Thresholds and what to do when we reach them

During the COVID-19  outbreak, I will be posting information and hosting discussions related to my work with the Parks Canada Agency. The lack of access to internal documents makes this interim measure necessary. The information shared here is not classified.  The opinions expressed do not represent those of the Government of Canada.

Well-considered thresholds in ecological monitoring are critical to success in having scientific observations make an impact on the management of protected areas. Thresholds perform the dual function of simplifying the reporting process on the health of protected areas and spurring land and marine managers on to effective action.  Setting thresholds is difficult. Though identifying important species or processes in a protected area is usually straightforward, it is not  so simple to decide what is a healthy abundance for a given species or a sustainable frequency for a process like fire or flooding. A further important step is interpreting the condition of landscapes and ecosystems as the result of a series of threshold-based assessments. Which indicators are most important? Which outcomes trigger conservation action?

How to choose a threshold?

Thresholds are required in the Ecological Integrity monitoring program for Canada’s National Parks.  The United States National Park Service’s Vital Signs monitoring program did not, at first,  require thresholds but has since found them useful in the related Natural Resource Condition program.  It is clear that the managerial convenience of thresholds outstrips the evidence that tidy thresholds for action occur in nature. Nonetheless, it is in the nature of decision support tools to bring the salient features of a system into focus.  Thresholds are very useful in that regard.

There are essentially three ways to justify a threshold:

1. Natural Variability – this statistical approach is the most common way to identify departures from natural levels.  With as little as four years of previous observations, a current observation can be assessed as being close to average (e.g. within 1 standard deviation) or exceptional (e.g. more than 2 standard deviations from the average).
2. Reference Condition – this approach relies on the existence of a well-studied, pristine site within the local region.  A departure from the reference condition either in standard deviations or percentiles is a good reason to signal concern for a protected area.
3. Absolute Threshold – this approach resembles the reference condition approach but is based on first principles (e.g. toxicity, nutrient dynamics) and is not based on local reference sites.

The document –  Thresholds Open Data 2020Jan30 – provides 135 examples of ecological integrity thresholds used in the Parks Canada monitoring program.

How to combine assessments at the landscape or ecosystem level?

Pulling together actionable information for a protected area manager means bringing together the relevant observations within the local ecosystem.  Ensuring that there is “enough” information to describe an ecosystem is quite difficult. Here, I assume that the monitoring program has prioritized the most relevant indicators in choosing its measurements.  This assumption is sometimes violated due to the cost of monitoring or the political implications of a given species or ecological process. Nonetheless,relevance is usually an important factor in monitoring design, so the information at hand should be at least a good starting place for decision making.  There are a couple of approaches for combining assessments of different indicators:

1. Equal Emphasis – this approach simply reports the average condition of the identified indicators. Parks Canada uses this approach for reporting on ecosystems combined with a precautionary principle that is added to avoid confusing mediocre results with a mixture of good and poor results.
2. Weighted Average – this approach has been used in the Natural Resource Condition program and is being considered by Parks Canada for summarizing ecological integrity. It requires a quantification of the importance or relevance of each indicator but mostly ignores the various connections between indicators.
3. Decision Tree or Model-based Approach – this approach, objectively the most detailed and appealing for scientists,  requires a consideration of which combinations of indicator assessments should lead to action. There are variety of approaches to building up the network of contributing factors leading to a negative conservation outcome.  Most of them work backwards from a desired state, including adverse outcome pathways, result chains, or decision trees.  All of these approaches require extensive information about the landscape or ecosystem in question.

How do you apply thresholds?

What is your experience with thresholds? Is the added effort of drawing a line between good and poor outcomes (and perhaps several other categories in between) worthwhile for decision making? Do policy leaders and managers tend to different conclusions than scientists regarding thresholds?

What level of information is necessary before it contributes usefully to decision making? Is the gold standard the only one worth striving for? Should we focus on intensively monitoring high value resources or work broadly from an ecosystem perspective?

Please add your comments below.