Albatross Modeling Workshop
November 7-9, 2007
Imin Conference Center
University of Hawaii at Manoa
1. Preliminaries
This workshop on modeling of Black-footed Albatross (BFA) populations is a follow-up to the PFRP-sponsored workshop in December 2001 on Protected Species Modeling. The 2002 workshop concluded that development of integrated statistical models of Hawaiian albatross population is feasible. Development of such models was included as a research priority in the 2002 PFRP request for proposals, and the PFRP subsequently funded two projects on modeling black-footed albatross. The general purposes of this workshop were to: (1) review results of the PFRP-funded projects, (2) compare these results to other population assessments, (3) identify problems in model development and application, and (4) suggest future directions for model development.
Kathy Cousins, Rebecca Lewison and Simon Hoyle kindly volunteered their services a rapporteurs. Simon Hoyle also volunteered to present an
outline of the workshop results to the PFRP PI Workshop on November 13, 2007.
2. Review of models and model results from PFRP funded projects:
Dan Goodman - Best Science and Best Data
Dr. Goodman's presentation stressed the importance of observation and emphasized the distinction between observed data and estimates used as data. As an example of the latter Dr. Goodman cited the discrepancy between the number of leaves on a strawberry plant recorded in medieval illustrations and the number of leaves on strawberry plants observable in the garden. Meta-analysis which often consists of analyzing results reported in the literature often does not allow for the possibility that estimates reported in the scientific literature may often not be closely connected to observation. Real observations have associated errors which are often lost in meta-analysis. Integrated models, which combine data from different sources with different associated errors, attempt to propagate these errors to correctly represent the uncertainty in the model conclusions through the use of joint likelihood functions.
Daubert Standard
In 2000, the Supreme Court approved amendments to the Federal Rules of Evidence relating to opinion evidence and expert testimony to conform to the "Daubert trilogy." In addition to amending Rules 701 and 703, Rule 702 now includes the additional provisions which state that a witness may only testify if, "1) the testimony is based upon sufficient facts or data 2) the testimony is the product of reliable principles and methods, and 3) the witness has applied the principles and methods reliably to the facts of the case."
http://en.wikipedia.org/wiki/Daubert_standard
Authors' summary not available; the following abstract is from the published paper.
Quantifying the impact of longline fisheries on adult survival in the black-footed albatross.
SOPHIE VÉRAN, OLIVIER GIMENEZ, ELIZABETH FLINT,WILLIAM L. KENDALL, PAUL F. DOHERTY JR and JEAN-DOMINIQUE LEBRETON
Summary
1. Industrial longline fishing has been suspected to impact upon black-footed albatross populations
Phoebastria nigripes by increasing mortality, but no precise estimates of bycatch mortality are available to ascertain this statement. We present a general framework for quantifying the relationship between albatross population and longline fishing in absence of reliable estimates of bycatch rate.
2.We analysed capture–recapture data of a population of black-footed albatross to obtain estimates of survival probability for this population using several alternative models to adequately take into account heterogeneity in the recapture process. Instead of trying to estimate the number of birds killed by using various extrapolations and unchecked assumptions, we investigate the potential relationship between annual adult survival and several measures of fishing effort. Although we considered a large number of covariates, we used principal component analysis to generate a few uncorrelated synthetic variables from the set and thus we maintained both power and robustness.
3. The average survival for 1997–2002 was 92%, a low value compared to estimates available for other albatross species. We found that one of the synthetic variables used to summarize industrial longline fishing significantly explained more than 40% of the variation in adult survival over 11 years, suggesting an impact by longline fishing on albatross’ survival.
4. Our analysis provides some evidence of non-linear variation in survival with fishing effort. This could indicate that below a certain level of fishing effort, deaths due to incidental catch can be partially or totally compensated for by a decrease in natural mortality. Another possible explanation is the existence of a strong interspecific competition for accessing the baits, reducing the risk of being accidentally hooked.
5. Synthesis and applications. The suspicion of a significant impact of longline fishing on the black-footed albatross population was supported by the combination of a low estimate of adult survival for the study period, and a significant relationship between adult survival and a synthetic measure of fishing effort. This study highlights the sensitivity of the black-footed albatross to commercial longline fishing, and should exhort fishery management authorities to find adequate seabirds avoidance methods and to
encourage their employment.
