Bildtext får vara max två rader text. Hela texten ska högerjusteras om den bara ska innehålla fotobyline! Photo: B. Christensen/Azote
Strategies for sustainable management
Back in 1916, Robert Frost spent an entire day ruminating about which path to take on a hike. Even after deciding, he spent an even longer time uncertain that he made the right choice and eventually wrote a poem about it entitled, The Road Not Taken.
More than a hundred years later and we are still trying to figure out: how does one make a decision when plagued by uncertainty?
A small group of researchers from the Stockholm Resilience Centre and the KTH Royal Institute of Technology might be getting closer to an answer.
In a study published in The Royal Society, centre researchers Emilie Lindkvist and Jon Norberg along with their colleague Örjan Ekeberg from KTH present a novel approach to understand how to simultaneously manage and learn about a system when its dynamics are unknown.
While climate change effects on species’ distributions and abundance is a known uncertainty, how climate change will disturb harvested species’ growth rates is often overlooked. Ignoring this latter type of uncertainty may have detrimental consequences such as eventual stock collapse
Emilie Lindkvist, lead author
Rising uncertainties pose additional challenges to the management of renewable resources. This means that management strategies that can deal with unexpected change are becoming increasingly important.
Adaptive management proposes the practice of learning-by-doing (LBD) to improve management during periods of uncertainty such as sudden fluctuations in an ecosystem. In their study, Lindkvist and her colleagues provide a useful insight into different aspects of learning-by-doing (LBD) and to identify robust strategies to improve adaptive management during turbulent times.
The researchers used methods from reinforcement learning to create an artificially intelligent agent. This programmed agent then attempted to learn sustainable management of a renewable resource.
The conceptual model of the agent-resource system is made up of three learning components: a learning model, a mental model and a decision-making model. Each component of the conceptual model can be set to influence the LBD process by varying their parameter values (γ, α, τ). First, the agent decides on an action, next the harvest is received, and in the last step the agent learns from its experience and updates its mental model. The learning goal is to optimise harvests between current and future time steps. (See methodology section below for more information)
They focused on modeling the trade-offs in decision-making with respect to choosing optimal actions (harvest efforts) for sustainable management during change.
The researchers found that different kinds of learning will be useful in three different scenarios of increasing, decreasing and fluctuating resource growth rates.
Increasing resource growth rates provides a tolerant system in which to learn. A fixed strategy can be efficient for increasing growth because adaptation to change is less critical. This activity risks promoting shortsightedness, sticking to past beliefs about the system and modest exploration.
Declining growth rates provides a more unforgiving resource system. This scenario requires faster adaption to lower harvest rates. Learning in this system can only be done efficiently with discount factors, high learning rates and modest exploration.
During periods of fluctuating growth rates, having a fixed strategy that does not respond to growth chances suffered the lowest performance. Management strategies that promoted specialized learning had a higher performance. High value on outcomes, higher learning and modest exploration were all favourable during these uncertain scenarios.
However, the researchers warn, because of the uncertainty inherent to the dynamics of resource systems, we can never accurately predict how the resource dynamics will change. To accommodate this, the researchers propose three key aspects renewable resource managers should consider amid increasing uncertainty:
● Place the same value for present and future outcomes
In most economic decision-making the value of expected future outcomes plays a large role. The results of this work show that a high value of future outcomes is crucial for effectively learning sustainable resource use.
● Strive for higher learning rates to new knowledge.
Change your understanding of what action is more sustainable when harvests are unexpected. For resources with an increasing growth rate, slower adaptation is adequate, however, faster adaptation is advantageous for resources with declining or inconsistent growth rates.
● Conduct modest exploration around what is perceived as the optimal strategy.
To increase the efficiency of the learning strategy in a stable period, a relatively high exploration rate is favourable. For the fluctuating scenarios a relatively low exploration rate was enough. This indicates that during unstable periods, learning by adapting to feedbacks is sufficient.
Based on this, Lindkvist and her colleagues believe this study makes a strong first step to operationalise selected core principles in sustainability science. These include dealing with unexpected events, surprise and radical uncertainty, and incorporate these operationalised principles into enhanced management practice by bridging adaptive management theory and practice.
The researchers applied a novel approach to studying resource growth problems using a computed form of adaptive management to find optimal strategies for prevalent natural resource management dynamics.
The agent model was created through the combination of three learning components, each of which is associated with a specific learning parameter. The mental model stores the agent’s subjective estimate of the total long-term reward for every possible pair of states and next actions. For the decision-making model, they used a weighted random method to select an action at each time step specifically, the softmax policy. For the learning method, they used a temporal difference method where the agent updates its mental model after each interaction with its environment. The resource system was represented by the Gordon-Schaefer model, a nonlinear function with logistic growth, in discrete time.
Emilie Lindkvist’s research interests are focused on using novel simulation models to understand diverse aspects of sustainability in social-ecological systems. Her first line of research explores the emergence of self-organized governance, and the dynamics of trust among fishers.
Jon Norberg works with conceptual development of the resilience and complex adaptive systems (CAS) frameworks. His work spans theoretical ecology, such as the interaction between different response processes such as evolution, species sorting and dispersal, as well as the role of human information networks in solving complex management tasks.