Events / Media
Back to Seminar Series

Data Assimilation Seminar Series

Dr. Daisuke Hotta (Sep. 2, 2015, 15:30-)

Affiliation JMA
Title Diagnostic methods for ensemble data assimilation
Abstract Authors: Daisuke Hotta, Yoichiro Ota (JMA)

Abstract: Advances in adjoint sensitivity techniques and application of information theory to variational data assimilation (DA) methods have given birth to a plethora of powerful diagnostic tools, including Degrees of Freedom for Signal (DFS) that measures how much information was extracted from observations by the analysis, Forecast Sensitivity to Observations (FSO) that quantifies how each assimilated observation has improved or degraded forecast and the forecast sensitivity to background- or observation- error covariances that estimates how much forecast errors would be reduced or increased by small perturbations to each element of the prescribed error covariance matrices. These diagnostics have recently been shown to be applicable also to the ensemble Kalman filters, either stand-alone or within hybrid systems. In this talk, I will present our recent work related these diagnostic techniques applied to quasi-operational global NWP systems. In the first part, we apply the ensemble-based FSO, or EFSO, to the National Centers for Environmental Prediction (NCEP)'s global hybrid DA system. We then show that the forecast can be dramatically improved by a sophisticated quality control (QC) procedure in which "flawed" observations that significantly degraded the forecast are detected by EFSO diagnostics with a short lead time and then the analysis and forecast are repeated without assimilating them. We call this new, fully flow-dependent QC procedure "Proactive QC". We also show that it is possible to formulate an ensemble version of the forecast sensitivity to observation error covariance matrix, which we call EFSR. We then show, by applying EFSR to the NCEP's global hybrid DA system, that tuning the observation error covariance matrix based on EFSR improves the EFSO impacts from the tuned observation types, indicating that the utility of observations can be improved by EFSR-based tuning. In the second part, we will present our preliminary results on EFSO, EFSR and DFS applied to Japan Meteorological Agency (JMA)'s global DA system. We found, through an examination of DFS and by partitioning forecast errors into the column- and null-spaces of the ensemble forecast perturbations, that in a situation where the member size is much smaller than the number of the observations that are locally assimilated, the member size critically restricts the amount of information that can be extracted from observations, potentially leading to suboptimality in the DA system. We will discuss the implication of this fact on how to design effective assimilation of dense observations such as satellite hyperspectral sounding and atmospheric motion vectors.
PDF Link to PDF file
Page Top