Journal of Applied Ecology (2007) 44: 942–952
The negative relationship between adult survival in the swordfish fishery depends on a linear regression between survival and a proxy measure of the effect of the fishery for the period 1997-2002. Only 6 estimates of annual survival were available for use as dependent variables in the regression because no other data were available at the time of the analysis. The low number of degrees of freedom constrained the complexity of the model. The working hypothesis of the analysis is that albatross bycatch is associated with swordfish catch. Unfortunately there are no available estimates of albatross bycatch, nor are there any available data on fishing method. Therefore, a single surrogate variable, based on principle component analysis of fishery data, was used as the independent variable in the analysis.
This analysis was severely constrained by the lack of available data. Estimation of adult survival depends completely on enumeration of re-sightings of banded birds on the nesting colony. Such data are not accommodated in the United States Geological Survey National Bird Banding Laboratory (BBL) and Dr. Veran was obliged to invest a substantial amount of time and effort to compile the necessary data. Data on sea bird bycatch are similarly difficult to obtain. Data collected by fisheries observers United States longline vessels are comprehensive, but considered confidential and access is difficult to obtain. Data on non-US vessels is either non-existent or unobtainable. Similar constraints are imposed on data pertaining to fishing operations, such as depth of fishing or hooks between floats, making it nearly impossible to estimate a measure of swordfish-directed fishing effort.
Dan Goodman - Bayesian Re-examination of the Tern Island Mark Recapture Study and the Correlation with Fishing Activity
Author's summary not available
Dr. Goodman offered a two-part presentation on the results of an exploratory inspection of the data at hand followed by preliminary Bayesian inferences. A number of inconsistencies in the data were noted. The number of breeding pairs on Tern Island is increasing exponentially at a rate that will lead to a doubling of the number in 5-6 years. In contrast, the population at French Frigate Shoals as a whole appears to have been stable for 25 years. Exploration of the resighting data suggest that resighting efforts were inconsistent until around 1995. Using only the most recent data, the survival of birds between ages 6 and 11 varies with both age and year from a low of approximately 75% up to 100%. The mode of the posterior distribution of survival over all age classes varies with year from 88% to 94%. The survival rate of these age classes appears to be negatively correlated with total North Pacific longline catches. The conclusions about adult survival and the effects of fishing on survival were similar to the results of Veran and Lebreton.
In order to cope with the inconsistencies in the data prior to 1995, Dr. Goodman was obliged to omit most of the early observations in the time series. Variability in resighting data suggest that there may be confounding between detectability on the colony and survivorship. Resighting effort appears to have been variable and efforts to standardize resighting effort should be continued. The lack of reliable fisheries data was again raised as an issue especially with respect to spatial location of both fishing effort and BFA interactions. Adding new data collected since 2005 would be very helpful in increasing the confidence in the conclusions.
Authors' summary not available
A state-space Kalman filter model was used to model the population trajectory using a classical stage-based population projection matrix in the transition equation. Some of the terms in the population projection matrix were estimated using the Kalman filter. The same PC2 covariate used by Lebreton and Veran was again used as a proxy for fishery effects. Survival of adults was inversely related to the fishing proxy, but survival of pre-breeders was not significantly related to the fishing proxy.
Discussion again focused on the paucity of reliable data. In the case of the state-space model, reliable data on juveniles would be particularly informative. Spatial issues were a concern in the discussion. BFA individuals are know to forage of the west coast of North America where there are few data on on interactions and lest stringent seabird bycatch mitigation measures.
Combined Abstract
Simon Hoyle began by summarizing progress on black-footed albatross modelling under the PFRP funded project ‘A General Bayesian Integrated Population Dynamics Model for Protected Species’. Two models are being developed using different approaches, by Hoyle & Maunder, and by Maunder, Alvarez-Flores, & Hoyle. With the modelling still proceeding, input from meeting participants was sought on the data used, and on model assumptions, structure, and implementation. Simon’s presentation summarized the data used in both models and the structure and preliminary results from the first model, while the following presentation by Mark Maunder summarized the second model and its preliminary results.
