Deep Learning in Geospatial Uncertainty Modeling

该项目将为分析复杂和不兼容的地理空间数据提供新的统计方法和工具，以推进不确定性意识和数据驱动的地理知识发现。随着地理空间数据变得越来越可用，以复杂模式和不确定性来源为特征的地理空间数据的复杂性急剧增加，远远超出了传统方法的能力。该项目将开发一个不确定性框架，以支持复杂时空模式下不兼容地理空间数据的统计学习。该框架将用于表征和量化科学建模和实际应用中的地理空间不确定性。该项目将为涉及地理空间数据的基础分析、建模和集成问题提供新颖的解决方案，并促进对地理空间不确定性本质的理解。该项目将提高地理空间数据的正确和具有成本效益的利用，并将对地理空间数据涉及的学科产生更广泛的影响。此外，通过对不确定性空间思维的公众宣传，该项目将通过提高公众对地理空间不确定性和关键地图阅读和使用的认识来促进公共利益。从地质统计学的角度来看，该项目将解决空间分析中长期未解决的挑战，处理复杂的时空模式，地理空间数据的异质性和不确定性。首先，开发的框架将提供一种全面和实用的方法来无缝集成异构地理空间数据源。其次，利用深度学习技术的最新进展，开发的框架将对异构数据中隐含的复杂时空模式进行表征和建模。第三，开发的框架将使地理空间不确定性的表征和建模以及域域的影响评估成为可能。将在两个领域的应用中评估所开发方法的性能：公共卫生中的时空疾病制图和土地覆盖变化的不确定性建模及其对大气模型的影响。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。

This project will contribute new statistical methods and tools for the analysis of complex and incompatible geospatial data to advance uncertainty-aware, data-driven geographic knowledge discovery. As geospatial data become increasingly available, the complexity of geospatial data marked by complex patterns and sources of uncertainty increases dramatically, and goes well beyond the capabilities of conventional methods. This project will develop an uncertainty framework to support statistical learning of incompatible geospatial data for complex spatiotemporal patterns. This framework will be used to characterize and quantify the geospatial uncertainty in scientific modeling and practical applications. This project will offer novel solutions to fundamental analysis, modeling, and integration problems involving geospatial data, and advance the understanding of the nature of geospatial uncertainty. This project will enhance the proper and cost-effective utilization of geospatial data, and will have broader impacts on disciplines that geospatial data are involved. Furthermore, with a public outreach on uncertainty-aware spatial thinking, this project will advance the public good by increasing the public awareness of the geospatial uncertainty and critical map reading and usage.From a geostatistics perspective, this project will address the long-standing unsolved challenges in spatial analysis dealing with complex spatiotemporal patterns, heterogeneity and uncertainty of geospatial data. First, the developed framework will provide a comprehensive and practical approach to seamlessly integrate heterogeneous sources of geospatial data. Secondly, capitalizing on recent advances in deep learning technologies, the developed framework will characterize and model the complex spatiotemporal patterns implied in heterogeneous data. Thirdly, the developed framework will enable the characterization and modeling of geospatial uncertainty and the impact assessment in domain fields. The performances of the developed methods will be evaluated in two domain applications: spatiotemporal disease mapping in public health and modeling uncertainty of land cover changes and the impact on atmospheric models.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.