Conditioned verification region selection in adaptive sampling

Adaptive or targeted observations supplement routine observations at a pre-specified targeting time. Adaptive observation locations are selected in an attempt to minimize the forecast error variance of a future target forecast within some predefined verification region at some predefined verification time. Ideally, the verification region is placed in a location where unusually large forecast errors or busts are likely. Here, we compare two methods of selecting verification regions. The unconditioned method is based on the raw verification time ensemble spread while the conditioned method is based on an Ensemble Transform Kalman Filter (ETKF) estimate of forecast error variance given the routine observations to be taken at the targeting time. To test the effectiveness of the two approaches, a series of Observation System Simulation Experiments (OSSEs) on a chaotic barotropic flow using an imperfect model were performed. When a hybrid error covariance model that linearly combined an isotropic and ensemble based error covariance model was used to assimilate routine observations then (a) conditioned verification regions captured a greater frequency of forecast busts than unconditioned verification regions; and (b) the assimilation of ETKF selected targeted observations reduced the frequency of forecast busts more with conditioned verification regions than with unconditioned verification regions. However, when an isotropic error covariance model was used to assimilate the observations then (a) conditioned verification regions again captured a greater frequency of forecast busts than unconditioned verification regions when only routine observations were assimilated; but converse to the hybrid case (b) the assimilation of targeted observations reduced the frequency of forecast busts less with conditioned verification regions than with unconditioned verification regions.

The sensitivity to the type of data assimilation scheme used on the ETKF's ability to distinguish occasions where the forecast change due to targeted observations is likely to be large from occasions where the change is likely to be small was also examined. It was found that the ETKF's ability to do this was greater when observations were assimilated using hybrid covariances than with isotropic covariances. To a lesser degree, the ETKF's ability to distinguish occasions where the forecast improvement due to targeted observations was large from occasions when it was small was also greater with hybrid covariances than with isotropic covariances. Forecast and analysis errors of target forecasts were substantially smaller with hybrid covariances than with isotropic covariances.