Complexity of Spatial and Categorical Scale in Landcover Characterization: A Statistical and Computational Framework

土地覆盖特征是陆地生态系统和气候建模、监测和预测的基础。这些特征主要是根据从卫星遥感测量中提取的信息得出的。 然而，全球和区域的土地覆盖大多由不同空间尺度的不同土地覆盖类型的异质混合物组成，而遥感土地覆盖特征描述的标准方法是在单一尺度上操作的，并使用一套固定的类别。 这项研究将开发一个综合的统计和计算框架，将产生土地覆盖特征的遥感数据，允许同时在空间分辨率和分类尺度的泛化。 这一土地覆被特征描述过程的结果将不再是一张土地覆被图，因为地图意味着单一的空间比例，而是一个土地覆被数据库，可以用传统方式和目前无法使用的方式查询。从本质上讲，用户可以自由选择最适合其需要的一系列空间和分类尺度。 制定这一框架的关键要素包括对地理和遥感、统计和计算机科学领域的新贡献。在基础上是一个新的多尺度统计模型，称为mixlets，从它将产生一个新的范式表示和可视化的土地覆盖分类。 反过来，这些进展将与空间数据库表示和空间查询系统的新发展相结合，土地覆盖特征对于涉及陆地生态系统和气候的许多研究以及人类对自然环境的影响至关重要，在许多不同的空间尺度上，各种用户群体都需要土地覆盖特征。 该项目将开发一个综合框架和原型系统，以便利用遥感和地理信息系统数据，以自动适应多种尺度信息的方式制作这类特征。 我们的研究在生态学，生物学，地理学，林业和环境科学领域有直接影响，涉及多尺度模式和过程。 更广泛地说，这些工具的开发和可用性将大大有助于提高科学界的理解，并最终在整个社会中，土地覆盖的复杂性。这个奖项是由数学科学司和社会，行为和经济科学理事会共同支持的数学科学优先领域的一部分。

Land cover characterization is essential for modeling, monitoring, and prediction of the terrestrial ecosystem and climate. Such characterizations are produced based largely on information extracted from satellite remote sensing measurements. However, much of the global and regional land cover consists of heterogeneous mixtures of different land cover types at varying spatial scales, while standard methods for remote sensing land cover characterization operate at a single scale and use a fixed set of categories. This research will develop an integrated statistical and computational framework that will produce land cover characterizations from remote sensing data that allow generalization in spatial resolution and categorical scale simultaneously. The result of this land cover characterization process will no longer be a land cover map, as a map implies a single spatial scale, but rather a land cover database, which can be queried in both traditional manners and ways currently unavailable. In essence, users will be free to select a range of spatial and categorical scales most appropriate for their needs. The key elements involved in the development of this framework include new contributions to the fields of geography and remote sensing, statistics, and computer science. At the foundation is a new class of multi-scale statistical models , called mixlets, from which there will result a new paradigm for representation and visualization of land cover categorization. In turn, these advances will be integrated with new developments in spatial database representations and spatial query systems.Land cover characterizations, critical for many studies involving terrestrial ecosystems and climate, and the human impacts on the natural environment, are needed at many different spatial scales and by a variety of user communities. This project will develop a comprehensive framework and prototype system for producing such characterizations in a manner that adapts automatically to multiple scales of information using remote sensing and GIS data. Our research has direct impact in the fields of ecology, biology, geography, forestry and environmental sciences dealing with multiscale patterns and processes. More broadly, the development and availability of these tools will contribute significantly to the improved understanding within the scientific community, and ultimately in the community at large, of the complexity of land cover.This award is jointly supported by the Division of Mathematical Sciences and the Directorate for Social, Behavioral, and Economic Sciences as part of the Mathematical Sciences Priority Area.