CAREER: Scalable Remote Sensing Computational Framework for Near-real-time Crop Characterization

地球观测卫星数量的不断增加，以及遥感数据的爆炸式增长，极大地促进了全球陆地表面特征的及时描述。在一些国家和国际农业倡议的推动下，近乎实时的作物类型特征已成为提供粮食不安全预警和及时作物产量预测以及提高全球粮食市场透明度的关键。然而，由于难以及时收集作物地面参考数据，现有特征模型的推广能力有限，以及缺乏适当的遥感网络基础设施，近实时作物类型表征仍然是农业遥感的一个挑战。卫星遥感和计算网络基础设施的最新进展为应对这一挑战开辟了新的途径。该项目的总体目标是建立一个可扩展的遥感计算框架，用于近实时作物类型表征，并促进计算遥感教育。计算框架可以改变大规模农业监测范式，以满足全球农业倡议的及时作物特征要求。该框架可以大大提高快速应对新出现的粮食危机的能力，并在推进广泛的遥感和农业研究方面产生跨领域影响。协同教育和推广活动为从K-12到研究生阶段的学生提供了关于计算遥感的独特学习机会，并将扩大代表性不足的学生对计算机的参与。这些活动还促进了跨一系列学科的计算框架的开放开发和采用。因此，这项研究与美国国家科学基金会促进科学进步和促进国家健康、繁荣和福利的使命是一致的。先进的遥感计算框架侧重于开发一个名为CropSight的基准数据存储库、一个作物表征建模系统和一个尖端的遥感网络基础设施，以促进近实时的作物和地表表征。CropSight是一个独特的国家尺度的作物地面参考数据库，包含了丰富的季节性遥感作物生长和美国大多数作物类型作物生长地点的环境属性。CropSight可以推广到大陆和全球尺度，并将作为一个大规模、系统和一致的地面参考数据库使用。作物表征系统包括一套新颖的基于深度学习的计算模型，这些模型可以融合来自一组地球观测卫星的图像，以便及时监测作物，并通过对复杂作物与环境相互作用的创新建模来识别不同的作物类型。该系统将增加作物类型表征的建模通用性，并具有在广泛地理区域外推的相当大的潜力。遥感网络基础设施将包括高度可扩展的云原生实现CropSight和近实时的按需作物监测系统。通过无服务器架构，该项目将构建集成各种地理空间数据源的云中间件，并实现数据密集型遥感数据分析，以便及时进行作物特征描述。网络基础设施将使模式从传统的计算有限的遥感分析转变为行星尺度的大规模图像分析，以便及时监测陆地表面。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。

The increasing proliferation of earth observation satellites, along with the explosive growth of remote sensing data, has dramatically facilitated timely land surface characterization worldwide. With several national and international agricultural initiatives, near-real-time crop type characterization has become vital for providing early warnings on food insecurity and timely crop yield forecasting, and for global food market transparency. However, near-real-time crop type characterization remains a challenge in agricultural remote sensing, due to the difficulty in collecting timely crop ground reference data, the limited generalizability of existing characterization models, and the lack of appropriate remote sensing cyberinfrastructure. Recent advances in satellite remote sensing and computational cyberinfrastructure open a new avenue to tackle the challenge. The overarching goal of the project is to establish a scalable remote sensing computational framework for near-real-time crop type characterization and to promote computational remote sensing education. The computational framework can transform the large-scale agricultural monitoring paradigm to meet the timely crop characterization requirements of global agricultural initiatives. The framework can substantially boost the ability to respond rapidly to emerging food crises, as well as create cross-cutting impacts in advancing a broad spectrum of remote sensing and agricultural research. The synergistic education and outreach activities offer unique learning opportunities about computational remote sensing to students from K-12 to the graduate level, and will broaden the participation of underrepresented students in computing. These activities also facilitate the open development and adoption of the computational framework across a range of disciplines. Therefore, this research aligns with the NSF mission to promote the progress of science and to advance the national health, prosperity, and welfare.The advanced remote sensing computational framework focuses on the development of a benchmark data repository called CropSight, a crop characterization modeling system, and a cutting-edge remote sensing cyberinfrastructure, to catalyze near-real-time crop and land surface characterizations. CropSight is a unique national-scale crop ground reference data repository, and embodies a wealth of season-long remotely sensed crop growth and environmental attributes across crop growing locations for most crop types in the U.S. CropSight can be generalized to continental and global scales, and will be used as a large-scale, systematic, and consistent ground reference data repository. The crop characterization system comprises a suite of novel deep learning-based computational models that can fuse the imagery from a set of earth observation satellites for timely crop monitoring, as well as identify varying crop types via innovative modeling of complex crop-environment interactions. The system will increase modeling generalizability for crop type characterization, and holds considerable potential to be extrapolated over wide geographical regions. The remote sensing cyberinfrastructure will include a highly scalable and cloud native implementation of the CropSight and a near-real-time on-demand crop monitoring system. With a serverless architecture, the project will build the cloud middleware that integrates various geospatial data sources and enables data-intensive remote sensing data analytics for timely crop characterization. The cyberinfrastructure will empower the paradigm shift from conventional compute-limited remote sensing analysis to planetary-scale massive imagery analysis for timely land surface monitoring.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.