Adapting environmental management to uncertain but inevitable change

What is it about?

We propose a new technique to manage uncertain systems that change over time. This approach uses partially observable Markov decision processes to learn the best system model over time. It is particularly useful for environmental systems subject to the impacts of climate change. We demonstrate our technique on a case study of protecting shorebird habitat from the impacts of sea level rise in the East Asian-Australasian flyway, and show that properly accounting for uncertainty can result protecting habitat for 25000 more birds than existing methods.

Featured Image

Why is it important?

Scientists are getting better at modelling the impacts of climate change but the models assume particular future scenarios. We don't know for sure what will happen (e.g. what action we'll take about rising CO2 levels), and this means we have a range of models that can predict quite different outcomes. This uncertainty makes it confusing to decide how to manage populations that will be impacted by climate change. Our method is a way to take good actions now while also learning from our actions to ensure we protect the population in the best way we can given the uncertainty.