Collaborative Research: IIBR Informatics: Data integration to improve population distribution estimation with animal tracking data

确定环境因素如何影响特定物种在何处发生，对于保护和维持生物多样性非常重要。具体来说，这些知识可以用来描绘物种的生态位，为测量变化提供基准，并帮助确定保护区域的优先顺序。鉴于这种重要性，生态学家已经开发了许多统计工具来确定环境因素和物种发生模式之间的联系。大多数这些工具可以分为两类，基于它们是使用传统的调查数据，还是使用动物跟踪数据。在这两种情况下，可用数据的数量和质量往往是有限的。该项目旨在将这两种方法统一在一个方法下，可以同时使用两种类型的数据。这很重要，因为它可以帮助克服每个数据源中的限制，而且这些不同的数据类型具有互补的优势，因此组合起来提供更多信息。项目工作将集中在濒危物种上，包括美洲虎和低地貘，这两种数据类型都是可用的，以展示这些技术如何为保护工作提供信息。通过结合多种数据源的优势，这些新方法将能够更好地确定这些脆弱物种的优先栖息地和区域。高级项目人员将参加AniMove.org动物运动分析课程，教授学生将这些方法应用于保护问题，并将在北卡罗来纳自然科学博物馆（NCMNS）主持一个数据整合研讨会。利用NCMNS每年接待的100万访客，这个项目？美国的推广工作将侧重于制作和展示沉浸式视频，将整个科学过程栩栩如生，从研究设计和实地工作，到分析和预测，再到知情的保护决策。识别环境驱动因素与物种发生模式之间联系的工具在生态学中经常使用，物种分布模型（SDMs）和资源选择函数（RSFs）是特别突出的例子。虽然这些方法密切相关，但sdm往往用于大规模的调查数据，而RSFs通常用于局部种群，并应用于动物跟踪数据。单个跟踪数据集之间普遍存在的自相关和频繁的相互关联违背了标准分布模型的关键独立性假设。为了统一这些方法，将开发一种新的加权对数似然函数来解释跟踪数据集内部和之间的非独立性，以及不同的采样计划和研究持续时间。在齐次泊松点过程框架中，该加权对数似然将与仅存在和不存在的调查数据相结合，用于分布建模。这种方法有两个主要优点。首先，它将允许积累跟踪数据的库存，以有效地为广泛的分布分析提供信息，从本地规模的rsf到地理范围规模的sdm。其次，它将抵消调查数据中经常出现的明显的空间偏差，因为被追踪的动物经常去调查者不去的地方。与传统的分布模型相比，这种新方法将从当地人口无缝扩展到地理范围，增加总体样本量，并利用不同数据类型的对比特性来减少空间偏差，更准确地估计不确定性。为了促进这种方法的广泛使用，将开发一个免费的软件工具，即分布数据集成模块（DDIM），以构建必要的多源数据集，并用相关的环境协变量对这些数据进行注释。项目结果将在http://biology.umd.edu/movement.html.This上公布。该奖项反映了NSF的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Identifying how environmental factors affect where particular species occur is important for the preservation and maintenance of biodiversity. Specifically, this knowledge can be used to delineate species' ecological niches, provide benchmarks for measuring change, and help prioritize areas for conservation. Given this importance, ecologists have developed many statistical tools for identifying linkages between environmental factors and species occurrence patterns. Most of these tools can be sorted into two categories, based on whether they use traditional survey data, or animal tracking data. In either category, the amount and quality of available data is frequently limiting. This project aims to unify these two approaches under a single methodology that can simultaneously use both types of data. This is important because it can help overcome limitations in each data source, and because these different data types have complementary strengths, and are thus more informative in combination. Project work will focus on at-risk species including jaguars and lowland tapirs, where both data types are available, to demonstrate how these techniques can inform conservation efforts. By combining the strengths of multiple data sources, these new methods will be able to better resolve priority habitats and areas for these vulnerable species. Senior project personnel will participate in the AniMove.org animal movement analysis courses to teach students to apply these methods to conservation problems and will also host a data-integration workshop at the North Carolina Museum of Natural Science (NCMNS). Leveraging the 1 million yearly visitors that NCMNS receives, this project?s outreach efforts will focus on creating and displaying immersive videos that bring to life the entire scientific process, ranging from study design and field work, through analysis and forecasting, and on to informed conservation decision making.Tools that identify linkages between environmental drivers and species' occurrence patterns are routinely used in ecology, with species distribution models (SDMs) and resource selection functions (RSFs) being especially prominent examples. Though these approaches are closely related, SDMs tend to be employed on large scales with survey data, while RSFs are typically used for local populations and applied to animal tracking data. The ubiquitous auto-correlation within, and frequent cross-correlation among, individual tracking datasets violates the key independence assumption of standard distribution models. To unify these approaches, a novel weighted log-likelihood function will be developed to account for non-independence both within and among tracking datasets, as well as for differing sampling schedules and study duration. This weighted log -likelihood will be integrated with both presence-only and presence-absence survey data in the very general in homogeneous Poisson point process framework for distribution modeling. This approach has two primary advantages. First, it would allow accumulating stockpiles of tracking data to validly inform a broad range of distribution analyses, from RSFs at the local scale, to SDMs at the geographic range scale. Second, it will counteract the often -pronounced spatial biases in survey data by leveraging the fact that tracked animals frequently go where surveyors don? t. Compared to conventional distribution models, this novel methodology will scale seamlessly from local populations to geographic ranges, increase overall sample size, and exploit the contrasting properties of the different data types to reduce spatial bias and more accurately estimate uncertainty. To facilitate broad use of this methodology, a freely available software tool, the Distribution Data Integration Module (DDIM), will be developed to both construct the necessary multi-source datasets, and annotate these data with relevant environmental covariates. Project results will be available at http://biology.umd.edu/movement.html.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.