There is currently much discussion within the meteorological community regarding the design and implementation of current and future observing systems, with particular emphasis on mesoscale observing systems. Assessing the relative value of observational platforms is useful for both scientific and budgetary interests. One way to evaluate the relative value in each observing system is to assess its contribution to the reduction of error in numerical weather prediction model forecasts. The Rapid Update Cycle (RUC), running operationally at the National Centers for Environmental Prediction (NCEP), is a good model to use for this evaluation because it assimilates a variety of asynoptic observations on an hourly basis.

Using the RUC on its operational continental United States (CONUS) domain, we have run a series of experiments where various data sources where systematically removed from the model in an attempt to measure the relative contribution of each in reducing forecast error. The data sources included in our tests were rawinsonde, automated aircraft (ACARS), profilers, surface (metars), and VAD winds. In addition, we have run the RUC with no data, other than that supplied indirectly through the lateral boundary conditions, as a “worse case” calibration.

In this paper we discuss the results of retrospective tests using the RUC model for the period 4-13 January 2001. This is the same period used by NCEP for testing the most recent operational versions of both the Eta and RUC models. Data were denied from the RUC over various portions of the RUC domain, including an area that contains the operational NOAA profiler network. For all the experiments, including a control run that contains all possible data, average verification statistics for RUC wind, temperature, height, and relative humidity forecasts against rawinsonde observations were compiled for the test period. In addition to the average errors, we examine the largest errors at individual rawinsonde locations to identify how the absence of each data type elates to the occurrence of large error events associated with active weather periods. Statistical tests were performed for the difference between each denial experiment and the control run in order to access the significance of the results.