SEI+II:Ecological Discovery & Inference: Tools for Data-driven Exploration and Testing of Observational Data

研究生态系统的研究人员努力了解影响极其复杂系统的因素，在这些系统中，数十个(如果不是数百个)环境因素可以影响一个物种的分布和丰度。在整个地理范围内管理一个物种的挑战尤其巨大，因为从小型局部研究到整个大陆的管理可能是不可能的。此外，最初的管理决定可能需要在很短的时间期限内做出，并且在缺乏对物种的太多先验知识的情况下做出。通常，为鲜为人知的物种的及时管理决策提供信息的必要数据已经存在；然而，限制的是随时可以获得这些数据，特别是探索这些数据所需的分析工具。该项目将数据挖掘和机器学习工具的优势与统计方法结合起来，在数据挖掘和机器学习的新分析框架中创建一套强大的新预测和推理工具，从而能够从现有数据中提取更多相关信息，并将先前的信息纳入混合的分层/数据挖掘模型。其结果将是新的数据挖掘技术，允许对大量环境变量及其潜在相互作用进行统计推断。这将大大提高模拟鸟类种群对多种风险因素的景观反应的能力，并通过土地管理制定扭转种群数量下降的处方。该项目将向广大新的受众展示新的数据资源和计算分析、数据可视化和操作方面的进展：从生物学家、保护机构和土地利用规划者到学校教室和参与环境监测的数万公民，包括全国数百万观鸟的人。此外，该项目还培训新的研究人员，将强大的统计技术与机器学习和数据挖掘相结合

Researchers who study ecological systems strive to understand the factors that influence extremely complex systems, in which tens if not hundreds of environmental factors can affect the distribution and abundance of a species. Challenges in managing a species across its entire geographic range are especially great, as scaling up insights from small local studies to management over an entire continent may be impossible. Further, initial management decisions may need to be made under very short time deadlines, and in the absence of much prior knowledge about a species. Often, the necessary data to inform timely management decisions for little-known species already exist; however, what is limiting is ready access to the data and especially to the analytical tools needed to explore these data. This project joins the strengths of data mining and machine learning tools with statistical methods, to create a suite of powerful new predictive and inferential tools in a new analytical framework for data mining and machine learning that will permit extracting more relevant information from the available data and by incorporating prior information into a hybrid hierarchical/data mining model. The result will be new data-mining techniques that allow statistical inferences about large numbers of environmental variables and their potential interactions. This will greatly enhance the ability to model the landscape-level response of bird populations to multiple risk factors and to develop prescriptions for reversing population declines through land management. This project will expose new data resources and advances in computational analysis, data visualizations, and manipulations to vast new audiences: from biologists, conservation agencies, and land-use planners to school classrooms and tens of thousands of citizens who participate in environmental monitoring, including the millions of people across the country who watch birds. Additionally, the project trains new researchers in the union of powerful statistical techniques with machine learning and data mining