Improving the sustainability of Northwest Atlantic fisheries via predictive modelling

野生商业鱼类种群对加拿大人极其重要，生态管理单位依靠对各种生态系统属性的准确预测来支持管理决策。对这些属性的预测构成了短期战术管理咨询意见和长期管理规划战略建议的基础。然而，复杂的种群动态往往产生于各种时空过程，如物种间的相互作用、与自然环境的相互作用、高内在增长率、密度依赖动力学、人为干预和其他随机过程误差。这使得能够分析复杂系统的准确和自适应的预测模型对成功的管理至关重要。地质统计模型是利用生态时间序列的空间性质的线性预测模型的主要工具。渔业种群评估认识到这一重要性，并已转向空间种群评估。预测分布模型利用时空模型结合环境协变量和鱼类运动来做出更精细的预测。另一方面，已经发现非线性动力学在海洋渔业中很常见，并且由于观察到的高维模式和开发系统的非线性，自然系统和人为干预之间的相互作用可能限制预测的潜力。近年来，非线性时间序列预测方法的发展为提高生态时间序列的预测能力提供了一种数据科学的替代工具，如经验动态模型。因此，利用西北大西洋地区研究调查、商业捕捞和气候监测中存在的丰富生态时间序列来检验这两种现代方法的预测能力至关重要。具体而言，本文将以非线性预测模型和地统计时空分布模型为重点，探讨以下具体目标：1)西北大西洋主要商业鱼类的时间序列是否具有非线性性质，是否可以使用非线性预测方法进行预测；2)利用地统计模型，这些鱼类及其群落在短期和长期内是否表现出可预测的分布变化；3)是否可以通过纳入这些模型的预测来改进种群评估的各种空间过程；4)从预测模型中获得的知识如何有助于更好地管理鱼类种群，以平衡物种、生态系统和产业。

Wild commercial fish populations are extremely important to Canadians and ecological management units rely on accurate forecasting of various ecosystem attributes to support management decisions. The forecasts of those attributes form the basis for tactical management advice in the short term and strategic recommendations in long-term management planning. However, complex population dynamics often arise from various spatial and spatiotemporal processes such as interactions among species, interactions with the physical environment, high intrinsic growth rates, density-dependent dynamics, human intervention, and other stochastic process errors. This makes accurate and adaptive predictive models that are capable to analyze complex systems very critical for successful management. Geostatistical modelling is a major tool of linear predictive models that utilizes the spatial nature of ecological time series. Fisheries stock assessment recognizes such importance and has been moving towards spatial stock assessment. Predictive distribution modelling utilizes spatiotemporal models to incorporate environmental covariates and fish movement to make finer projections. On the other hand, it has been found nonlinear dynamics are common in marine fisheries, and the interaction between the natural system and human interventions can limit the potential for predictions given the observed patterns of high dimensionality and the nonlinearity in exploited systems. Recent developments in nonlinear time series forecasting method provide an alternative tool from data science to improve the predictive ability of ecological time series such as Empirical Dynamic models. Thus, it is crucial to test the predictive ability of the two modern methods using the rich ecological time-series exist from both research surveys, commercial catch, and climate monitoring in the Northwest Atlantic region. Specifically, I will focus on the nonlinear forecasting modelling and geostatistical spatiotemporal distribution modelling to investigate the following specific objectives: 1) are the time series of the major commercial fish in the Northwest Atlantic nonlinear in nature and can they be forecasted using nonlinear forecasting methods 2) do those fish species and their community display predictable distribution shifts in both the short term and long term using geostatistical models; 3) can we improve various spatial processes for stock assessment by incorporating predictions from those models; and 4) how the knowledge gained from predictive modelling contribute to better fish stock management to balance the species, the ecosystem, and the industry.