DRAFT REVISED STELLER SEA LION RECOVERY PLAN Eastern and Western Distinct Population Segments (Eumetopias jubatus) National Oceanic and Atmospheric Administration National Marine Fisheries Services Office of Protected Resources

This is the report of a project to explore the use of a Bayesian population viability analysis, in a decision theory framework, to define recovery criteria that satisfy the requirements of the Endangered Species Act, for the western US distinct population segment of Steller sea lion. The project was undertaken by the author, as a consultant to the Steller Sea Lion Recovery Team that was in the process of writing a draft for a new Recovery Plan. A subgroup of the Recovery Team provided three crucial inputs as expert opinion. These were: (1) quantification of necessary policy elements that are not fully specified in existing agency guidelines, (2) specification of some uncertain factual elements that were needed for the modeling but could not be fully determined by statistical analysis of hard data, and (3) specification of the probability of the essential correctness of the core assumptions of the PVA model that was used versus alternative hypotheses that greatly discount the risk to the population. The standard adopted for downlisting from Endangered to Threatened was 99% probability of the population persisting for 100 years without declining below a quasi-extinction threshold. The quasi-extinction threshold adopted was 4743 individuals, corresponding to a genetically effective population size of 1000. The belief in the essential correctness of the model was quantified at 80% probability that the alternative hypotheses discounting the risk to the population are not correct. The subgroup drew up a table representing their opinion of the intensity, during respective past time intervals, of factors responsible for past threats to the population as used in the model. The central hard data used in the modeling were the 6 available population wide estimates of population size, that span time intervals averaging approximately 10 years in duration. The salient features of the available information, from the standpoint of assessing extinction risk, are the combination of large, but much reduced population size, continuing and volatile decline for many decades until just a few years ago, unexplained dynamics, failure to recover as expected, and a context of very large fisheries operations and large natural ecosystem variability. The basic PVA model captures this state of knowledge by assuming that the population is subject to random changes in its growth rate, at random intervals, where the distribution of exponential growth rates is normal, and the distribution of interval length is exponential with a mean duration of 10 years. The dynamics in the model are not density dependent, and a specific analysis was done to elucidate the circumstances under which a population might go through such a wide range of population sizes without displaying density dependence. The PVA model represents process variation through stochasticity of the changes in population growth rate and stochasticity of the time intervals between changes. The PVA model incorporates parameter uncertainty by representing the parameters of the distribution of population growth rates as a joint distribution of the uncertain mean and uncertain standard deviation. The joint distribution of these uncertain parameters was obtained by Bayesian inference from the past data, as adjusted by the subgroup’s expert opinion concerning correction for threat factors that are believed to have had different intensity in the past than they will in the future. The primary reason for belief in these differences is the changes in implementation of regulatory protection for sea lions, and changes in the operation of the fisheries. The basic model, applied to the data as adjusted by the subroup’s inputs, and using the subgroup’s policy specifications and appraisal of overall correctness, predicts almost 30% probability of quasi-exinction within 100 yrs from 2004, if the current level of protections is maintained. The population grew at roughly 2.8% per year in the interval 2000-2004. If that growth continues till 2024, the population size will then be 83,352 (roughly doubling the population size from 2004, in a little less than two generations). At that population size, and if the population growth were known to have stayed constant at 2.8% through two intervals, with all the other inputs the same, the assessment becomes almost 13% probability of quasi-exinction within 100 yrs from 2024, if the current level of protections is maintained. Absent knowledge of the rates of growth between 2004 and 2024, the attainment of a population size of at least 83,352 in 2024, with all other inputs the same, the assessment becomes about 19% probability of quasi-exinction within 100 yrs from 2024, if the current level of protections is maintained. All these scenarios fail to meet the risk standard for a downlisting criterion. Sensitivity analysis shows that the observed, but unaccounted for, steep rate of decline in the period 1985-1989, contributes a very large component of the calculated risk, both through its influence on the inferred mean rate of population growth and its influence on the inferred random variation in growth rate. If this interval were removed from the analysis, a recovery criterion of 83,352 as the population size in 2024, with all other inputs the same nearly meets the standard (achieving 1.43% probability of quasi-exinction within 100 yrs from 2024, if the current level of protections is maintained). The subgroup was not able to justify exclusion of the 1985-1989 decline rate from the analysis, though it is acknowledged that there may have been some peculiarities in the fishery unique to this interval. The large role of uncertainty in forcing the analysis toward very stringent recovery criteria (high population values), indicates that a new approach to extinction risk assessment, taking account of planned experimentation and a firm commitment to adjust management in response to future monitoring and results of experiments, may be the only way to obtain more readily attained recovery criteria while still satisfying the chosen standards. Reconciling this approach to present interpretations of the legal requirements of the Endangered Species Act may take some careful examination, particularly of the legal nature of commitments to future implementation of a plan with contingencies based on results of future monitoring and experimentation, and the required demonstration that management measures built into the plan really are adequate to contain the risk. The theory of how to technically quantify the total risk of such a plan is known.