BIGDATA: F: Advancing Deep Learning to Monitor Global Change

世界人口的增长以及工业化和城市化的加速使本已稀缺的自然资源和粮食供应更加紧张，这些资源和粮食供应必须扩大规模以满足日益增长的需求。由此产生的大规模变化的后果包括对环境的巨大压力，以及对我们养活世界人口的能力的挑战，如果不能以可持续的方式加以管理，以目前的变化速度，这可能是灾难性的。应对这一挑战需要及时提供有关全球变化的信息；例如，农业土地生产力绩效的变化；森林向农田或人工林的转变，以及城市化导致的生产性耕地的流失；以及水土退化。为了应对监测全球变化的这些挑战，该项目将开发先进的机器学习技术，特别是深度学习。该项目的主要重点将是分析通过美国和国际机构从卫星上的各种仪器和传感器获得的遥感数据。这些丰富的数据集捕捉了塑造地球自然景观和环境质量的自然过程和人类活动的多个方面，从而为研究和更好地塑造全球变化的性质和影响提供了机会。该项目旨在极大地推进最先进的机器学习技术，用于分析地球系统过程的多尺度、多源、时空数据。具体而言，该项目将推进深度学习技术，以应对使用遥感数据进行全球变化监测所面临的挑战。深度学习已经成功地解决了许多领域的问题，这些领域涉及具有空间和时间（顺序）信息的复杂数据集，如视觉、视频和自然语言处理。深度学习的前景主要源于其利用复杂映射关系的能力，以及使用大量训练数据在空间和时间上提取判别特征的能力。因此，在将深度学习技术应用于遥感数据方面出现了一系列活动。然而，由于环境应用所特有的挑战，为计算机视觉等相关应用开发的现成深度学习技术的实用性有限。本研究将解决这些挑战，并通过开发能够利用潜在的复杂、多尺度、时空地球系统过程进行全球变化监测的技术，推进最先进的深度学习技术。本项目开发的方法预计也会对处理大规模时变数据的许多学科产生影响。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。

Growth in the world's population and the acceleration of industrialization and urbanization are straining already scarce natural resources and food supplies, which must scale up to keep pace with growing demand. The consequences of the resulting large-scale changes include tremendous stresses on the environment, as well as challenges to our ability to feed the world's population, which could be calamitous at the current rate of change if not managed sustainably. Meeting this challenge will require timely information on global changes; for example, the changing productivity performance of land in agriculture; the conversion of forest to farmland or plantations, and the loss of productive farmland due to urbanization; and soil and water degradation. To address these challenges in monitoring global change, this project will develop advanced machine learning techniques, especially deep learning. The project's primary focus will be on the analysis of remote sensing data, available from a variety of instruments and sensors aboard satellites through United States and international agencies. These rich datasets capture multiple facets of the natural processes and human activities that shape the physical landscape and environmental quality of our planet, and thus offer an opportunity to study and better shape the nature and impact of global changes.This project seeks to greatly advance the state-of-the-art in machine learning techniques for analyzing the multi-scale, multi-source, spatio-temporal data about earth system processes. Specifically, this project will advance deep learning techniques to meet the challenges involved in using remote sensing data for global change monitoring. Deep learning has been successful in addressing problems in a number of domains involving complex data sets with spatial and temporal (sequential) information such as vision, video, and natural language processing. The promise of deep learning mainly stems from its capacity to exploit complex mapping relationships and extract discriminative features over space and time using large volumes of training data. Consequently, there has been a flurry of activity in applying deep learning techniques to remote sensing data. However, due to challenges that are unique to environmental applications, off-the-shelf deep learning techniques developed for related applications such as computer vision have limited utility. This research will address these challenges and advance the state-of-the-art in deep learning techniques by developing techniques that can make use of the underlying complex, multi-scale, spatio-temporal earth system processes for global change monitoring. Methods developed in this project are expected to also have an impact across many disciplines that deal with large-scale time-varying data.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.