Early Warning Signs of Risk to Populations
The previous chapters have attempted to establish that scientists may anticipate the nature of some interacting effects, but in most situations they are not currently able to forecast the cumulative effects of all stressors with any accuracy. Therefore, there is a pressing need for early detection of unexpected population declines and, where possible, rapid diagnosis of the main factors contributing to them. This requires some form of population monitoring. The parameters monitored must be informative about the status of the population; it is also helpful if they are informative about the contributing factors for any decline in status, although that could become part of a secondary, more intensive, data-gathering effort that is instigated if the first stage of monitoring indicates a problem. (An alternative view is given in the following paragraph.) Detecting a deleterious situation involves testing for long-term declines in status over time (trend analysis; see, e.g., Thomas et al., 2004), or a recent sudden drop (sequential surveillance; see, e.g., Anderson and Thompson, 2004; Frisén, 2009). Alternatively a comparison could be made with reference to populations thought to be in good status, although such comparisons need to consider natural variability. The parameters monitored must also be measured with sufficient accuracy and precision that there is a good chance a deleterious change of magnitude large enough to cause concern will be detected (i.e., good statistical power, if a statistical hypothesis test is the detection mechanism).
The above approach has been criticized as being inefficient and ineffective by Nichols and Williams (2006), who refer to it as “surveillance monitoring.” They argue that a focus on detecting declines, often using statistical hypothesis testing, is unlikely to lead to optimal conservation decisions and introduces unnecessary time lags, and that identifying the causes of declines is less important than identifying the most effective remedy (although recognizing the cause can often help identify possible solutions). Instead, they advocate embedding monitoring within a larger framework of conservation-oriented science or management, where monitoring is used to enable discrimination between multiple competing hypotheses about the biological system being monitored and hence facilitate better management decisions. Monitoring therefore becomes an integral part of an adaptive management framework, as defined in the previous chapter. This also implies that monitoring programs will change what is measured as the scientific hypotheses under consideration are updated—a paradigm called “adaptive monitoring” by Lindenmayer and Likens (2009).
The committee believes that there is merit in both of these frameworks. Adaptive management, and hence adaptive monitoring, potentially can be effective in situations where there is enough knowledge of the system to formulate working hypotheses about the link between each potential management action and the outcome, to evaluate the a priori probability of each hypothesis, and where learning through focused monitoring will be useful. However, there are at least two reasons not to rely exclusively on such adaptive monitoring. First, there are many cases where the above criteria will not be met and adaptive management will not be helpful. Second, as described in Chapter 6, there is a strong potential for “ecological surprises,” for example, unexpected declines in species that had not previously been considered to be of conservation concern. Hence, a dual approach is advocated, where the principles of adaptive management and adaptive monitoring are applied where possible, but where, in addition, a “light touch” surveillance program is undertaken in order that very large changes in conservation status of species are not missed until it is too late to do anything about them. It is recognized that such a surveillance program will have low
power, but its aim is to detect only large changes in status. The chance of detecting a change in status will be improved if a sensitive indicator can be found that is also relatively inexpensive to monitor.
The committee has previously recommended the use of adaptive management (Recommendation 6.1) to focus data collection and guide management actions. The following recommendation concerns a “light touch” surveillance program.
Recommendation 7.1: Responsible agencies should develop relatively inexpensive surveillance systems that can provide early detection of major changes in population status and health. Surveillance systems should be developed first for populations that currently lack adequate stock assessments.
In the following sections, the population parameters that might best be measured in either of the above frameworks are discussed. One form of ecological surprise described earlier is that of an ecological tipping point. In the last section, suggestions from the literature on the early detection of a species or system approaching a tipping point are described.
