论文标题
部分可观测时空混沌系统的无模型预测
Towards hourly three-dimensional ensemble data assimilation of screen-level observations into coupled atmosphere-land models
论文作者
论文摘要
储层计算是预测湍流的有力工具，其简单的架构具有处理大型系统的计算效率。然而，其实现通常需要完整的状态向量测量和系统非线性知识。我们使用非线性投影函数将系统测量扩展到高维空间，然后将其输入到储层中以获得预测。我们展示了这种储层计算网络在时空混沌系统上的应用，该系统模拟了湍流的若干特征。我们表明，使用径向基函数作为非线性投影器，即使只有部分观测并且不知道控制方程，也能稳健地捕捉复杂的系统非线性。最后，我们表明，当测量稀疏、不完整且带有噪声，甚至控制方程变得不准确时，我们的网络仍然可以产生相当准确的预测，从而为实际湍流系统的无模型预测铺平了道路。
We explore the potential of three-dimensional data assimilation for assimilating sparsely-distributed 2-metre temperature observations across the coupled atmosphere-land interface into the soil moisture. Using idealised twin experiments with the limited-area modelling platform TerrSysMP and synthetic observations, we avoid model biases and directly control errors in the initial conditions and observations. These experiments allow us to test hourly data assimilation with a localised ensemble Kalman filter, as often used for mesoscale data assimilation. We find here an error reduction of such an ensemble Kalman filter approach compared to daily-updating with a one-dimensional simplified extended Kalman filter. We attribute this improvement to the ensemble approximation of the sensitivities and the more frequent updates with the ensemble Kalman filter. The hourly updates result hereby into a positive assimilation impact during daytime and a neutral impact during night. With a three-dimensional ensemble Kalman filter, we can directly assimilate screen-level observations at their respective position into the soil moisture, skipping the otherwise needed spatial interpolation step. These findings suggest an emerging potential for the localised three-dimensional ensemble Kalman filter to hourly assimilate screen-level observations into coupled atmosphere-land models.
