Data-driven comparison of spatio-temporal monitoring techniques

Abstract

Monitoring marine ecosystems is challenging due to the dynamic and unpredictable nature of environmental phenomena. In this work we survey a series of techniques used in information gathering that can be used to increase experts' understanding of marine ecosystems through dynamic monitoring. To achieve this, an underwater glider simulator is constructed, and four different path planning algorithms are investigated: Boustrophendon paths, a gradient based approach, a Level-Sets method, and Sequential Bayesian Optimization. Each planner attempts to maximize the time the glider spends in an area where ocean variables are above a threshold value of interest. To emulate marine ecosystem sensor data, ocean temperatures are used. The planners are simulated 50 times each at random starting times and locations. After validation through simulation, we show that informed decision making improves performance, but more accurate prediction of ocean conditions would be necessary to benefit from long horizon lookahead planning.