CAREER: AI-enabled Integrated Nutrient, Streamflow, and Parcel sImulation for Resilient agroEcosystems (INSPIRE): a framework for climate-smart crop production and cleaner water

气候智能型农业实践有望减少农业中的碳(C)排放，但其实施往往需要复杂的权衡，特别是影响氮(N)和磷(P)的管理。碳、氮和磷的综合管理，以确保气候智能型作物生产，同时保持清洁水源，受到若干知识和技术差距的阻碍。为了解决这一重大挑战，该项目旨在显著推进对密西西比河上游流域相互关联的碳、氮、磷和水循环的整体理解和建模。为了实现这一目标，将开发一个基于人工智能的框架，用于综合营养、水流和弹性农业生态系统的包裹模拟（INSPIRE），该框架可以轻松地吸收多源观测数据，并提供从田间到流域尺度的准确和快速量化。该项目的成果有望为政策制定者和农业社区提供有价值的见解，特别是在优化管理实践以改善美国中心地带的碳固存、土壤健康和水质方面。此外，该项目将其研究目标与教育议程相结合，其特点是开发一种计算工具，以促进大学生广泛参与大规模计算。该项目还将引入一个信息物理分水岭mesocosmos，作为使用数字孪生技术加强与农业和环境可持续性相关的STEM教育的创新试验。与现有的基于过程的建模方法相比，人工智能辅助的碳、氮、磷和水通量综合模拟能够更好地捕捉高分辨率的环境变化，并确定在不牺牲作物生产的情况下实现气候智慧型农业和水质目标的最佳做法，该项目将基于这一总体假设进行开发。科技创新将通过四个目标来实现。首先，将开发基于知识引导机器学习（KGML）的INSPIRE-Field模型，以显著提高现场水平C、N、P和水文相互作用的预测精度。其次，INSPIRE-Field将与基于图神经网络（GNN）的水文替代模型相结合，该模型首先汇总小流域（即INSPIRE-Watershed）内的现场水和养分通量，然后将流域输出路由到整个密西西比河上游流域（即INSPIRE-Basin）。为了降低INSPIRE的不确定性，本文将探索一种新的表征学习方法，通过低维嵌入有效地同化遥感和原位遥感数据。第三，将开发一个用户友好的网络界面，使利益相关者能够预览不同气候智能管理实践的结果，并根据碳、氮、磷和水文目标的多目标优化确定特定领域的首选管理策略。最后，在农业生态系统工程师、教育工作者和决策者的未来劳动力中，计算、传感和机器学习的教育和实践将通过项目活动得到加强。通过整合遥感、机械建模和人工智能，引领可持续农业数据分析的前沿，实现对每一块农田的监测和管理，跟踪污染物，预测农业风险，为农民提供最佳解决方案，最大限度地减少对环境的负面影响，最终帮助世界实现可持续的粮食未来。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。

Climate-smart agricultural practices hold the promise of reducing carbon (C) emissions from farming, yet their implementation often presents complex trade-offs, particularly affecting nitrogen (N) and phosphorus (P) management. Integrated management of C, N, and P to ensure climate-smart crop production while preserving clean waters is hindered by several knowledge and technology gaps. To approach a solution for this grand challenge, this project aims to significantly advance the holistic understanding and modeling of the interconnected C, N, P, and water cycles in the Upper Mississippi River Basin. This goal will be pursued by developing an AI-based framework of integrated nutrient, streamflow, and parcel simulation for resilient agroecosystems (INSPIRE) that can easily ingest multi-source observations and provide an accurate and speedy quantification from the field to basin scale. The outcomes from this project are expected to provide valuable insights for policymakers and farming communities, particularly in optimizing management practices for improved carbon sequestration, soil health, and water quality in the America's heartland. Additionally, this project intertwines its research objectives with an educational agenda, which is featured by developing a computational tool to foster broad participations in large-scale computing among undergraduates. The project will also introduce a cyber-physical watershed mesocosm as an innovative trial of using the digital twin technology to enhance STEM education related to agricultural and environmental sustainability.This project will develop under the overarching hypothesis that AI-assisted integrated simulation of C, N, P, and water fluxes, compared with existing process-based modeling approach, is better able to capture high resolution environmental variability and identify best practices for achieving climate-smart agriculture and water quality goals without sacrificing crop production. The scientific innovations will be achieved through four objectives. First, a Knowledge-Guided Machine Learning (KGML)-based INSPIRE-Field model will be developed to significantly improve the prediction accuracy of field-level C, N, P, and hydrological interactions. Second, INSPIRE-Field will be coupled with Graph Neural Network (GNN)-based hydrologic surrogate models that first aggregate field water and nutrient fluxes within small watersheds (i.e., INSPIRE-Watershed), and then routing watershed outputs throughout the Upper Mississippi River Basin (i.e., INSPIRE-Basin). To reduce the uncertainty of INSPIRE, a novel representation learning method to efficiently assimilate remote and in-situ sensing data via low-dimensional embeddings will be explored. Third, a user-friendly web interface will be developed that allows stakeholders to preview outcomes of different climate- smart management practices and identify field-specific preferred management strategies based on multiobjective optimizations for C, N, P, and hydrological goals. Finally, the education and practice of computing, sensing, and machine learning among the future workforce of agroecosystem engineers, educators, and decision-makers will be enhanced through project activities. The investigator aims to lead the frontier of data analytics for sustainable agriculture by integrating remote sensing, mechanistic modeling, and artificial intelligence, with the aspiration to enable monitoring and managing every cropland, track pollutants, forecast agricultural risks, provide farmers best solutions to minimize negative environmental impacts, and ultimately help the world to achieve a sustainable food future.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.