MONITORING POPULATION SIZE
Population size is the most basic measure of population state. However, for most marine mammal species, monitoring total population size (or density) over time or space is not a sensitive way to obtain early warning of problems (for surveillance monitoring) or distinguish between different possible management actions (for adaptive monitoring). One issue is that it is often difficult to define what constitutes a biologically appropriate unit of assessment because many local populations are not genetically or demographically isolated. Another is that most marine mammal species are long lived and slow to reproduce, so any negative impact that causes reproductive failure or juvenile mortality, or any beneficial management action, will take a very long time to cause a significant population trend. However, the main issue is that population (or stock) size is a parameter that is notoriously difficult to measure precisely, particularly for marine mammals that often range over a large area and are invisible when underwater. Visual methods requiring human observers remain the most commonly used for marine mammals, particularly cetaceans—either shipboard or aerial line transect surveys or photographic capture–recapture (Buckland and York, 2009). For colonial pinnipeds, colony counts are sometimes used, with a correction factor (derived from animal-borne tags) for those at sea (Buckland and York, 2009); for some pinnipeds such as grey seals, pup production at breeding colonies is estimated and a population dynamics model is used to scale up to total population size (e.g., Thomas et al., 2005). For animals that are widely dispersed, it tends to be the spatial variation that causes low precision; for rare or hard-to-see animals it is the low sample size; for colony counts it is estimating the scaling factor. The result is that the ability to detect all but the most drastic population trends is often limited. For example, Taylor et al. (2007) reviewed the precision of abundance estimates for 127 stocks under U.S. management and concluded that, overall, 70% were not precise enough to detect a precipitous decline of 50% over 15 years of monitoring. Jewell et al. (2012) examined the utility of combining results from multiple abundance surveys worldwide: for the best-fitting model, the smallest population decline detectable with high (>0.8) power was more than 50% for 5 out of the 11 taxonomic and geographic groupings used.
Despite this pessimistic message, more precise monitoring is possible for some stocks, particularly those that live in restricted areas relatively close to shore (e.g., southern resident killer whales) or all pass close to shore at some point in their life cycle (e.g., gray whales). New technology may also play a part in enabling more precise population estimation—for example, potentially replacing visual surveys with remote aerial vehicle surveys using high-definition cameras or video recorders (Buckland et al., 2012) or passive acoustic surveys from fixed or floating sensors, or remote underwater vehicles (Marques et al., 2013). Many of these techniques are still under active development; for passive acoustic methods a critical limitation is knowledge of the acoustic biology of the target species required to convert call density into animal density and abundance. New statistical methods that make better use of existing or emerging data streams also offer the potential for better precision—for example, the recent ability to extend capture–recapture analysis to utilize information about the location of the captures (Borchers, 2012; Royle et al., 2013; Pirotta et al., 2015c). Taylor et al. (2007) discuss some other potential routes to increased precision. However, it is important to emphasize that, at the current time, estimation of population size remains a very imprecise science for almost all marine mammal stocks.
One possibility sometimes suggested for obtaining more precise estimates of population status is to measure indices of population size, such as uncalibrated acoustic detections and sightings from shore-watch schemes or from platforms of opportunity. However, straightforward interpretation of changes in the index as changes in population numbers requires that the relationship between the two is linear and has constant variance over the range of both indices, or that the shape of the relationship and variance is known (Williams et al., 2001, Section 12.7). In practice, the relationship is rarely linear (indeed it may not even be monotonic) or with constant variance. Nevertheless, carefully chosen indices may still be effective as early warning metrics, for example, if they are sensitive to changes in population size or disturbance for the species of interest and are relatively inexpensive to deploy at the population scale. Passive acoustic detections may be a good candidate in this regard, in that large amounts of data can be collected at moderate
expense (for vocal species); however, its efficacy has yet to be demonstrated.
In determining the cause of population declines, it is often insightful to focus on the components of the population likely to be affected first. This is discussed in the next section.
MONITORING DEMOGRAPHIC PARAMETERS
Population dynamics are governed by four fundamental demographic parameters: survival, fecundity, immigration, and emigration. One or more of these must decline (or increase in the case of emigration) for population declines to occur. Hence, measuring these parameters may make for a more sensitive monitoring system than waiting for a detectable change in population size. However, it is typically infeasible to monitor all of these parameters with good precision, so one will typically need to prioritize. To do so, one needs to consider which of these parameters is expected to be most strongly affected by cumulative impacts of stressors, the influence changes in these parameters have on population size, and the feasibility of accurately measuring the parameter.