The Hoyle and Maunder model is intended to be general enough to model the important aspects of population dynamics for a range of albatross species. Its structure includes sub-populations on island groups, with bycatch being taken from a pooled population. The statistical approach is to use integrated analysis, fitting to multiple data types within the one model. These include nest counts, fledgling counts, estimates of breeding success, survival and state transition rates from mark-recapture data, and fishery bycatch. Fishing effort and bird distribution data are also included in the model, and assumed to be known without error.
The Maunder, Alvarez-Flores, Hoyle model is based on including multiple covariates to explain survival and other population processes. Fishing effort is a special case covariate. Other covariates can be used to include human induced (e.g. culling of birds) and natural changes in survival. The total amount of kills attributed to a covariate can be fit to estimates of the kills. Impact plots are then created to illustrate the impact of each covariate on the population.
Long term nest and fledgling count data show some interesting trends and dynamics, which suggest hypotheses and sensitivity analyses. For example, population trends appear to differ among islands, with long-term stability on Laysan island, but substantial increases on most other island groups. This suggests that effects on or close to islands must be considered. Also, a decline in nests counted at French Frigate Shoals from the mid-1980’s to the mid-1990’s was contemporary with the disappearance of Whale-Skate island (which held up to 40% of FFS breeding pairs), so to some extent may have represented skipping breeding until nest sites were re-established on other islands.
Preliminary model results were presented, including diagnostic fits to the data, and estimates of total bycatch by fishery. These results are expected to change substantially as model structure and input data are improved.
3. Review of results from other models:
Abstract
We evaluated the status and trends in Laysan and Black-footed Albatross populations using linear regression, population viability analysis (PVA), and age-structured population models. All analyses were based on nest count data gathered by the U.S. Fish and Wildlife Service at French Frigate Shoals (FFS), Laysan Island, and Midway Atoll, three colony sites containing the majority of the World population of Laysan and Black-footed albatross. Regression analysis was used to evaluate linear trends in populations over the time intervals of 1922-2005, 1957-2005, and 1992-2005. Population viability analyses were conducted by fitting stochastic, discrete time, logistic and exponential models to observed data, and then making population projections 60 years into the future. Age-structured population models, based on a matrix modeling framework, were used to evaluate the effect of fishery bycatch on populations, and to determine the potential number of birds that could be removed without causing a population decline (Potential Biological Removal, or PBR). Estimates of albatross bycatch were obtained either by projecting measured rates for U.S. fisheries to international fleets, or by using published figures.
The Laysan Albatross population increased from 18,000 pairs in 1922, to 590,000 pairs in 2005, and currently does not appear to be at risk of decline due to fishery bycatch. The large increase in population size during the past 83 years, is likely a response to cessation of persecution by feather hunters and the military, and an increase in nesting area at some colonies. Analysis of linear trends in the population showed a positive change over the 1922-2005 and 1957-2005 time periods, and a stable size during 1992-2005. PVA results for FFS indicate that the population is currently stable and has a 50% probability of increasing 8.36 times in the next 60 years, but it also has a 28% probability of decreasing 0.24 times. For Laysan Island, the PVA predicts that the median population size will increase by 15.7% in the next 60 years, but there is a 45% chance that the population will decline by > 50%. No PVA analysis was conducted for Midway due to the paucity of data. Matrix modeling results indicate that the Laysan Albatross population, summed across all three colonies, is increasing at 6.7% a year, and that the estimated bycatch of 2,500 birds/year is below the PBR.
