Over the past 30 years, researchers have used various approximations to address the impact of measurement uncertainty on optimal management policy. This literature has consistently suggested the counter-intuitive proposition that increasing harvest levels in the presence of measurement error is often optimal. Here, we use state-of-the-art algorithms for Partially Observed Markov Decision Processes (POMDPs) to provide the first complete solution to this classic problem, to resolve this paradox.
We show that contrary to previous work, the optimal policy under measurement error is usually more conservative than without it. We demonstrate that management which underestimates the measurement error results in both low economic returns and high frequency of stock collapses, while overestimating the role of measurement error can still result in nearly-optimal economic performance while avoiding stock collapse. We further show on real-world time series data of marine ecosystems that the policy which accounts for measurement error using POMDPs differs significantly from previous methods which ignore such uncertainty.