Many marine mammals are relatively long lived and reproduce infrequently but over multiple occasions. Under these circumstances, ecological theory leads us to predict that reproductive-age adult females should evolve strategies that enable them to delay breeding or abandon investment in young when conditions are harsh in order to prioritize their own survival and hence maximize their future reproductive output when conditions may be better. Therefore, there is an expectation that adult female survival will remain high and relatively constant in fluctuating environments, while fecundity and calf or pup survival should fluctuate with the conditions. A similar phenomenon occurs as populations approach carrying capacity and, based partly on empirical observations, Eberhardt (2002 and references therein) proposed the following sequence of changes as conditions worsen:
- increase in mortality rate of immatures
- increase in age of first reproduction
- reduction in reproductive rate of adult females
- increase in mortality rate of adults
The committee’s opinion is that there is no strong theoretical reason to suggest that pup or calf mortality should always increase before fecundity-related parameters decrease; this may depend on the cost of pregnancy and gestation, and whether the species is adapted to uncertainty in the ability to provision young. For species where these costs are low, and that are adapted to uncertain provisioning conditions, adult females may tend to continue to produce pups or calves but then not be able to successfully rear them. Hence, from an early warning perspective, fecundity (including age at first breeding) and calf or pup survival are all parameters to target.
To determine influence on population size, it is useful to consider the findings of matrix population modeling (Caswell, 2001), in particular from sensitivity analysis, which quantifies how much population growth will be affected by identically sized changes in each demographic parameter in the model. Exact results depend on the model, but in general, population growth is most sensitive to changes in adult survival, with changes of the same magnitude in fecundity and pup or calf survival having much less effect (Eberhardt, 2002).
Putting these last two threads together it is expected that birth rates and/or pup or calf survival are likely to be first affected by cumulative stressors, but that they will have the least effect on population growth rate. This provides a strong justification for monitoring these parameters as part of an early warning system, where they may show a strong signal of population stress before the population trajectory is strongly affected. However, it is important to recognize that natural population processes such as density dependence will also result in low birth rates and/or with pup or calf survival, and hence measurements need to be put into the context of natural population dynamics. Also, as stated earlier, these demographic parameters are expected to show the highest levels of natural variation, so picking out a declining trend among strong interannual variation may be difficult.
The last consideration is the feasibility of accurately monitoring the parameters. Many demographic parameters can be estimated from an intensive capture–recapture survey; typically for marine mammals this involves photographic identification, although genetic identification from biopsies or fecal samples (or even potentially blow samples) is possible. Each of these methods is labor intensive, and only feasible in situations where animals are accessible and a reasonable recapture rate is likely. In planning a study, the expected precision can readily be evaluated using a straightforward simulation approach (Devineau et al., 2006).
Age-specific mortality can also be derived from analysis of age structure of a population, assuming a stable age structure (as in when the population is growing exponentially, or has reached carrying capacity); this is the basis of life-table analysis. One example of this is Moore and Read (2008), who used the age structure of harbor porpoise deaths from all mortality sources and the age structure of deaths from fisheries bycatch to estimate the effect of bycatch on vital rates and the likelihood of population decline. The use of strandings is, however, problematic due to the length of time required to obtain a sufficient number of carcasses for age structure analysis, and the fact that it can only be used on inshore populations in areas where stranded carcasses are reported and can be investigated. For this reason it cannot be recommended as a general monitoring method.
Fecundity (or at least pregnancy) can also potentially be estimated from hormone analysis (e.g., Kellar et al., 2006;
Hunt et al., 2014) and from looking at pregnancy rates (and possibly pregnancy history) of stranded or sampled animals. However, high pregnancy rates alone may not mean good population status: if calf or pup survival is low then females do not need to devote energy to provisioning their young and hence may recover and breed again more quickly—thus elevating pregnancy rates. Hence pup or calf survival should also be measured.
