Methods for modeling population dynamics

对动物种群的现代管理和保护需要对种群动态参数进行可靠、准确的估计。对来自新标记设备的数据的分析(例如来自相机陷阱的照片、较小动物的轻型标记、GPS标记、遗传标记)以及多种数据来源(例如捕获-重新捕获、公民科学计数数据、EDNA)的结合给生态学家和管理者提出了统计方法方面的挑战。这种知识鸿沟的后果往往是基本的分析，无法利用数据中的所有信息，并可能提供误导性的结果。我的研究计划专注于开发新的统计方法和计算产品，以回答与种群动态有关的重要科学问题，用于保护和管理。具体的短期目标是：1)使用利用多种数据类型的方法对种群规模和生命率进行可靠的估计；2)开发新的方法，在捕获-重新捕获模型中纳入标签丢失，包括完全标签丢失；3)开发扩展批次标记实验的方法，以纳入协变量，如年龄和大小信息，以及标签丢失；以及4)探讨与捕获-重新捕获估计值相比，随时间和/或多个地点复制的计数数据模型的稳健性，这些数据具有野生动物和流行病学应用的协变量信息，包括确定参数冗余问题。这些目标将通过使用最先进的方法开发新的方法来实现，例如隐马尔可夫模型(用于批次标记和计数数据模型)、潜在多项模型(用于建模完全标签损失)和贝叶斯分层模型(用于综合总体模型)，同时通过模拟对这些方法和竞争方法进行严格评估。将努力开发评估拟合度的方法，并探索偏离模型假设的情况。这些方法将得到广泛应用，通常是通过在生态学(例如，塔库河上的鲑鱼、石灰石岛上的古马哈鱼)和流行病学(温哥华岛上的无家可归人口)方面的合作现实世界研究。我的长期研究目标是使这些新的统计方法可用来解决加拿大渔业科学家、野生动物管理者、生态学家和流行病学家目前面临的应用问题。

Modern management and conservation of animal populations requires robust, accurate estimates of population dynamics parameters. The analysis of data from new tagging devices (e.g. photographs from camera traps, light-weight tags for smaller animals, GPS tags, genetic tags) as well as the incorporation of multiple data sources (e.g. capture-recapture, citizen science count data, eDNA) has presented statistical methodology challenges for ecologists and managers alike. The consequence of this knowledge gap is often rudimentary analyses that do not harness all information in the data and may provide misleading results. My research program focusses on developing novel statistical methodology and computational products to answer important scientific questions related to population dynamics for conservation and management. Specific short-term goals are to: 1) develop robust estimates of population size and vital rates using approaches that take advantage of multiple data types; 2) develop new methods to incorporate tag loss in capture-recapture models including complete tag loss; 3) develop methods for extended batch marking experiments to incorporate covariates such as age and size information, and tag loss; and 4) explore the robustness of models for count data replicated through time and/or multiple sites with covariate information for wildlife and epidemiological applications in comparison to capture-recapture estimates including determination of parameter redundancy issues. These objectives will be achieved by developing new methods using state of the art methodology such as hidden Markov models (for batch marking and count data models), latent multinomial models (for modelling complete tag loss), and Bayesian hierarchical models (for integrated population models) while critically assessing these and competing approaches through simulation. Efforts will be made to develop methods to assess goodness-of-fit and explore deviations from model assumptions. These methods will be widely applicable, often through collaborative real-world studies in both ecology (for example salmon on the Taku river, Ancient Murrelets on Limestone Island) and epidemiology (homeless populations on Vancouver Island). My long-term research goal is to make these novel statistical methods available to tackle current applied problems faced by Canadian fisheries scientists, wildlife managers, ecologists, and epidemiologists.