Much of the native prairie managed by the U.S. Fish and Wildlife Service (FWS) in the Prairie Pothole Region (PPR) of the northern Great Plains is extensively invaded by the introduced cool-season grasses, smooth brome (Bromus inermis) and Kentucky bluegrass (Poa pratensis). Management to suppress these invasive plants has had poor to inconsistent success. The central challenge to managers is selecting appropriate management actions in the face of biological and environmental uncertainties. In partnership with the FWS, the U.S. Geological Survey (USGS) developed an adaptive decision support framework to assist managers in selecting management actions under uncertainty and maximizing learning from management outcomes. This joint partnership is known as the Native Prairie Adaptive Management (NPAM) initiative. The NPAM decision framework is built around practical constraints faced by FWS refuge managers and includes identification of the management objective and strategies, analysis of uncertainty and construction of competing decision models, monitoring, and mechanisms for model feedback and decision selection. Nineteen FWS field stations, spanning four states of the PPR, have participated in the initiative. These FWS cooperators share a common management objective, available management strategies, and biological uncertainties. Though the scope is broad, the initiative interfaces with individual land managers who provide site-specific information and receive updated decision guidance that incorporates understanding gained from the collective experience of all cooperators. We describe the technical components of this approach, how the components integrate and inform each other, how data feedback from individual cooperators serves to reduce uncertainty across the whole region, and how a successful adaptive management project is coordinated and maintained on a large scale.

During an initial scoping workshop, FWS cooperators developed a consensus management objective: increase the composition of native grasses and forbs on native sod while minimizing cost. Cooperators agreed that decision guidance should be provided annually and should account for local, real-time vegetation conditions observed on the ground. Over the course of development, two prototypes of the decision framework were considered. The final framework recognized four alternative actions that managers could take in any given year: (1) Graze—targeted use of grazing ungulates to achieve defoliation, (2) Burn—application of prescribed fire as the single form of defoliation, (3) Burn/Graze—a combination treatment, and (4) Rest—no action. The study area included northern mixed-grass and tallgrass prairie. Native vegetation in mixed–grass prairie has a strong cool-season component and thus the dominant native species have a phenology similar to that of smooth brome and Kentucky bluegrass, making management of those species challenging. In contrast, tallgrass prairie has a strong warm-season native component, leading to an existence of cool-season windows, periods of time in the fall and spring when cool‐season invasive grass species are actively growing and vulnerable to damage via select management actions, but warm‐season grass species are not active and are thus less susceptible to damage via the same actions. This dichotomy between prairie types necessitated the development of separate but parallel decision support systems for mixed-grass and tallgrass biomes.

Management units are parcels of native prairie that receive a single management treatment at any one time over their entire extent. At any particular time, the vegetation state of each management unit is characterized by the amount of cover of native grasses and forbs and the type of invasive grass that is dominant. In addition, each unit has a defoliation state which reflects the number of years since the last defoliation event and an index to how intensively the unit was managed during the previous 7 years. State-transition models are used to predict the state of a management unit in year t+1 from its state in year t and a prescribed management action that was applied between the two monitoring events. Alternative models are built around key uncertainties that make choice of a management action difficult. Three uncertainties revolve around whether the effect of management actions depends on (1) type of dominant invader, (2) past defoliation history, and (3) level of invasion. Two additional uncertainties are considered when choosing a management action for tallgrass units: (4) the effectiveness of grazing within the cool-season window as a surrogate for burning when smooth brome is the dominant invader, and (5) the differential effect of active management outside the window as compared to rest.

Because data on the probability of transitioning from one state to another under the various models were lacking, expert opinion and elicitation were used to parameterize the models. In addition, cooperators participated in elicitation exercises to extract their beliefs regarding the value of having native prairie compared to the cost of achieving it. Quantifying the subjective expression of utility in this way allowed for mathematical representation of the management objective into an objective function. By maximizing the objective function, cumulative utility is maximized, leading to the identification of a sequence of decisions that will achieve the management objective.

The NPAM system adopted a vegetation monitoring protocol that was rapid, inexpensive, and familiar to many of the cooperators. The monitoring protocol served three purposes: (1) determining current vegetation and defoliation states of each unit, (2) evaluating progress toward the management objective, and (3) assessing predictive performance of the alternative models. The management year runs from September 1 to August 31. Management can be applied anytime during that period and monitoring takes places from late June to mid-August. Cooperators enter vegetation data and management information into a centralized database by August 25 of each year. Given the current state of the system (vegetation and defoliation states) and the current understanding of the system (or the belief state), identifying the current best management decision is a matter of looking up the combination (that is, system state and belief state) in the appropriate (mixed-grass or tallgrass) optimal decision table. Given complete uncertainty at the outset of decision-making, initial assignment of equal belief weights to each model was believed reasonable. The decisions in the optimal decision table that correspond to the current belief state constitute the current optimal decision policy. By August 31 of each year, individual cooperators are provided with a recommended management action for each of their management units for the upcoming management year. Upon receiving the management recommendations for their units, managers consider the recommendation, along with other relevant information, and at some point during the year one of the management alternatives is carried out. This iterative cycle of making and implementing a management decision, predicting the response, monitoring the outcome, comparing predicted and observed outcomes, updating model weights, and recommending a management action for the next cycle is expected to result in an accumulation of weight on a representative model of system dynamics, thereby increasing understanding needed to effectively manage native prairies.

The NPAM system is now entering its second full year of complete operation, and represents one of only a few fully implemented applications of adaptive management within the U.S. Fish and Wildlife Service. NPAM is truly unique in that it originated from the ground up as a result of the leadership and steadfastness of several refuge biologists and managers confronted with a common problem. These biologists recognized that working together across a large landscape presented perhaps the best opportunity for halting and reversing the invasion of native grasslands by non-native cool-season grasses. Importantly, the NPAM system encapsulates the collective thinking and experience of tens if not hundreds of individuals who have battled this vexing problem for much of their careers.

The NPAM initiative is rooted in principles of adaptive management, thereby affording the opportunity for grassland managers to pursue management objectives while acquiring information to reduce uncertainty and improve future management. The project introduced a number of technical innovations that will serve as templates for conservation efforts throughout and beyond the U.S. Fish and Wildlife Service. First, NPAM is an on-the-ground implementation of active adaptive management—possibly the first of its kind in conservation management—in which recommended management actions result from a prospective analysis of future learning (Williams, 1996). Second, by the use of dynamic optimization, NPAM demonstrates how decisions can be made that take into account possible future transitions of the system. Third, NPAM demonstrates how models of partial controllability are an effective means of accommodating unpredictable circumstances that cause a manager to follow a different course than was intended. Finally, the database developed for NPAM is an unparalleled system that enables the rapid integration of data from the field for the generation of ‘just-in-time’ management recommendations. In all, NPAM provides an example of how a science-management partnership can be forged to achieve large-scale conservation objectives.