Inferring surface and soil variables from assimilation of atmospheric boundary layer observations

大气边界层的演变是通过能量和水汽通量与地面和土壤状态强烈耦合的。这种相互作用被用于气象数据同化，以从观测到的屏蔽级大气变量的预测偏差中得出土壤变量，特别是土壤湿度。一些气象部门也在业务上使用这一既定方法，将在研究股的综合数据同化计划中加以实施，以量化大气边界层观测的信息量。作为一种新的研究方法，我们将在两个主要方向上扩展ABL观测的同化范围：第一，通过数据同化来额外估计模式参数，例如叶面积指数，以避免预报误差过于归因于土壤的状态变量。其次，我们不仅将预报和观测的瞬时差异视为所谓的创新，还将考虑预报误差的时间结构。这将解释地表和土壤变量变化的非常不同的时间尺度：例如，表面温度在几秒钟内变化，而土壤湿度在几小时到几个月的尺度上变化(取决于深度)，以及参数，如干燥土壤的导热系数，在时间上是恒定的。因此，这些量的误差会导致典型的大气预报误差，我们称之为“指纹”，因为它们允许识别误差源。虚拟集水区非常适合确定这些指纹：在选定的网格点，简化的一维陆地-大气模式将被用来执行集合模拟，其中指纹变量的系统扰动和其他模型变量的随机扰动。我们将确定哪些大气量在哪些条件下以及在哪些时间尺度上受系统扰动的影响最大。作为第二步，将为一维模式安装集合卡尔曼滤波(EnKF)同化系统，并对其进行扩展，以允许模式参数的同化。最后，将新的指纹作为观察算子在该框架中实现。这第一个实现将进一步扩展到考虑非瞬时观测。我们将审查这一多时间尺度扩展的现有概念，并确定最合适的方法。计划在第二项目阶段实施这一改进，并将首先为单柱模型制定的方法调整为三维耦合模型系统。

The evolution of the atmospheric boundary layer is strongly coupled via energy and moisture fluxes with the state of the surface and soil. This interaction is used in meteorological data assimilation to derive soil variables, especially soil moisture, from deviations of forecasts of screen-level atmospheric variables from the observations. This established approach, which is also used operationally by some weather services, will be implemented in the integrated data assimilation scheme of the Research Unit to quantify the information content of atmospheric boundary layer (ABL) observations. We will first consider screen-level temperature and humidity as ABL observations and add later boundary layer height and skin temperature.As a novel research approach, we will extend the assimilation of ABL observations in two main directions: First, model parameters, e.g. leaf area index, will be estimated additionally by the data assimilation to avoid a too strong attribution of forecast errors to state variables of the soil. Second, we will consider not only instantaneous differences of forecasts and observations as so-called innovations, but also take the temporal structure of forecast errors into account. This will account for the very different temporal scales at which surface and soil variables vary: Surface temperature, for example, changes within seconds, whereas soil moisture varies at scales of hours to months (depending on the depth), and parameters, like e.g. the heat conductivity of a dry soil, are constant in time. Accordingly, errors of these quantities result in characteristic atmospheric forecast errors, which we call “fingerprints”, since they allow identifying the error source.The virtual catchment is perfectly suited to determine these fingerprints: At selected grid points, a simplified one-dimensional land-atmosphere model will be used to perform ensemble simulations with systematic perturbations of the fingerprint variable and random perturbations of other model variables. We will identify which atmospheric quantities under which conditions and at which timescales are affected most by the systematic perturbations. As second step, an Ensemble Kalman Filter (EnKF) assimilation system will be installed for the one-dimensional model and expanded to allow for an assimilation of model parameters. Finally, the novel fingerprints will be implemented as observation operator in this framework. This first implementation will be further expanded to account for non-instantaneous observations. We will review existing concepts for this multi-timescale extension and identify the moist suitable approach. The implementation of this enhancement as well as the adaptation of the methodology, which will be developed first for single-column models, to the three-dimensional coupled model system is planned for the second project phase.