Assessing the value of monitoring to biological inference and expected management performance for a European goose population
Johnson, Fred et al. (2022), Assessing the value of monitoring to biological inference and expected management performance for a European goose population, Dryad, Dataset, https://doi.org/10.5061/dryad.j3tx95xjg
1. Informed conservation and management of wildlife require sufficient monitoring to understand population dynamics and to direct conservation actions. Because resources available for monitoring are limited, conservation practitioners must strive to make monitoring as cost-effective as possible.
2. Our focus was on assessing the value of monitoring to the adaptive harvest management (AHM) program for pink-footed geese (Anser brachyrhynchus). We conducted a retrospective analysis to assess the costs and benefits of a capture-mark-resight (CMR) program, a productivity survey, and biannual population censuses. Using all available data, we fit an integrated population model (IPM) and assumed that inference derived from it represented the benchmark against which reduced monitoring was to be judged. We then fit IPMs to reduced sets of monitoring data and compared their estimates of demographic parameters and expected management performance against the benchmark IPM.
3. Costs and the precision and accuracy of key demographic parameters decreased with the elimination of monitoring data. Eliminating the CMR program, while maintaining other monitoring instruments, resulted in the greatest cost savings, usually with small effects on inferential reliability. Productivity surveys were also expensive and some reduction in survey effort may be warranted. The biannual censuses were inexpensive and generally increased inferential reliability.
4. The expected performance of AHM strategies was surprisingly robust to a loss of monitoring data. We attribute this result to explicit consideration of parametric uncertainty in harvest-strategy optimization and the fact that a broad range of population sizes is acceptable to stakeholders.
5. Synthesis and applications: Our study suggests that existing or potential monitoring instruments for wildlife populations should be scrutinized as to their cost-effectiveness for improving biological inference and management performance. Using Svalbard pink-footed geese as a case study, we show that the loss of some existing monitoring instruments may not be as adverse as commonly assumed if data are jointly analyzed in an integrated population model. Finally, regardless of the monitoring data available, we suggest that conservation strategies that explicitly account for uncertainty in demography are more likely to be successful than those that do not.
Population counts conducted in spring and autumn – Internationally coordinated population counts of pink-footed geese have been performed annually since 1990 in Denmark, Belgium, and the Netherlands in late October or early November (hereafter referred to as the November count) (Madsen et al., 1999). Over time, the population has expanded its distribution and the spatial coverage of the count has repeatedly been extended to capture new sites occupied by geese (Madsen, Christensen, Balsby, & Tombre, 2015). Since 2005, the population has also been counted in Norway, and since 2016 in Sweden. Because of increasing challenges in monitoring the autumn population, an additional count was introduced in May in 2010, which includes Norway, Denmark, Sweden and, since 2016, Finland. The known sites are covered by a network of trained observers who coordinate the coverage. The May census costs €5,327 per year for academic staff salary and travel costs of volunteers. The cost of the November census is similar, costing €3,350 per year.
Harvest estimates – Pink-footed geese are subject to an open hunting season in Norway, including Svalbard, and in Denmark. The species is protected in the Netherlands, Belgium, Sweden, and Finland. In both Norway and Denmark, reporting the harvest is mandatory and hunters report their harvests online. Harvest monitoring imposes no additional costs on the pink-footed goose management community.
Temporal distribution of harvest in Denmark – The November count occurs after the start of hunting seasons in Norway and Denmark and therefore it is important to be able to partition harvest into that occurring before and after the count. We assumed all the harvest in Norway occurs prior to the November count. In most years, the harvest in Norway occurs prior to the November count (Gundersen, Clausen, & Madsen, 2017; Jensen, Madsen, & Tombre, 2016). Moreover, the average number of pink-footed geese in Norway during the November count is very low, consisting of less than five percent of the total. The temporal distribution of the pink-footed goose harvest in Denmark is derived from wings submitted to the Danish wing survey. Danish hunters voluntarily submit wings from harvested individuals, providing information on date, location, and age of harvested geese. Monitoring the temporal distribution of the harvest in Denmark imposes no additional costs on the pink-footed goose management community.
Proportion of young in the autumn – Based on age-specific plumage characteristics, random counts of the number of young of the year and of older geese in flocks have been conducted by trained observers in the Netherlands and Denmark since 1980 (Madsen, 1982; Madsen et al., 1999), and in more recent years in Belgium, Norway, and Sweden as migratory behavior has changed. To minimize the effect of seasonal changes in age ratios, we used only counts occurring between October 12 and November 4, inclusive. Fall counts of young and adults are relatively expensive, and at the current level of effort cost €21,897 per year for academic staff salary, contract observers, and travel costs.