The Black-footed Albatross population increased from 18,000 pairs in 1923, to 61,000 pairs in 2005, and currently may be at risk of decline due to fishery bycatch. As with Laysan Albatross, the increase in population size over the past 83 years, is probably in response to the end of persecution at the nesting colonies. Analysis of linear trends in the population showed a positive change over the 1922-2005 time period, no change during 1957-2005, and a slight decrease during 1992-2005. PVA results for FFS indicate that the population is increasing at 1% a year and has a 50% probability of increasing by 74% in the next 60 years, but it also has a 35% chance of decreasing by > 47%. For Laysan Island, the population is stable, and the PVA predicts the median population growth rate will increase by 54.7%, but this should be taken with caution since there is a 35% chance that the population would decrease by >50%. For Midway Atoll, the population is increasing by 1.5% annually, and the PVA predicts a 50% chance of a 1.36 times increase in the population size, and a 22.5% chance of the population decreasing by any amount. Matrix modeling results indicate that the Black-footed Albatross population, summed across all three colonies, is stable, or slightly increasing, with a population growth rate of 0.3% a year. The current estimate of bycatch is 5,228 birds/year, but if this value is doubled, as a safeguard for underestimating bycatch, it approaches the PBR of 11,980 birds/year.
Bill Kendall, Paul Doherty, Sarah Converse, Maura Naughton, and Beth Flint
Demographic estimates and banding database issues from Midway and Tern Island
Authors' summary not available
Maura Naughton, Marc Romano, Elizabeth Flint, William Kendall, and Sarah Converse.
Current Demographic Monitoring of Laysan and Black-footed albatrosses at the Hawaii Breeding Colonies
Authors' summary not available
4. Problems encountered implementing these models
A. Data issues
Problems with data was a recurring theme throughout the workshop discussions. Problems were cited for counts of nesting pairs, band resighting, fishery bycatch statistics, and descriptive information about fishing methods and location. These problems appear to rooted in legal, institutional and technical constraints.
(1) Legal constraints on access
The United States National Marine Fisheries Services imposes sever constraints on access to self-reported fisheries data from fishers' log books and to data recorded by on-board fisheries observers. Procedures for non-NMFS scientists to gain access to these data are cumbersome, time consuming and inhibit creative analysis of the data. The National Research Council found "Confidentiality of fisheries data is restrictive to the point of hindering both research and management. ... These policies neglect the rights of the public to have greater information about the use of public-trust resources. The privilege to exploit marine fish resources should carry some obligation on the part of fishermen, balancing reasonable protection of proprietary information against the large need of managers to be well informed and able to manage the fishery." (p160, "Improving the Collection, management, and use of marine fisheries data", National Academy Press, 2000.)
Similar restrictions prevent access to data collected by non-US fishing fleets. Foreign governments are reluctant to divulge details of fishing activities by vessels flying their flags. Most foreign fleets do not carry observers, and thus the observer coverage on non-US fleets is generally less than one tenth of the coverage on US fleets. Thus there are no available data on sea bird bycatch by non-US pelagic longline fleets operating in the north Pacific.
(2) Budgetary and institutional constraints
Albatross banding data are nominally housed at the USGS Patuxent Wildlife Research Center Bird Banding Laboratory (BBL). However albatross are a long-lived species, and albatross banding data typically consist of multiple resightings of the same bird (which may carry multiple bands). In contrast, the BBL was set up to accommodate a single resighting of a banded bird, such as a band returned by a hunter from a dead duck.
Fish and Wildlife Service scientists often collect large amounts of data in field notebooks and in conjunction with specific projects. The existence of these data is often unknown outside of FWS offices and is not generally recorded in a machine readable format.
Individual researchers created
ad hoc data bases from the banding data in order to develop population models. Paul Doherty transfered resighting data through 2001 from the original data sheets into an "Access" database. Sophie Veran and Jean-Dominique Lebreton used the Doherty database, extracted the releases and recaptures that could be used in her analyses, and stored these in the form of capture histories. Veran also visited FWS offices to gather data from them, which were added to the capture histories. The work creates a second
ad hoc data base which was also used by Goodman and by Maunder and Hoyle.
(3) Progress on data issues during the workshop.
Counts of adults and fledglings: Sievert agreed to provide on-island counts to Maunder and Hoyle for comparison with currently used counts. Discrepancies will be clarified by Flint and Naughton.
Survival rate estimates: Maunder and Hoyle obtained mark-recapture survival rate estimates for pre-breeders from Veran's thesis and have requested the full covariance matrices for both mark-recapture models.
Distribution: The importance of getting better estimates of BFA distribution was clarified and strategies for obtaining distribution estimates from colleagues in animal telemetry were explored.
