Monitoring with multiple goals: Bayesian methods for changing objectives

Long-term monitoring is essential for wildlife conservation. Most wildlife population attributes require long-term monitoring to evaluate. Over the time for attributes to resolve through monitoring, however, information needs change. Existing frameworks to accommodate information need changes, such as adaptive monitoring and management, are built for large-scale, programmatic changes. Often, smaller, rapid changes are necessary. Fortunately, information needs can change predictably in wildlife monitoring, even when little is known about populations. Predictable changes include the desire to answer: 1) is the species present?; 2) are multiple individuals present?; 3) is breeding occurring?. We suggest long-term monitoring can accommodate these changes. We propose Goal Efficient Monitoring (GEM), an approach that uses a Bayesian integrated population model (BIPM) to accommodate changing information needs through: a BIPM that links population state changes (e.g., present, multiple individuals present) to population dynamics (e.g., abundance, demographic rates); and sampling rules to allocate effort observation effort based on current knowledge. To test the efficacy of a GEM approach, we ask two research questions: 1) can implementing a GEM approach provide robust population estimates?; and 2) do GEM sampling rules in multiple long-term monitoring settings (i.e., population sizes) accommodate changing questions while providing continual, reliable population inference? To answer these questions, we built a BIPM and conducted a simulation study for a rare species in the US, Canada lynx (Lynx canadensis). We simulated lynx populations under five different starting conditions and simulated a GEM approach (10 years of simulated observations with GEM sampling rules), then used our BIPM model to produce estimates and predictions. In 93 % of simulations, 95 % credible intervals for BIMP estimates contained the true value for all biological (abundance of all sexes and age classes, birth events, survival, state transition probabilities) and observation variables (detection probabilities). We demonstrate how a GEM approach can provide reliable long-term inference while being responsive to shifting information needs.