Collaborative Research: Integral Projection Models for Populations in Varying Environments: Construction and Analysis

所有的环境都是多变和不确定的：有些年份更热，有些年份更潮湿；捕食者大量消长。生物体如何对这种变异性进行缓冲并加以利用，是生态学面临的主要挑战。该项目将把新的统计和数学理论与半干旱植物群落的长期数据集结合起来，以解决在理解可变环境如何影响种群方面的两个挑战。第一个目标将是确定哪些环境变量对植物种群比率影响最大，如存活、生长和繁殖力。这项工作将借鉴对自然群落的长期研究，这些研究提供了数十年来对数千种植物个体的观察，以及以高空间和时间分辨率测量的许多环境变量。有这么多潜在的解释变量，标准的统计变量选择方法是不稳定和不可靠的。该项目将把机器学习技术中的数据挖掘方法与传统统计理论相结合，以确定重要的环境驱动因素，并建立更可靠的人口统计模型。将讨论的生态问题包括：人口比率是否主要取决于资源可获得性(例如土壤湿度)而不是温度等非资源变量；环境和竞争如何相互作用；以及过去的条件对当前业绩的影响有多大。经验数据集将被用来回答这些问题，并寻找多个社区的共性。第二个目标是了解环境变化影响的个体机制。例如，如果植物种群的增长最受降雨的影响，是因为许多植物立即死亡，还是因为长期影响，如预期寿命缩短和一生中体型变小？为了解决这些问题，离散状态随机过程理论中的统计方法将扩展到连续状态(例如，单个植物大小)和变化的环境，并应用于拟合的植物人口统计模型。该项目的一个预期的变革性成果将是人口建模和分析的新方法，其更广泛的意义将是这些工具在生态学、保护生物学和入侵物种管理中的应用。这项工作的统计产品将作为R代码模块免费分发，用户可以根据自己的研究系统进行调整。该项目还将支持统计学博士生和数量生态学博士后的研究、培训和指导。

All environments are variable and uncertain: some years are hotter, some wetter; predators wax and wane in abundance. How organisms buffer themselves against this variability, and exploit it, are major challenges for ecology. This project will combine new statistical and mathematical theory with long-term data sets on semi-arid plant communities to address two challenges in understanding how variable environments affect populations. The first goal will be to identify which environmental variables most strongly affect plant demographic rates such as survival, growth, and fecundity. This work will draw upon long-term studies on natural communities that provide observations on thousands of individual plants over multiple decades, along with numerous environmental variables that are measured at high spatial and temporal resolution. With so many potential explanatory variables, standard statistical methods for variable selection are unstable and unreliable. This project will combine data mining methods from machine learning techniques with traditional statistical theory to identify important environmental drivers, and build more reliable demographic models. Ecological questions that will be addressed include whether demographic rates respond primarily to resource availability (e.g., soil moisture) versus non-resource variables such as temperature; how environment and competition interact; and how strongly past conditions affect current performance. The empirical data sets will be used to answer these questions and look for generalities across multiple communities. A second goal is to understand the individual mechanisms that underlie effects of environmental variation. For example, if a plant population's growth is most affected by rainfall, is it because many plants die immediately, or because of long-term effects such as decreased life expectancy and smaller size throughout life? To address such questions, statistical methods from discrete-state random process theory will be extended to continuous states (e.g., individual plant size) and varying environments, and applied to the fitted plant demographic models. An expected transformative outcome from this project will be new methods for demographic modeling and analysis, and much of its broader significance will be the applications of those tools in ecology, conservation biology and invasive species management. Statistical products from this work will be disseminated freely as R code modules that users can adapt to their own study systems. The project will also support the research training and mentoring of a doctoral student in statistics, and a postdoctoral student in quantitative ecology.