Breeding success: Flint and Naughton clarified these data and their potential biases. Arrangements were made to obtain more detailed data.
B. Covariates
Most presenters emphasized high catches of BFA in longline fisheries as a constraint on the growth of BFA populations. The only reliable data on sea-bird bycatch in the north Pacific come from observers on the US longline fleet, which accounts for less than 5% of the total number of pelagic longline hooks deployed in the north Pacific. Since bycatch data are not available from other fleets, total BFA bycatch cannot be included in population models.
The US observer data show that the shallow-set method of fishing used to target swordfish also can have a high bycatch of sea-birds. Therefor attempts were made to find a proxy for sea-bird bycatch by manipulating the available data on longline catch. The "PC2" covariate utilized by Veran and Lebreton appears to represent some aggregate measure of swordfish fishing. Access to detailed information on fishing depth and location is severely restricted making it impossible to contrive a more useful proxy for BFA mortality using fishing data.
Workshop participants noted that BFA nesting grounds in the NWHI sit astride the Subtropical Frontal Zone, an important and very dynamic biogeographic boundary. Changes in the position of the Frontal Zone and the biological productivity Transition Zone where BFA forage are known to influence the reproductive success of BFA. Readily available climatic indices such as the
Pacifc Decadal Oscillation Index and the
Southern Oscillation Index should have been included as covariates in population models.
5. "Listing" considerations
The fish and wildlife service has been petitioned to list the BFA as threatened or endangered. Karen Rosa, who will be in responsible for compiling and reviewing information and preparing the FWS review, presented some background on the listing process. One of the functions of the process is to identify potential threats and their likely effects on BFA populations. The process will include a thorough review of all available information including published literature, "gray" literature, and expert comment. Some of the questions that will be considered in the decision to whether to list list the BFA as threatened or endangered are:
1. What are the population trends? It the population increasing or decreasing? It the population is stable or not increasing, is it on the verge of collapse?
2. What is the trend in recruitment success?
3. What likely changes in habitat will cause a decrease in the population?
This presentation provoked a lengthy discussion of whether the field data collection programs provide the information needed for rigorous population dynamics analyses to inform the listing decision.
6. Future directions:
Mark Maunder and Simon Hoyle -
Planning the next generation general population assessment model
Summary
Circumstances are arising that make this the time to start planning the next generation general population assessment model. Advances in computer power (e.g. computer performance, parallel processing, and distributed computing) and computational methods (e.g. automatic differentiation, random effects, and MCMC) have enabled the use of complex population dynamics models that can be fit to multiple types of data. Consequently, there has been a convergence among fields towards fitting to data and integrated population dynamics modeling. Meanwhile, pressure is mounting from management to assess an immense number of populations, but there are too few qualified analysts. However, the current general models are reaching there limitations and new ones are needed.
A general model provides a common language that allows communication among the different groups involved. This common language facilitates discussion, promotes development, and reduces duplication. It also greatly facilitates the review process and increases the number of reviews that can be conducted. Adoption of a general model by the scientific community will, hopefully, lead to comprehensive analysis and testing that can be used to generate guidelines for best practices.
There are many existing general models used for assessing populations. These include the integrated models used in fisheries stock assessment (e.g. Coleraine, MULTIFAN-CL, Stock Synthesis 2, CASAL, Gadget, Xsurvivers, ADAPT), multi-species/Ecosystem models (e.g. Ecopath/Ecosim), mark recapture based models (e.g. MARK, M-SURGE), and population viability analyses (ALEX, RAMAS, VORTEX, GAPPS, INMAT). All of these models have limitations. For example, ecosystem models often have simple population structure, mark-recapture models are limited in the type of data they use, and PVAs don't fit to data at all. The next generation general population assessment model should learn from these previous models, encompassing and extending their functionality.
Important features that need to be consider for the next generation general population assessment model include: the ability to integrated multiple data types, use of priors, inclusion of process error, spatial structure, sub-population structure, covariates, multi-species, meta analysis, inclusion of genetics, estimation of uncertainty, model selection and averaging, simulation of data for model testing and MSE, ability to include user defined functions, and the ability to run each component of the model separately. In addition, some features required for protected species that should be considered include: alternative stock-recruitment curves; density dependence in other processes, mate pairing, widowing, skipping breeding, stage structure, random variation in population processes due to small population sizes, mark-recapture data, occupancy data, minimum counts, habitat data, and individual characteristics. Issues with developing a general model that need to be considered include: the tradeoff between generality and computational efficiency, users using the model incorrectly, how to weigh the different data sets, and missing data in covariates.
