Title: Observation Impact on Forecast Error in a Regional Model

Authors: Thomas Auligné (National Center for Atmospheric Research)
Hongli Wang (National Center for Atmospheric Research)
Xin Zhangs (National Center for Atmospheric Research)
Xiaoyan Zhang (National Center for Atmospheric Research)
Qingnong Xiao (National Center for Atmospheric Research)
Xiang-Yu Huang (National Center for Atmospheric Research)

A new capability to calculate the sensitivity of forecast error to observations, based on adjoint techniques, has been included into the Weather Research and Forecasting (WRF) model and data assimilation suite. This presentation will focus on an effort to understand some of the main sources of uncertainty in the calculation of observation impact. We believe that most of the methodology is applicable to all systems focusing on adjoint sensitivity and observation impact.

The reference for the calculation of the forecast error is often chosen as the system own analysis after 24 hours, assuming it is uncorrelated with the analysis at initial time. We will assess the sensitivity of the observation impact to the choice of reference. We will then compare results using several norms for the forecast error. In particular, we will introduce a new verification using future observations as reference and we will show its connection with another common metric called the Degree of Freedom for Signal (DFS). The ability of the tangent-linear (resp. adjoint) code to propagate real-size perturbations (resp. sensitivities) will be evaluated and we will study the amount of information lost through the lateral boundary conditions.

Finally we will show an attempt to predict the impact of observations on forecast error. This can be very useful for operational systems where some of the detrimental observations could be discarded a priori.

GMAO Head: Michele Rienecker Global Modeling and Assimilation Office NASA Goddard Space Flight Center

Curator: Nikki Privé Last Updated: March 1 2011