CPS: Medium: Making Every Drop Count: Accounting for Spatiotemporal Variability of Water Needs for Proactive Scheduling of Variable Rate Irrigation Systems

我们都依靠农业来维持生计。与海产品和牲畜相比，种植系统提供了主要的营养来源。种植系统的产量和生产力必须提高，以满足不断增长的人口的需求。一旦有了种子，一个成功的种植季节是由水决定的。这有两个来源：灌溉和降水。灌溉用水是农业的主要投入，特别是在半干旱和干旱地区。在最近对《水土资源保持法》的评估中，美国农业部将灌溉用水保护确定为一项全国性的需求。缺水引起应激，对作物生长和产量产生不利影响。另一方面，过度浇水会导致养分流失、土壤侵蚀和水资源浪费。农场还受到干旱、降水变化和生长季节延长的不利影响。拟议的努力强调水资源管理和保护，这是对农场经常遇到的逆风的适应。这项工作解决了过度浇水（土壤侵蚀和养分流失）和浇水不足（不利作物产量和压力）的相关问题，同时确保农业系统的可持续性和盈利能力。该项目的总体目标是开发一个端到端的网络物理智能系统，该系统可以预测给定田地的时空作物需水量，并实施可变速率灌溉策略，以优化整个田地的作物产量。我们用有限数量的原位土壤含水量传感器对田地进行测量；这些实地观测得到雷达和卫星遥感数据的补充。这项工作包括设计基于深度神经网络（DNN）的新型AI（人工智能）方法，以生成对水需求的预测。这些深度神经网络在多模式、高维数据上运行，以识别田地不同部分的土壤水分缺失和变化。生成的预报考虑了作物、土壤类型、降水事件和作物生长阶段。该项目在人工智能引导的网络物理系统中关闭了感知环境和驱动之间的循环。这些预测在基于博弈论的算法中得到利用，通过处方计划来精确驱动浇灌臂，控制喷嘴和区域水平的浇灌速率。该算法对降水事件、预报的不确定性和驱动开销具有自适应和响应能力。这项多方面的研究通过创新地结合传感环境、算法博弈论、科学模型和领域科学以及人工智能/深度神经网络，推进了网络物理系统的科学。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。

We all depend on agriculture for sustenance. When compared to seafood and livestock, cropping systems provide the primary source of nutrition. Yields and productivity of cropping systems must grow to meet the demands of a growing population. Once seeds are available, a successful cropping season is determined by water. There are two sources for this: irrigation and precipitation. Irrigation water is a major input to agriculture, especially in semi-arid and arid regions. In a recent appraisal for the Soil and Water Resources Conservation Act, the USDA identified irrigation water conservation as a national need. Under-watering induces stresses and adversely impacts both crop growth and yields. Over-watering, on the other hand, leads to nutrient runoff, soil erosion, and water waste. Farms are also impacted by the adverse effects of droughts, variability in precipitation, and lengthening of the growing season. The proposed effort with its emphasis on water management and conservation represents an adaptation to the head winds often encountered at farms. The effort addresses the interrelated aspects of over-watering (soil erosion and nutrient runoff) and underwatering (adverse crop yields and stress) while ensuring sustainability and profitability of agricultural systems.The overarching objective of this project is to develop an end-to-end cyber-physical intelligence system that forecasts space-time crop water needs in a given field and implements variable rate irrigation strategies to optimize crop yield throughout the field. We instrument the field with a limited number of in-situ soil moisture content sensors; these in situ observations are complemented with remotely sensed data from radars and satellites. The effort includes design of novel AI (Artificial Intelligence) methods based on deep neural networks (DNN) to generate forecasts of water needs. These DNNs operate on multimodal, high-dimensional data to identify soil moisture deficits and variability in different parts of the field. The generated forecasts account for crop, soil type, precipitation events, and the crop growing phase. The project closes the loop between the sensing environment and actuation within the AI-guided cyber physical system. These projections are leveraged within a game theory based algorithm to inform precise actuations of the watering arm with prescription plans that control watering rates at the nozzle and zone level. The algorithm is adaptive and responsive to precipitation events, uncertainty in the forecasts, and the actuation overheads. This multifaceted research advances the science of cyber-physical systems by innovatively combining sensing environments, algorithmic game theory, scientific models and domain-science, and AI/DNNs.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.