Overall, although birth rates and pup or calf survival seem at first glance to be the best parameters to monitor for early warnings, it will be important to undertake some form of power or precision analysis to determine whether a signal of the expected magnitude can be detected given expected levels of interannual variation and measurement error.
Another generally applicable approach is to focus on indices of demography that can readily be measured in the field. One prominent example is the ratio of adults to juveniles in a sightings survey (or, relatedly, the proportion of mother–calf pairs in populations where this is an appropriate metric). Calves or pups are typically readily distinguishable from adults; it may also be possible to distinguish juveniles and record similar metrics on them. In conclusion, collection and analysis of stage-structured population data may provide a useful early warning of poor population status.
MONITORING POPULATION HEALTH
Chapter 5 provided a definition of individual health, as well as reviewing some of the various indices used to assess individual health. However, it is important to distinguish between assessing the health of an individual versus assessing the health of a population, the latter being focused on the measurement of the distribution of health outcomes in a population or a subset of a population, as well as the determinants or factors that influence those outcomes (Ryser-Degiorgis, 2013). The term “health outcomes” is used rather than the more narrow term “health status” because the latter refers to health at a single point in time rather than over a period of months or even years that it may take for a disease to develop (and demographic consequences to become manifest) (Kindig and Stoddart, 2003). As a field of research, population health focuses on multiple potential contributing factors for health outcomes; it considers the complex interactions among factors, the biological mechanisms underlying a given health outcome, and the influence of different factors over time and throughout an organism’s life cycle (Kindig and Stoddart, 2003; Ryser-Degiorgis, 2013). In this respect, population health studies not only address the detection of changes in health outcomes, but also simultaneously address the potential causal factors.
The concept of population health involves different criteria from population status. The National Marine Fisheries Service (NMFS) assesses the status of a marine mammal population or “stock” by assessing its range, minimum population estimate, current population trends and productivity rates, human-caused mortality, and other factors that may cause a decline or impede recovery (NMFS, 2004). Populations that are large and near carrying capacity will usually have a good population status but could have a lower level of population health. A population that is at or nearing carrying capacity may exhibit a high prevalence of disease (e.g., malnutrition or infectious disease), and the population’s size in relation to its expected carrying capacity should be considered as a potential driver when poor population health is observed. In this context, population health (i.e., the distribution of health outcomes in a population or a subset of a population) may produce a false-positive indication of population decline. While this chance of false positives for populations for which status is completely unknown decreases specificity, population health will in most cases provide greater sensitivity and is a more tractable approach as compared to monitoring population status, which requires precise estimation of population size and current productivity rate in relation to an expected productivity rate. Carrying capacity is generally not known and is difficult to estimate. However, the objective of monitoring as outlined in this chapter is early detection of population declines. If poor population health is observed, continued monitoring over time would allow the hypothesis of carrying capacity being the underlying driver to be confirmed or rejected.
Population health monitoring can take two primary forms: passive health surveillance (also referred to as scanning surveillance) and targeted health surveillance. Passive health surveillance focuses on in-depth investigation of disease incidence and for wild marine mammals is generally conducted using carcasses or tissues collected from stranded animals. In the United States, under the 1992 Amendments to the Marine Mammal Protection Act, the Marine Mammal Health and Stranding Response Program (MMHSRP) was formalized to coordinate efforts to investigate marine mammal strandings.1 The intent of the program is to improve the knowledge of rates and causes of mortality and morbidity to gain a better understanding of population threats and stressors, and to detect emerging or unusual events. Since 1991, 62 marine mammal unusual mortality events (UMEs) have been recognized in the United States,2 and in those where causes have been attributed (only 56%), these have included biological toxins, infections, human interactions, oil spills, and changes in oceanographic conditions (Gulland and Hall, 2007). An additional important component of the MMHSRP is biomonitoring, i.e., sampling, archiving, and analysis of tissues to allow for examination of geographic and temporal patterns in exposure to chemical contaminants, biological toxins, and/or pathogens (e.g., Fire et al., 2009; Twiner et al., 2012; Simeone et al., 2015). A real-time, nationally centralized system for reporting marine mammal health data has been proposed (Simeone et al., 2015) and would
