III: Small: A Discriminative Modeling Framework for Mining of Spatio-Temporal Data in Remote Sensing

了解地球大气、陆地和海洋的物理和化学过程是重大的科学挑战。部分由于应对这些挑战的社会影响，近年来对大量用于地球观测的卫星和地面传感器进行了大量投资。这些传感器的观测结果用于估计重要的地球物理特性，如温度、云、气溶胶、温室气体、雪和冰，并用于旨在气候建模、天气预报、空气质量监测和疾病管理的科学研究。由于计算限制、建模和参数估计困难或无法提供不确定性估计，目前空间统计和数据融合技术达不到所需的水平。该项目的目标是开发方法，以帮助最好地利用来自卫星和地面仪器的大量多源观测资料，用于在覆盖范围、分辨率和质量方面具有不同能力的地球观测。本课题开发了一种基于高斯条件随机场模型的多传感器遥感数据融合判别建模框架。该框架具有灵活性、鲁棒性和计算效率，因此适合用于大型时空数据集。它允许在存在抽样偏差的情况下，从带有部分可观察属性的标记和未标记数据中学习。这些方法将在一个开源的、用户友好的软件工具中实现，供从业者使用。所产生的工具将在大气气溶胶和地表污染估计问题上进行评估，这些问题是气候研究和环境科学中的一些重要问题。这个项目的更广泛影响包括在时空数据挖掘和地质统计以及地球遥感方面的最新技术的方法学进展。它将改进气溶胶对地球辐射收支和气候影响的表征。数据融合框架可直接应用于许多其他大气、陆地和海洋性质的估计。本项目的研究活动与教育相结合。他们将帮助扩大从博士到K-12级别学生的参与，并通过天普大学已经建立的渠道，吸引来自代表性不足群体的学生，增加计算机科学的多样性。有关该项目的更多细节可以在http://www.dabi.temple.edu/~vucetic/nsf_fusion.htm上找到

Understanding the physical and chemical processes of Earth's atmosphere, land, and ocean presents significant scientific challenges. Due in part to the societal impact of addressing these challenges, recent years have seen significant investments in a large number of satellite and ground-based sensors dedicated to Earth observation. The observations from these sensors are used to estimate important geophysical properties such as temperature, clouds, aerosols, greenhouse gases, snow and ice, and used in scientific studies aimed at climate modeling, weather forecasting, air quality monitoring, and disease management. Current techniques from spatial statistics and data fusion fall short of what is needed because of computational constraints, difficulties in modeling and parameter estimation, or inability to provide uncertainty estimates. The objective of this project is to develop methods that help best utilize large quantities of multi-source observations from satellite and ground-based instruments for Earth observation having different capabilities regarding coverage, resolution, and quality. This project develops a novel discriminative modeling framework for fusion of multi-sensor remote sensing data based on the Gaussian conditional random field model. The framework is designed to be flexible, robust, and computationally efficient, and hence suitable for use on large spatio-temporal data sets. It allows learning from a mixture of labeled and unlabeled data with partially observable attributes, in the presence of sampling bias. The methods will be implemented in an open source, user-friendly, software tool for use by practitioners. The resulting tools will be evaluated on the problems of atmospheric aerosol and surface level pollution estimation, which are some of the important problems in climate research and environmental science. The broader impacts of this project include methodological advances in the current state of the art in spatio-temporal data mining and geostatistics as well as Earth remote sensing. It will enable improved characterization of the effects of aerosols on the Earth's radiation budget and climate. The data fusion framework is directly applicable to estimations of many other atmospheric, land, and ocean properties. The research activities in this project are integrated with education. They will help broaden the participation of students ranging from doctoral to K-12 level and increasing diversity in Computer Science through already established channels at Temple University for engaging students from underrepresented groups. Additional details about the project can be found at: http://www.dabi.temple.edu/~vucetic/nsf_fusion.htm