Capture-mark-recapture program – Estimates of annual survival rate (i.e., reflecting all sources of mortality) are available from a capture-mark-recapture (CMR) program (Madsen et al. 2002). During the period 1991 – 2020, 4,984 pink-footed geese were captured and fitted with neck collars and tarsus metal rings during spring staging in Denmark and Norway and, in four years (2007, 2008, 2012, and 2018), on the Svalbard breeding grounds. At marking, all birds had survived at least one fall migration and one hunting season. Re-sightings were made by a network of professional and volunteer observers outside the breeding grounds (September-May). Dead recoveries are reported by members of the public to the ringing centers involved (Denmark and Norway). They include both geese reported shot by hunters and those found dead by other members of the public. Using program MARK, we fit a number of Burnham’s joint recovery-recapture models (Burnham 1993) by including all possible combinations of fixed and fully time-dependent survival (S), resighting probability (p), recovery probability (r), and fidelity (F). The fidelity parameter expresses the proportion of geese surviving from year to year that are available for resighting and might, therefore, (among other things) reflect marker retention and emigration to poorly covered sites or other populations. Models were evaluated using AICc (Burnham & Anderson, 2003) and we chose for our use model S(t) p(t) r(t) F(.), indicating a fixed F-parameter and full temporal variability in survival, detection and recovery probability. Estimates of annual survival and their standard errors (inflated using a variance inflation factor) from this best-performing model were used to specify prior beta distributions for the IPM using the method of moments (Bolker, 2008). We omitted the last survival estimate in the time series due to identifiability issues.
It should be noted that using survival estimates to generate informative prior distributions is not necessarily equivalent to using the encounter data directly in the IPM joint likelihood. Although we investigated the possibility of including the encounter histories as data in the joint likelihood, we eventually abandoned this approach for several reasons. First and foremost, the CMR data are needed to estimate both annual survival and May population size. To do this simultaneously from the raw data within the IPM requires use of the Jolly-Seber model (Jolly, 1965; Seber, 1965). A major obstacle with use of this model, however, is a key assumption that the “capture” rate of marked and unmarked birds is the same. This assumption is certainly violated when the initial capture is a physical capture and subsequent “captures” are sightings from a distance. Failure of this assumption biases the estimates of population size (although not of survival) (Nichols, Hines, & Pollock, 1984). We also considered embedding the Cormack-Jolly-Seber model (Cormack, 1964; Jolly, 1965; Seber, 1965) to estimate survival from live encounters, but mortality and permanent migration are confounded and, thus, survival estimates can be biased low (especially if re-sightings occur primarily in only a portion of the animal’s range). Finally, we considered the use of Burnham’s joint model (Burnham, 1993) for live and dead encounters. There were practical problems with this model related to excessive computing time (especially when many data-reduction scenarios had to be run) and to the necessity of specifying a single set of constraints for the structure of the encounter data in the IPM. The latter contrasts with the independent analysis in MARK, in which a large number of constraints could be tested to provide the most parsimonious model. Using informative priors based on previous studies is a widely accepted approach in Bayesian analysis (Hobbs & Hooten, 2015; King, Morgan, Gimenez, & Brooks, 2009). McCaffery & Lukacs (2016) provide an example involving the use of informative priors for survival in birds. Nonetheless, the ideal approach would be to use the encounter histories as data in the IPM joint likelihood when possible and practical (Kéry & Schaub, 2012).
The CMR data are also used to derive an estimate of pink-footed goose population size (N) by dividing the number of marked geese in the population (M) by the ratio of the number of marked geese to the number of geese observed (R) (Sheaffer & Jarvis, 1995): N = M/R . This approach has been used in the monitoring program since 1991 (Ganter & Madsen, 2001) and resighting effort has increased significantly since 2011. As described by Clausen et al. (2019), the CMR data are used to derive estimates of pink-footed goose population size in spring, which are independent of the May counts. M is estimated from the number of marked birds seen alive in any given year corrected for annual variation in detection probability, and R is estimated as the average of the ratios from observations of marked individuals throughout the flyway. The time series of population estimates were used as input data for the IPM using a log-normal likelihood.
The CMR program is the most expensive of all monitoring instruments, costing €93,569 per year for materials, salary, and travel.
Spring temperature in Svalbard – Warm May temperatures on the breeding grounds, used as a proxy for snow melt in relation to the timing of egg-laying by geese, tend to improve reproductive success of pink-footed geese (G H Jensen, Madsen, Johnson, & Tamstorf, 2014). Therefore, daily average temperatures during May at two locations in Svalbard (Ny Ålesund and Svalbard Airport) are retrieved each year from the Norwegian Center for Climate Services (https://seklima.met.no/observations/). The number of days in which the average temperature is above freezing are tabulated and the number of “thaw days” are calculated by averaging values from the two stations. There is no additional cost incurred in acquiring these data.