1 See http://www.nmfs.noaa.gov/pr/health/MMHSRP.html.
greatly facilitate the conduct of epidemiological analyses to more rapidly detect and identify contributing factors for UMEs, as well as to explore more subtle changes in population health over space and/or time in relation to one or more stressors. Standardization of databases for marine mammal health within and across nations could facilitate more global analyses. However, with the exception of nearshore species, the utility of passive surveillance for marine mammal populations will still be limited due to the extremely low probability of recovering carcasses (Williams et al., 2011; Barbieri et al., 2013; Carretta et al., 2015).
Recommendation 7.2: A real-time, nationally centralized system for reporting marine mammal health data should be established.
In contrast, targeted health surveillance is carried out proactively, focusing on live animals that in some cases are apparently healthy, and relying primarily on cross-sectional study designs that require only a single sampling occasion (Ryser-Degiorgis, 2013). Targeted health surveillance in the form of capture–release health assessment has been successfully conducted for a number of species along the U.S. coast (e.g., Wells et al., 2004; Aguirre et al., 2007; Greig et al., 2010). Physical examination, diagnostic ultrasound, and blood sampling for hematology, serum biochemistry, and hormone analysis can be conducted and synthesized to determine the prevalence of specific disease conditions (Schwacke et al., 2014a), and serology (to determine antibody prevalence) can help to evaluate prior pathogen exposure, or lack thereof, assisting in the development of management plans (M. Barbieri, personal communication). Portable auditory evoked potential systems also allow for hearing tests (Finneran and Houser, 2007) to be performed, which are particularly relevant for understanding hearing loss among various populations. Unfortunately, capture–release studies can only be conducted on relatively small, tractable marine mammal species, and to date have focused on the nearshore where individuals can be temporarily caught and restrained on land (e.g., seals and polar bears; Stirling et al., 1989; Polischuk et al., 2001) or in shallow waters (e.g., small delphinids, and manatees; Bonde et al., 2012). However, methods could and should be developed to extend such sampling to other coastal, continental shelf, and/or oceanic species, although an extension of these types of approaches to large cetaceans will be complicated by the logistical challenges of capturing and restraining them. Nevertheless, remote sampling techniques are rapidly advancing and can be applied to large cetaceans. Hunt et al. (2013) review currently available techniques for obtaining physiological information on large whales that include remote collection of respiratory (“blow”) samples, skin/blubber samples, and fecal samples. Perhaps most promising is the collection of blow, as techniques for analysis of metabolites, hormones, and pathogens have been demonstrated using cetacean respiratory samples (Acevedo-Whitehouse et al., 2009; Hunt et al., 2013; Aksenov et al., 2014; Cumeras et al., 2014), and recent developments in human breath analysis indicate promise for eventually obtaining a broad array of physiologically relevant indicators of health (reviewed by Hunt et al., 2013). However, collection methods are still being refined and will require extensive validation as well as collection of baseline samples to understand the inherent variability for the suite of measures across species, life-history stages, and varying environmental conditions. Likewise, “-omics” approaches (primarily proteomics and transcriptomics) are being pursued using sampling matrices that can be remotely collected (blow, skin/blubber; reviewed by Hunt et al., 2013), but characterization of expression profiles is still in its infancy, and identifying patterns that provide meaningful information on health state is complicated by lack of information on cetacean genomes (Hunt et al., 2013), variation among life-history stages, genetic stock, and varying environmental conditions (e.g., Van Dolah et al., 2015), and the fact that some remotely collected samples (i.e., skin/blubber) simply may not be appropriate matrices for detecting expressional changes associated with many health conditions.