Creating the next generation population assessment model is an ambitious project and substantial planning is needed to do it properly. The model should be open source and free to facilitate its uptake and development. This will create a community for development, testing, training, and assistance. Expertise is scattered among countries, organizations, and disciplines. Therefore, the project will require collaboration among statisticians and mathematicians to provide efficient algorithms, computer scientists to provide efficient code, statisticians to provide the appropriate statistical framework (e.g. likelihood functions), and ecologists and biologists to outline the population dynamic processes. Funding will be needed to make the project a success. Many institutions allow their staff to work on this type of project and cover their salaries. However, others require outside funding and funding will be needed to convene meetings and working groups. Many sources of funding have indicated a keen interest in developing the next generation population assessment model. Things look promising. Now is the time to start planning so that it is done properly.
7. General Findings and Recommendations
Workshop participants agreed that the following issues should be treated with urgency:
a. Data availability
Finding: Data availability and legal constraints impede research and model development.
Recommendation: Legal constraints need to be resolved within agencies and between governments. A metadata inventory could be hosted on the North Pacific Albatross Working Group (
NPAWG) website.
b. Data interpretation and analysis
Finding: Spatial and temporal overlap between fishing effort and bird distribution is critical to estimate bycatch and understand fishery impacts.
Recommendation: Develop uniform means to characterize longline fishing depth by identifying usable proxies. Attempt to access bird tracking results from TOPP and other sources.
c. Data storage and dissemination
Finding: The existing institutions do not provide a stable means to store and update albatross band resighting data.
Recommendation: Upgrade and maintain the USGS/BBL database so that it serves the needs of resource managers. Consolidate existing
ad hoc research data collections and make them available to all researchers.
Finding: Change in population size is a critical piece of information on which the status determination rests.
Recommendation: Collect more information on “walkers”, changes in age of first reproduction, and changes in survival by age to improve estimates of population size.
e. Data collection on-colony
Finding: It may be possible to collect additional data without compromising current protocols.
Recommendations: Clarifying which parameter(s) of interest are needed to monitor population change would facilitate improvements in data collection protocols. Additional statistical consultation on optimizing data collection would help to optimize banding efforts among islands and age classes. Initiate a PIT tag pilot study. Collect and archive feathers/egg shells. Collect and analyze mating pair data on known-age plots.
8. Convener's Conclusions
The PFRP Program Manager bears sole responsibility for the following conclusions and recommendations. Some workshop participants may hold other opinions.
The PFRP committed a substantial amount of time and money to BFA modeling projects with very little to show for it.
To a large extent, this lack of progress is attributable to the unsatisfactory state of the available data - a problem that has persisted for more than 10 years. Given the possibility that BFA may be soon be listed as threatened or endangered and the costs to the public that will be incurred as a result, the deplorable state of the available data needs to be urgently addressed. It is not simply a matter of finding additional funding; it is a question of adjusting priorities and the responsible use of existing funds. Upgrading the BBL capabilities to accommodate albatross band resighting should be treated as urgent. The existing and undocumented data sitting in FWS files needs to be brought into the light and descriptive metadata should be posted made available to the research community (perhaps, as was discussed in the workshop, by posting on the NPAWG website). These criticisms of the BFA data are also applicable to many other protected species and better planning of data collection, storage and dissemination is needed in general.
Several workshop participants agreed to attempt to update the data used for the analyses presented at the workshop. Whether the agencies involved will actually product the data is unclear.
It is widely acknowledged that albatross breeding success is mediated by environmental factors, yet the only covariates included in the population models pertained to fishing activities. BFA conservation plans that do only consider selected threats will not meet expectations and detract from the credibility of the responsible management agencies.
![[toggle]](javascript:icntoggle('mod-mnu_application_menul1','module.png');){kind=small}
