Designing forecasts for adaptive management: calibrating language, scale, and need

[Testudinidude]

Although related to #19, #17, #9 (and likely others), this seemed like a sufficient tangent to warrant branching off into its own issue. This will, admittedly, be a little bit broad at present, but should hopefully be conveyed sufficiently to facilitate discussion down-the-line.

The Problem

Ecological forecasting and prediction is well-recognized as a crucial component of adaptive management cycles. Dietze et al. (2018)'s Fig. 1 illustrates quite nicely the way in which iterative forecasting informs consideration of alternative management strategies; in turn, post-management monitoring can provide updated information for recalibrating forecasting cycles.

While the coupling of the iterative forecasting and adaptive management cycles represents a crucial means of translating modelling outputs into management actions, ecological forecasts are not always framed in ways that encourage integration of model outcomes into management activities. This could stem from a few possible causes, which I have tried to list out here:

1. Poorly calibrated spatial and temporal scales: discrete management activities often operate on the scales of weeks and months; as pointed out in Dietze et al. (2018), an increased emphasis on near-term forecasts is necessary for bridging this gap. However, while land management decisions may be guided by larger-scale policy imperatives, boots-on-the-ground management activities are shaped by local/site-level context that may complicate translation of broader-scale model outputs into tangible management recommendations.
2. A lack of cross-walking between management- and forecast-relevant covariates: Williams and Brown (2014) may be a bit dated, but they do argue this point quite well: "the monitoring program must be designed to ensure a tight connection between management objectives and specific monitoring metrics and protocols, so that the data collected are relevant to assessment, learning, and future decision-making." In reality, though, covariates that can be conveniently monitored may not be tractable to manage directly (or vice versa). As one example, urban deer densities are notoriously difficult to estimate, necessitating indirect estimates of relative deer activity through passive approaches (i.e., wildlife cameras). In contrast, most management strategies for deer (culling, sterilization) involve direct manipulation of densities and abundance that may (or may not) affect activity levels across the landscape.
3. Ecological complexity and unintended outcomes: Even when covariates and scales have been appropriately calibrated, the inherent complexity of ecological systems can complicate attempts to reach desired management outcomes. For example, while habitat suitability estimates are used for near/mid-term forecasting, the multi-dimensional nature of habitat suitability means that management activities may have unanticipated consequences for habitat suitability. For example, while habitat suitability for a target species may be higher in areas with open canopy cover, mechanical clearing without sufficient post-management follow-up may lead to encroachment of invasive species and net loss of suitable habitat.

The proposed project

Using existing forecasts available from the NEON forecasting challenge (the tick population forecasts may be particularly relevant here!), generate a heuristic framework of how outputs from the model could be used to guide decision-making. Some elements worth focusing on could be:

• Scale: Are the predictions at a useful scale for application to management needs? If not, can predictions be downscaled/upscaled to an appropriate scale for management?
• Covariate choice: Are the covariate(s) involved in the forecast feasible target(s) for management? If not, are they secondarily related to covariates that could be targeted for management? -Complexity: What management actions could be successfully undertaken to affect the process/pattern of interest, and is that process/pattern strictly governed by covariates included in the model?

A brief plug for other monitoring datasets

While using NEON datasets and projects would be the most efficient use of participant expertise, I will be presenting a poster on the sorts of monitoring datasets commonly collected by regional park systems (such as my own). I would welcome opportunities to discuss the ways in which NEON-wide forecasts could be leveraged for local scale monitoring efforts.

[Testudinidude]

Excited to see this as a focal project! While there are multiple directions that this could take, I think a productive scope of work for the EFI unconference would be the following:

• collaboratively select illustrative NEON challenge themes that have readily recognizable applications for applied management. While I have initially selected the tick dataset, the terrestrial carbon/water flux dataset and challenge would work as well.
• For these selected NEON challenge theme(s), briefly create/adapt conceptual diagrams that describe the input datasets, modelling approaches, and potential outputs for these datasets (in essence, the iterative forecast cycle). At this stage, focus on the more strictly forecasting/modelling components of this.
• Attempt to characterize a probable adaptive management cycle stemming from this iterative forecast cycle. As we do so, we will need to think about how scale, covariate choice, and complexity influence the application of forecasts to adaptive management.

Happy to discuss this further prior to the Unconference, too! Looking forward to chatting about forecasting and adaptive management next week!

[melissakenney]

On the adaptive management side it will be useful to ensure that there are people who actively manage natural resources because the operationalization of forecasts into adaptive management is something that moves it from a framework to practice...

[trashbirdecology]

Following