SBIR Phase I: Operational Seasonal Forecasting of Environmental Data using Machine Learning and Statistical Methods

SBIR第一阶段项目的更广泛的影响/商业潜力是将天气预报能力扩展到目前的14天窗口之外。该项目将利用来自全球网格季节预报数值模型的公开信息和来自物联网（IoT）传感器的现场观测，用于基于机器学习的预报系统，提供季节和亚季节预报。 由于天气，特别是极端天气事件，继续影响各个行业和广泛的单个资产，因此需要这种特定于资产的信息。 准确的预测可以改善物流，供应链，劳动力调度，并影响对美国经济至关重要的各种作物的种子选择，包括核果，苹果和梨。 此外，技术革新有可能改进当地对与农业、能源和保险业有关的环境变量的估计，这一项目将使环境变量的现场季节性和次季节性预测系统投入运作。 所提出的创新包括使用基于Koopman算子的机器学习近似的数据驱动的模拟预测方法进行环境预测的系统和方法，该Koopman算子控制非线性动态系统中的可观测量的演变。该项目将通过探索次月时间尺度，并通过将预测器集扩大到大多数统计方法所使用的传统海面温度之外，提高预测精度，使其超过粗分辨率动力学模型。具体来说，该项目将创建传递函数，将来自动态模型的粗分辨率预测器与来自支持物联网的环境传感器的本地信息（预测值）联系起来。此外，该项目将应用和评估这一框架，对美国太平洋西北部和夏威夷的气候进行季节性和区域性预测，如累积降水或特定季节的开始，以及对霜冻或干旱等极端事件的概率预测。该奖项反映了NSF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。

The broader impact/commercial potential of this SBIR Phase I project is to extend weather forecasting capabilities beyond the current 14-day window. The project will leverage publicly available information from global gridded seasonal forecasting numerical models and in-situ observations from sensors for the Internet of Things (IoT) for a machine-learning based forecasting system that delivers seasonal and sub-seasonal forecasts. This asset-specific information is needed as weather, especially extreme weather events, continue to affect various industries and a wide range of individual assets. Accurate forecasting can improve logistics, supply-chain, labor scheduling and affect seed selection of a variety of crops important in the US economy, including stone-fruits, apples and pears. In addition, the technical innovation has the potential to improve local estimates of environmental variables relevant to the agricultural, energy and insurance industries.This project will bring into operation an in-situ seasonal and sub-seasonal forecasting system of environmental variables. The proposed innovation includes systems and methods for environmental forecasting using data-driven analog forecasting methods based on machine learning approximation of Koopman operators, governing the evolution of observables in nonlinear dynamical systems. The project will improve the forecasting accuracy beyond that of coarse resolution dynamical models by exploring sub-monthly timescales and a by expanding the predictor set beyond the traditional sea surface temperature used by most statistical approaches. Specifically, the project will create transfer functions that link the coarse resolution predictors from the dynamical models with local information (predictands) from IoT enabled environmental sensors. Furthermore, the project will apply and assess this framework for seasonal and regional predictions of the U.S. Pacific Northwest and Hawaii’s climate such as cumulative precipitation or onset of the given season, and probabilistic forecast of extreme events such as frosty days or drought, across a range of sectors.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.