Observing System Simulation Experiments and Objective Analysis Tests in Support of the Goals of the Atmospheric Radiation Measurement Program

Abstract

Time continuous data assimilation or four-dimensional data assimilation (FDDA) is a collection of techniques where observations are ingested into a numerical model during the simulation in order to produce a physically balanced estimate of the true state of the atmosphere. Application of FDDA to the mesoalpha and subalpha scales is relatively new. One of many strategies for undertaking FDDA on the mesoscale is to employ Newtonian relaxation on increasingly finer horizontal grids. Encouraging results were found using this technique by Kuo et al. on a 40-km grid and by Stauffer and Seaman in a nested model with a 10-km inner grid. In these studies, the model is nudged toward the observations through adding an extra term(s) based on the difference between observations and the model predictions to the model?s prognostic equation(s). Since the model must retain a balance, this adjustment is spread over relatively large spatial and long temporal scales, and the nudging term is also multiplied by a coefficient that keeps the adjustment relatively small. Despite the positive findings of past studies, a number of questions arise in the application of this technique to fine grids. One area yet to be tested is how nudging will behave on fine grids under conditions with sharp horizontal and temporal gradients. Little improvement or even degradation of the model by the nudging might be expected as the timescale of nudging is relatively slow compared to the rapid evolution of the atmosphere, and spreading the observations out in time and space may not be representative of the actual atmospheric conditions. Other questions include 1) how the behavior of nudging at these scales and in active convection depends on boundary conditions, network density, and areal extent; 2) how the results depend on variations in the nudging coefficients; and 3) how nudging compares to simple objective analysis of the observations. In this study, Newtonian relaxation is used in a moist, full physics, nonhydrostatic mesoscale model to conduct simulations with horizontal resolutions as fine as 5 km in environments with deep convection and in mountainous terrain. Observing system simulation experiments were designed to address the previously mentioned questions. The authors show that nudging on these scales and in these conditions tends not to produce any large degradations but instead leads to improvements in the simulations even with a small number of observing sites. In applying nudging to a limited mesoscale area, the authors found that the results were more favorable if the nudging was undertaken over larger regions, which supports the nested approach used by Stauffer and Seaman. Some negative aspects of nudging were also uncovered with locally high rms errors due to data representativity problems and predictability issues. The accuracy of objective analysis was also explored and discussed in the context of the Atmospheric Radiation Measurement (ARM) Program. In agreement with Mace and Ackerman, the errors associated with objective analysis can be too large for the goals of ARM. However, the authors also found that a method proposed by Mace and Ackerman to detect time periods where significant errors exist in the objective analysis was not valid for this case. Based on this work, the authors propose that for a modest network of observing sites FDDA has a number of advantages over objective analysis.