A Multiarmed Bandit Approach to Adaptive Water Quality Management.

摘要: Nonpoint source water quality management is challenged with allocating uncertain actions and monitoring their performance in the absence of state-dependent decision making. This adaptive context can be expressed as a multiarmed bandit problem. Multiarmed strategies attempt to balance exploitation that appear maximize exploration uncertain, but potentially better, actions. We performed test inform Massachusetts, USA. Conservation restoration practitioners were tasked household wastewater treatments minimize N inputs impaired waters. obtained time series data from 3 treatment types organized them chronologically randomly. The chronological set represented nonstationary based on recent data, whereas random stationary performance. tested 2 hypothetical experiments sample through 20 sequential decisions. A deterministic probability-matching strategy allocated highest probability success regarding at each decision. randomized randomly according compared nonadaptive equally Results indicated equal allocation useful for learning situations tended overexplore inferior thus did not when other strategies. Deterministic matching maximized many situations, adequately explore converged situations. Randomized balanced could converge These findings provide evidence are management. Integr Environ Assess Manag 2020;16:841-852. © 2020 Authors. Integrated Environmental Assessment Management published by Wiley Periodicals LLC behalf Society Toxicology & Chemistry (SETAC).