CAREER: Next generation mixed membership models for highly multivariate biodiversity data

职业：高度多元生物多样性数据的下一代混合成员模型

• 批准号: 2040819
• 负责人: Denis Valle
• 金额: $ 68.17万
• 依托单位: University of Florida
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 2021
• 资助国家: 美国
• 起止时间: 2021-07-15 至 2026-06-30
• 项目状态: 未结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2040819&HistoricalAwards=false
• 关键词: CAREER; Next; generation; mixed; membership

Understanding and predicting how human activities impact biodiversity is challenging given the often large number of species in any given location. Importantly, despite the increasing amount of data generated by the remote-sensing revolution currently underway in Ecology (e.g., photos from satellites and camera traps), the use and integration of this information with field data on biodiversity is limited by the absence of modeling methods that can integrate multiple data streams and that can properly account for the characteristics of the data generated by these sensors. The long-term goal of this project is to advance cyberinfrastructure by creating broadly applicable methods for biodiversity datasets and by training the next generation of quantitative environmental scientists. This project will focus on substantially improving Mixed Membership (MM) models. These models were originally developed for text-mining purposes but have been widely used for biodiversity research in a wide range of ecosystems. Unfortunately, the current formulation of these models still has important limitations. This project will develop improved MM models that can account for the characteristics of the data generated by these sensors, can integrate multiple sources of data, and enable biodiversity predictions to be made. Ultimately, these improved MM models will be critical to enhance our ability to quantify and predict impacts on biodiversity. This project will also increase the awareness of the impact of climate change on biodiversity among high-school teachers and students. Evaluating and forecasting how species composition has been and will be altered by anthropogenic stressors is key to sustaining biodiversity and ecosystem functioning, but existing methods to quantify biodiversity change have important limitations. Biodiversity data are highly multivariate (e.g., an assemblage can contain hundreds of species in tropical forests) but many of the dimension-reduction methods typically used to interpret these data often generate results that are not easily interpretable (e.g., nonmetric multidimensional scaling axis scores), rely on unrealistic assumptions (e.g., hard clustering of sites), and are ill suited for wildlife studies because they do not account for imperfect detection. Critically, many of these methods do not allow for formal inference and/or predictions to be made and these methods do not leverage multiple data streams. To circumvent these limitations, this project will develop methods to generate new insights on the drivers of spatial and temporal variation of biodiversity. The overall objective of this project is to significantly improve MM models for biodiversity research. The specific objectives of this project consist of a) creating MM models that can generate reliable inference and predictions, integrate disparate data streams, and account for detection issues; and b) disseminate and train scientists on the developed models; and increase awareness of the impact of climate change on biodiversity among high-school students while addressing important science, math, and statistics standards. The results of this project will be stored in the stable URL https://denisvalle.weebly.com/mm-models.htmlThis 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.

考虑到在任何特定地点往往有大量物种，理解和预测人类活动如何影响生物多样性是具有挑战性的。重要的是，尽管目前在生态学领域正在进行的遥感革命产生了越来越多的数据（例如，来自卫星和相机陷阱的照片），但由于缺乏能够整合多个数据流并能适当解释这些传感器产生的数据特征的建模方法，这些信息与关于生物多样性的实地数据的使用和整合受到限制。该项目的长期目标是通过创建广泛适用的生物多样性数据集方法和培训下一代定量环境科学家来推进网络基础设施。该项目将着重于实质性地改进混合会员（MM）模型。这些模型最初是为文本挖掘目的而开发的，但已广泛用于各种生态系统的生物多样性研究。不幸的是，目前这些模型的表述仍然有重要的局限性。该项目将开发改进的MM模型，该模型可以解释这些传感器产生的数据的特征，可以整合多个数据来源，并使生物多样性预测成为可能。最终，这些改进的MM模型将对提高我们量化和预测生物多样性影响的能力至关重要。该项目还将提高高中教师和学生对气候变化对生物多样性影响的认识。评估和预测物种组成如何被人为压力源改变是维持生物多样性和生态系统功能的关键，但现有的量化生物多样性变化的方法存在重要的局限性。生物多样性数据是高度多元的（例如，热带森林中的一个组合可能包含数百种物种），但通常用于解释这些数据的许多降维方法通常产生的结果不容易解释（例如，非度量的多维尺度轴分数），依赖于不切实际的假设（例如，站点的硬聚类），并且不适合野生动物研究，因为它们没有考虑到不完善的检测。关键的是，这些方法中的许多不允许进行正式的推理和/或预测，并且这些方法不利用多个数据流。为了规避这些限制，本项目将开发方法，对生物多样性时空变化的驱动因素产生新的见解。本项目的总体目标是显著改进生物多样性研究的MM模型。该项目的具体目标包括a)创建MM模型，该模型可以生成可靠的推断和预测，集成不同的数据流，并考虑检测问题；b)传播和培训科学家开发的模型；提高高中学生对气候变化对生物多样性影响的认识，同时解决重要的科学、数学和统计标准问题。该项目的结果将存储在稳定的URL https://denisvalle.weebly.com/mm-models.htmlThis中，该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的智力价值和更广泛的影响审查标准进行评估，被认为值得支持。

项目成果

A new LDA formulation with covariates

一种新的带有协变量的 LDA 公式

• DOI: 10.1080/03610918.2023.2171059
• 发表时间: 2023
• 期刊: Communications in Statistics - Simulation and Computation
• 作者: Shimizu, Gilson Y.;Izbicki, Rafael;Valle, Denis
• 通讯作者: Valle, Denis

Identifying latent behavioural states in animal movement with M4, a nonparametric Bayesian method

• DOI: 10.1111/2041-210x.13745
• 发表时间: 2021-10-31
• 期刊: METHODS IN ECOLOGY AND EVOLUTION
• 影响因子: 6.6
• 作者: Cullen, Joshua A.;Poli, Caroline L.;Valle, Denis
• 通讯作者: Valle, Denis

The Latent Dirichlet Allocation model applied to airborne LiDAR data: A case study on mapping forest degradation associated with fragmentation and fire in the Amazon region

• DOI: 10.1111/2041-210x.13836
• 发表时间: 2022-03-16
• 期刊: METHODS IN ECOLOGY AND EVOLUTION
• 影响因子: 6.6
• 作者: Valle，Denis;Silva，Carlos Alberto;Brando，Paulo
• 通讯作者: Brando，Paulo

Automatic selection of the number of clusters using Bayesian clustering and sparsity‐inducing priors

使用贝叶斯聚类和稀疏性诱导先验自动选择聚类数量

• DOI: 10.1002/eap.2524
• 发表时间: 2022
• 期刊: Ecological Applications
• 影响因子: 5
• 作者: Valle, Denis;Jameel, Yusuf;Betancourt, Brenda;Azeria, Ermias T.;Attias, Nina;Cullen, Joshua
• 通讯作者: Cullen, Joshua