The Causal Promise of Decision Intelligence for Wildlife Management

Every wildlife agency runs on decisions, and it rarely makes just one. Should we allow more or fewer deer to be killed this season? Should we expand a habitat corridor? Should we adjust a fishing quota? These repeat year after year, which makes wildlife management one of the best proving grounds for decision intelligence. It is also home to a framework that looks a lot like decision intelligence already: adaptive management. The resemblance is worth examining closely, because the differences point to additional power still on the table for wildlife agencies.

What Adaptive Management Does Well

Since 1995 the U.S. Fish and Wildlife Service has used a framework called Adaptive Harvest Management to set duck hunting regulations.¹ A similar approach has also been used elsewhere, for example the multispecies management of horseshoe crabs and red knots in Delaware Bay.² The idea traces back to Carl Walters' 1986 work on adaptive management of renewable resources, and it treats regulation-setting as a genuinely repeated problem: kill rate affects survival, survival affects next spring's breeding population, and that population feeds into the following year's kill decision, with the cycle repeating indefinitely. What made this a real advance is how it handles uncertainty. Biologists maintain a small set of competing population models, for instance whether hunting-related mortality is additive or compensatory to natural mortality, and use Bayesian updating to shift confidence toward whichever model best matches new banding and survey data each year.³ The regulatory strategy itself is computed with algorithms called stochastic dynamic programming to find the policy that performs best given that uncertainty.⁴ This is rigorous, transparent in its math, and has run continuously for three decades.

Differences with Decision Intelligence

Despite the resemblance, adaptive management and decision intelligence are not the same approach, for several reasons.

1. The causal structure is buried, not elicited.

A CDD is developed by a mixed group of stakeholders, including non-specialists, before any computational modeling begins. Adaptive management's structure instead lives inside statistical population models built by quantitative ecologists. There is no separate, plain-language diagram that a hunter, tribal representative, or state commissioner can look at and revise. The feedback loop between kill rate, survival, and population size is real, but it is a property of any demographic model over time, not evidence of a stakeholder-drawn causal map.

2. Uncertainty is resolved by model averaging, not by testing a human hypothesis.

Decision intelligence uses statistical associative methods to check whether a causal claim proposed by people actually holds up against data. Adaptive management instead compares a short, expert-specified list of candidate model forms and updates the weight on each. That is a powerful kind of learning, but it operates within a fixed menu built by specialists, not against an open causal structure that anyone in the room could have contributed to. This sometimes results in wildlife managers making decisions without fully considering the models' predictions, relying instead on context and experience with the ecosystems in question that lie beyond the models' parameter set.

3. Adaptive management is one optimization pipeline, not a modular structure.

Decision intelligence allows different causal relationships in the same diagram to be filled by different tools: a rule here, an econometric model there, machine learning somewhere else. Adaptive management is a single integrated pipeline of population models plus dynamic programming, purpose-built for one kind of decision (kill quotas) rather than a general architecture that could just as easily host a rule, a market model, or additional optimization algorithms for a different part of the same problem.

4. Adaptive management optimizes for a narrow objective, not the wider decision landscape.

The Adaptive Harvest Management program solves for one thing: the kill policy that best balances population goals given uncertainty, where population goals are set to maximize the kill rate without causing population decline. It was not built to causally reflect how kill regulation connects to habitat funding, disease risk, drought, or land-use change, the kind of broader causal picture a CDD is designed to hold, and, arguably, the context within which such decisions must exist when adopting an ecosystem services approach.

What Decision Intelligence Could Add

None of the above diminishes what adaptive management has accomplished, but a decision intelligence approach could extend it in ways the current framework structurally cannot.

First, CDDs could make the causal structure visible and contestable beyond the population-dynamics community. Habitat biologists, hydrologists, tribal co-managers, and social scientists could add boxes and arrows the current mallard survival models never capture, such as drought severity, disease outbreaks, or land conversion, and challenge assumptions in a shared diagram rather than in a technical report.

Second, CDDs could pull in other repeated wildlife decisions that already use exactly the kind of pluralistic tools decision intelligence is built around, and connect them to the kill decision instead of leaving them siloed. Fisheries agencies already set annual catch limits using bioeconomic models such as Gordon-Schaefer to translate stock size and effort into sustainable, economically sound quotas.^5,6 Conservation planners already use fixed rules, such as IUCN Red List population-decline thresholds, to classify extinction risk without needing a model at all.⁷ Further, optimization software such as Marxan already selects protected-area networks at minimum cost using integer programming.⁸ Right now these sit in separate technical silos. A CDD could tie them into one causal picture of how kill regulation, habitat protection, and species listing decisions interact with each other.

Third, CDDs could make the framework more adaptable to genuinely new causal drivers. Researchers have noted that the fixed model set underlying adaptive kill management struggles as environmental non-stationarity from climate change accelerates, since new dynamics do not fit neatly into the specified models.⁹ A CDD is not locked into a small, pre-built set of candidate structures. New causal connections, and new tools to estimate them, can be added as understanding evolves without redesigning the entire programming apparatus underneath. In this sense, decision intelligence better adapts to shifting resilience constraints at both the species and ecosystem levels as climate change pressures increase.

Integrating DI with Adaptive Management

Adaptive management is a genuine achievement in decision science, and Adaptive Harvest Management specifically has spent thirty years proving that iterative, uncertainty-aware decision making works in practice. But it is a specialized optimization pipeline built by and for population biologists, not a human-centered causal structure open to the full range of people and tools a wildlife decision actually touches. Decision intelligence would not replace it. It would give it a shared, legible spine, one that could connect kill regulation to habitat, funding, and listing decisions that today are optimized separately, and that could welcome new causal knowledge as fast as the world changes. It would fully integrate human-in-the-loop decision-making with quantitative rigor and decades of accumulated data in ways that extend adaptive management's capabilities to true ecosystem resilience levels.

References and Further Reading

1. Adaptive Harvest Management | U.S. Fish & Wildlife Service. https://www.fws.gov/project/adaptive-harvest-management (2022).
2. McGowan, C. P. et al. Multispecies modeling for adaptive management of horseshoe crabs and red knots in the Delaware Bay. Nat. Resour. Model. 24, 117–156 (2011).
3. Johnson, F. A., Kendall, W. L. & Dubovsky, J. A. Conditions and Limitations on Learning in the Adaptive Management of Mallard Harvests. Wildl. Soc. Bull. 1973-2006 30, 176–185 (2002).
4. Complex decisions made simple: a primer on stochastic dynamic programming - Marescot - 2013 - Methods in Ecology and Evolution - Wiley Online Library. https://besjournals.onlinelibrary.wiley.com/doi/full/10.1111/2041-210X.12082.
5. Fisheries, N. Setting an Annual Catch Limit | NOAA Fisheries. NOAA https://www.fisheries.noaa.gov/insight/setting-annual-catch-limit (2022).
6. Aladetan, J. B., Idoko, F. A. & Bamigwojo, O. V. Mathematical approaches to fisheries quota management: ensuring sustainable practices in U.S. commercial fishing. World J. Adv. Res. Rev. 24, 1019–1053 (2024).
7. The IUCN Red List of Threatened Species. IUCN Red List of Threatened Species https://www.iucnredlist.org/en.
8. Marxan. GitHub https://github.com/Marxan-source-code.
9. Chapman, M., Xu, L., Lapeyrolerie, M. & Boettiger, C. Bridging adaptive management and reinforcement learning for more robust decisions. Philos. Trans. R. Soc. B Biol. Sci. 378, 20220195 (2023).

Feature Image by Kev from Pixabay