The Estimation of Analysis Error Characteristics Using an Observation System Simulation Experiment

University Of Maryland Baltimore County, Baltimore MD

Investigators

Abstract

Before atmospheric observations can be applied as initial conditions for numerical weather forecast, they must be analyzed. Modern analyses are carried out using data assimilation systems. This raises the question of what errors are introduced by the analysis. This question cannot be answered in an operational setting, for which the only data available for checking the analysis are those from which it was prepared. In this project the error statistics of the National Center for Environmental Prediction's operational assimilation system will be assessed by applying it to the output of a numerical model, for which, unlike the observed atmosphere, the true state is perfectly known. The "nature run" of the model, which will provide the "data" to be assimilated is an already extant 13-month simulation of the European Centre for Medium Range Weather Forecasting (ECMWF) high resolution forecast model. The observing system simulation experiment (OSSE) will assimilate simulated observations of the nature run. The results of the OSSE will be validated, interpolated to the same grid as the nature run, and the statistics of analysis errors will be computed. These include basic statistics, such as biases and error variances, and more complex statistics that will reveal the spatial structures of analysis errors and their dependence on the atmospheric flow. The broader impacts of this project are in providing to the community estimates of uncertainty in analyzed data sets that are widely used in operational weather forecasting and for research.