Targeted surveillance could also be supported through photographic studies. Photographic monitoring has been used to identify emerging zoonotic disease (Rotstein et al., 2009) and support epidemiological investigations of skin disease in both terrestrial (e.g., Oleaga et al., 2011) and marine mammals (e.g., Hart et al., 2012; Van Bressem et al., 2015). Visual health assessment based on body and skin condition, and the presence of cyamids and rake marks, has been applied for right whales (Eubaleana glacialis), and an index of health based on these criteria has been developed that is predictive of survival and reproduction (Schick et al., 2013). In addition, Fearnbach et al. (2015) have applied photogrammetry to assess body condition based on proportional head width in endangered Southern Resident killer whales (Orcinus orca). Furthermore, recent development of techniques to obtain photographs using unmanned aircraft systems (Durban et al., 2015) will greatly facilitate photographic monitoring to measure body condition and/or assess parasites, skin disease, or other externally visible indicators of compromised health.
These novel health assessment methods are primarily designed to be applied to individuals, but because population health emerges from the health status of a population’s members, appropriate sampling at the individual level can lead to inferences about population status. In this vein, body condition, as measured by a visual health assessment or photogrammetry (see above paragraph), could represent a first-pass metric for overall population health. Sampling would need to include a sufficiently large number of animals to assess the health of groups critical to population growth, such as a large cross-sectional sample of adult females across a variety of life-history stages or of juveniles. A broad measure of health, such as body condition, would not necessarily
be sensitive to quick changes because fat reserves may not be affected until the late stage of a disease; however, because most pathways of declining health eventually affect body condition, it could capture the consequences of a variety of potential stressors.
One important caveat here, just as with measuring demographic parameters, is that care needs to be taken not to misinterpret poor health caused by natural demographic processes, such as reaching carrying capacity, with poor health that is of concern; in other words, measurements need to be put in the context of expectation given the population status.
EARLY WARNING OF TIPPING POINTS
As described in Chapter 6, the existence of multiple stable states and tipping points in natural ecosystems is now beyond reasonable doubt. However, the real challenge for managers and scientists alike is the ability to anticipate and predict regime shifts, especially as the impacts of anthropogenic stressors and drivers on ecosystem function and processes appear to be increasing. The potential for predicting regime shifts in marine environments and their management depends on the characteristics of the regime shifts: their drivers, scale, and potential for management action.
Recent theoretical findings (Drake and Griffen, 2010; Dai et al., 2012; Dakos et al., 2015) suggest that ecosystems tend to recover more slowly from small perturbations if they are in the vicinity of tipping points. This phenomenon is referred to as “critical slowing down,” and its temporal and spatial indicators may under some conditions provide early warning signals of a system approaching a tipping point where it could easily pass through a critical transition into an alternate state (Dakos et al., 2015). However, applying these theoretical insights to the management of marine mammal populations is limited by a lack of critical ecological data in many species: without these data it is challenging to characterize baseline variability in populations and resources well enough to detect changes that might indicate a potential tipping point. There is also the important consideration that many population parameters for marine mammals are measured with such low precision that detecting any signal among the noise may be nearly impossible.
Levin and Möllmann (2015) argue that “accounting for marine regime shifts in management clearly requires integrative, cross-sectoral ecosystem-based management (EBM) approaches.” EBM is widely used for ocean management worldwide and is well suited for dealing with regime shifts, as it considers the multiple interacting drivers and ecosystem linkages that generate ecosystem shifts. They make a case for the use of Integrated Ecosystem Assessment (IEA) (Levin et al., 2009), an EBM framework used by a number of management agencies in the United States.3 IEAs are becoming more common, but they are still new enough in their development to allow the inclusion of regime shift concepts in an emerging EBM framework. IEAs could provide a transparent means of characterizing the status of ecosystem components, “prioritizing potential risks and evaluating alternative management strategies against a backdrop of actual environmental conditions.” To be useful, IEAs will need to identify ecosystem attributes and anthropogenic stressors; “develop and test indicators and reference levels that reflect key ecosystem attributes and the drivers; explore the susceptibility of an indicator to natural or human threats as well as the ability of the indicator to return to its previous state after being perturbed; evaluate the potential different management strategies to influence the status of key ecosystem components and the pressures that affect these ecosystem components”; and consider the precision with which the indicator can be measured, relative to the expected strength of the signal generated.
3 See http://www.noaa.gov/iea.