CAREER: CAS-Climate: Forecast-informed Flexible Reservoir System Modeling Enabled by Artificial Intelligence Algorithms Using Subseasonal-to-Seasonal Hydroclimatological Forecasts

突发的极端天气、变化的气候和频繁的自然灾害，如洪水和干旱，给我国水库系统的有效、可持续和灵活运行带来了新的挑战。为了避免由于极端事件期间运行灵活性不足和不可预测的水通量而导致水库失效，大坝运营商需要两个基本项目：(1)在较长的提前期（从几天到几个月不等）进行准确可靠的水文预报；(2)强大的自适应决策支持工具，不仅可以帮助实时决策在特定时间释放多少水，还可以使水库运营商灵活地将工程约束和水文气候预测情景纳入灵活的释放计划。在过去的四十年里，在确定性预测、线性规划、优化算法和基于规则的模拟模型方面取得了重大的科学进展，以指导油藏作业。然而，由于目前在理解亚季节到季节（S2S）水文气候预报的变动性方面的局限性，以及缺乏利用集合预报进行更有效的放水决策的建模能力，这些方法无法解决未来的业务挑战。因此，本CAREER项目的目标是双重的：1)开发一个综合解决方案，可以解释降水的时空变化及其不确定性；2)开发一种新的人工智能数据挖掘（ai & DM）决策支持工具，使油藏运营商能够使用改进的集成预测来制定自适应释放策略。本研究的目标是在水库运行和规划中更好地响应和减轻极端天气事件和气候不确定性的影响。该项目将(1)利用最先进的深度学习模型的优势，从北美多模式集成数据集中的多个预测模型中发现并纠正与S2S降水预测相关的时空误差；(2)建立基于自适应集成提升树的预测控制模型，将改进的集成预测有效地纳入基于场景的水库释放模拟中，以实现规划目的。将进行水文建模和不确定性分析，以帮助了解气象不确定性如何从大气条件传播到水资源规划和基础设施管理。将在美国进行大规模水文验证实验（超过671个流域）和水库模拟（超过316个水坝）。研究结果将用于验证改进后的预测，量化集合水文预测的不确定性，并评估基于预测的油藏决策支持工具。AI&amp；DM模型将与美国垦务局（USBR）和美国陆军工程兵团（USACE）合作进行全面测试，这两家机构是美国的两大水库机构。预期结果旨在使水库运营商能够制定合适的水库储存和释放策略，以解决来水突然变化或供水不足的问题，同时满足各种需求和限制。教育任务与研究紧密相连。积极的学习活动将帮助研究生培养解决复杂研究问题的能力。本科生将获得编程技能。外展活动包括在俄克拉荷马州诺曼的国家气象博物馆和科学中心（NWMSC）举办一年一度的“水节”展览。在这个职业项目期间和之后，博物馆的参观者和孩子们将见证水文学、气象学、水资源管理的重要性，以及极端天气和气候的影响。美国国家科学基金会资助的CUAHSI组织也将与该项目合作，通过各种教育和推广活动，最大限度地扩大已开发数据、模型和算法的影响。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。

Abrupt weather extremes, changing climate, and frequent natural hazards, such as floods and droughts, have created new challenges for the effective, sustainable, and flexible operation of our nation’s reservoir systems. To avoid reservoir failures due to insufficient operational flexibility and unpredictable water fluxes during extreme events, dam operators need two essential items: (1) accurate and reliable hydrological forecasts at extended lead times (ranging from days to months in the near future); and (2) powerful and adaptive decision support tools, which not only could assist real-time decision making about how much water to release at a certain time, but also allow reservoir operators to nimbly incorporate engineering constraints and hydroclimatological forecast scenarios into flexible release planning. Over the past four decades, significant scientific advancements have been made in deterministic forecasts, linear programming, optimization algorithms, and rule-based simulation models to guide reservoir operations. However, these approaches are unable to address future operational challenges due to current limitations in understanding the variabilities of Subseasonal-to-Seasonal (S2S) hydroclimatological forecasts and a lack of modeling capabilities that utilize ensemble forecasts for more effective water release decision-making. Therefore, the goals of this CAREER project are twofold: 1) to develop an integrated solution that can account for the spatial and temporal variability of precipitation and its uncertainty; and 2) to develop a novel Artificial Intelligence & Data Mining (AI&DM) decision support tool that allows reservoir operators to use improved ensemble forecasts to develop adaptive release strategies. This research targets enabling better response to, and mitigation of the impacts of, extreme weather events and climate uncertainty in reservoir operation and planning.The project will (1) leverage the advantages of state-of-the-art deep learning models to discover and correct the spatial and temporal errors associated with S2S precipitation forecasts from multiple forecasting models in the North American Multi-Model Ensemble dataset; and (2) develop an adaptive Ensemble Boosting Tree-based Predictive Control Model, which can effectively incorporate improved ensemble forecasts into scenario-based reservoir release simulations for planning purposes. Hydrological modeling and uncertainty analysis will be performed to help understand how meteorological uncertainty propagates from atmospheric conditions into water resources planning and infrastructure management. Large-scale hydrological validation experiments (over 671 watersheds) and reservoir simulations (over 316 dams) across the U.S. will be conducted. The results will be used to validate the improved forecasts, quantify the ensemble hydrological forecast uncertainty, and evaluate the forecast-informed reservoir decision support tool. The AI&DM models will be comprehensively tested in collaboration with the U.S. Bureau of Reclamation (USBR) and the U.S. Army Corps of Engineers (USACE), which are two major reservoir agencies in the USA. The expected outcomes are aimed to allow reservoir operators to develop suitable reservoir storage and release strategies that address sudden fluxes of incoming water or a lack of water supply, while simultaneously meeting various demands and constraints. Educational tasks are tightly coupled with research. Active learning activities will help graduate students develop the ability to tackle complex research problems. Undergraduate students will obtain skills in programming. Outreach includes hosting an annual “Water Festival” exhibit at the National Weather Museum and Science Center (NWMSC) in Norman, Oklahoma. During and beyond this CAREER project, museum visitors and children will witness the importance of hydrology, meteorology, water resources management, and the impacts of extreme weather and climate. The NSF-funded CUAHSI organization will also collaborate with the project to maximize the broader impacts of developed data, models, and algorithms via various educational and outreach activities.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.