

Abstract

Bayesian Optimization is a sequential method for obtaining the maximum of an unknown function that has gained much popularity in recent years. Bayesian Optimization is commonly used to monitor the surface of large-scale aquatic environments using an Autonomous Surface Vehicle. We propose to model water quality parameters using Gaussian Processes, and propose three different adaptations of classical Acquisition Functions in order to explore an unknown space, considering surface vehicle restrictions. The proposed Sequential Bayesian Optimization system uses the aforementioned information in order to monitor the Lake and also to obtain a water quality model, which has an associated uncertainty map. For evaluation, the Mean Squared Error of the resulting approximated models are compared. Afterwards, they are compared with other monitoring algorithms, like the Traveling Salesman Problem, using Genetic Algorithms and Lawnmower. Concluding remarks indicate that the proposed method not only performs better while minimizing the Mean Squared Error (via active monitoring), but also manages to quickly identify an approximate of the black-box function, which is very useful for monitoring lakes like Ypacarai Lake ($60\: km^{2}$ ) in Paraguay. Additionally, the proposed method reduces the MSE by 25% when compared with Traveling Salesman Problem-based monitoring algorithms and also provides a more robust solution, i.e., 30% more independent of initial conditions, when compared with known robust coverage methods like the lawnmower method.