Experimental 2D-Var assimilation of ARM cloud and precipitation observations

A two-dimensional variational method (2D-Var) based on 12-hour integrations of the single-column version of the European Centre for Medium-Range Weather Forecasts model for adjusting initial temperature and specific-humidity profiles is used to identify the issues related to the assimilation of either high temporal frequency or time-accumulated observations that are directly affected by clouds and precipitation. 2D-Var experiments have been run using cloud-radar reflectivity profiles and microwave-radiometer brightness temperatures in nonprecipitating cloudy cases, and accumulated rain-gauge measurements and total column water-vapour retrievals from the Global Positioning System in precipitating situations. All observations have been obtained from the Atmospheric Radiation Measurement Program. In cloudy situations, the use of time sequences of cloud-radar reflectivities at half-hourly time steps is problematic because the variational assimilation minimization becomes overconstrained. Instead, it is better to assimilate time-averaged profiles of cloud radar reflectivities when background reflectivities are generally greater than the observations. When background reflectivities are lower than observed, the falling of the extra precipitation produced by 2D-Var hampers the convergence of the cost-function minimization. The assimilation of brightness temperatures seems more straightforward even at a half-hourly frequency. As regards the application of 2D-Var to accumulated rain-gauge measurements, the minimization is biased towards observations that are available at the beginning of the 12-hour assimilation window, due to the reduction of precipitation sensitivities in time. However, using three-dimensional adjoint sensitivity computations, it is shown that this problem should not be as critical in four-dimensional variational assimilation. It is also demonstrated that the combination of precipitation data with information about the moisture field produces more realistic 2D-Var increments than with rain gauges solely. Finally, the implications for the future assimilation of cloud and precipitation affected observations in direct 4D-Var are presented.