Non‐linear Error Compensation Technique with Associative Restoration (NECTAR) is a novel approach to the assimilation of fragmentary sensor data to produce a global nowcast of the near‐Earth space weather. NECTAR restores missing information by iteratively transforming (“morphing”) an underlying global climatology model into agreement with currently available sensor data. The morphing procedure benefits from analysis of the inherent multiscale diurnal periodicity of the geosystems by processing 24‐hr time histories of the differences between measured and climate‐expected values at each sensor site. The 24‐hr deviation time series are used to compute and then globally interpolate the diurnal deviation harmonics. NECTAR therefore views the geosystem in terms of its periodic planetary‐scale basis to associate observed fragments of the activity with the grand‐scale weather processes of the matching variability scales. Such approach strengthens the restorative capability of the assimilation, specifically when only a limited number of observatories is available for the weather nowcast. Scenarios where the NECTAR concept works best are common in planetary‐scale near‐Earth weather applications, especially where sensor instrumentation is complex, expensive, and therefore scarce. To conduct the assimilation process, NECTAR employs a Hopfield feedback recurrent neural network commonly used in the associative memory architectures. Associative memories mimic human capability to restore full information from its initial fragments. When applied to the sparse spatial data, such a neural network becomes a nonlinear multiscale interpolator of missing information. Early tests of the NECTAR morphing reveal its enhanced capability to predict system dynamics over no‐data regions (spatial interpolation).

中文翻译：

---

NECTAR模型变形对稀疏连续近地天气预报的吸收

带有联合恢复的非线性误差补偿技术（NECTAR）是一种新颖的方法，用于对碎片传感器数据进行同化以产生近地空间天气的全球临近预报。NECTAR通过将基础的全球气候学模型迭代转换（“变形”）成与当前可用的传感器数据一致，来恢复丢失的信息。通过处理每个传感器站点的实测值与气候期望值之间的差异的24小时历史记录，可以通过对地球系统固有的多尺度昼夜周期性进行分析来进行变态过程。24小时偏差时间序列用于计算然后全局内插日偏差谐波。因此，NECTAR从周期性的行星尺度角度来看待地球系统，以将观测到的活动碎片与匹配变率尺度的大尺度天气过程联系起来。这种方法增强了同化的恢复能力，特别是当临近天气预报的观测站数量有限时。NECTAR概念最有效的方案在行星尺度近地天气应用中很常见，尤其是在传感器仪器复杂，昂贵且因此稀缺的情况下。为了进行同化过程，NECTAR采用了在关联记忆架构中常用的Hopfield反馈递归神经网络。联想记忆模仿人类从其原始片段中恢复完整信息的能力。当将神经网络应用于稀疏空间数据时，它将成为丢失信息的非线性多尺度插值器。NECTAR变形的早期测试表明，它增强了预测无数据区域（空间插值）系统动态的能力。