Data Assimilation Seminar

Dr. Tse-Chun Chen (July 3, 2019, 11:00-)

Affiliation University of Maryland
Title How to improve forecasts by identifying and deleting detrimental observations
Abstract Ensemble Forecast Sensitivity to Observations (EFSO; Kalnay et al. 2012) estimates the short-term (e.g. 6-h) forecast error change introduced by each assimilated observation based on the gain matrix and the ensemble forecasts from ensemble Kalman filter, making it economical and efficient. EFSO identifies whether each observation is beneficial or detrimental to the following short-term forecast, and has two major applications: 1) Data Monitoring and Selection: EFSO provides efficient monitoring of the observational impact and reveals detrimental subsets of observations that allow designing a detailed data selection and QC for both existing and new instruments. 2) Proactive QC: This is a fully flow-dependent QC based on immediate EFSO impacts that avoids forecast skill dropouts (Hotta et al., 2017, Ota et al., 2013) by rejecting observations with very detrimental EFSO impact in each DA cycle, using the next 6-h analysis as verifying truth. PQC provides improvements to the analysis and subsequent forecast that accumulate with time, increasing over many cycles. An experiment with realistic GFS forecasts shows that cycled PQC improves the 1-day forecasts by ~10-20%, and that the improvement does not vanish until after 10 days. We found that the improvement from past PQCs accumulates with time and dominates over the immediate improvement.

Back to Seminar Series

Page Top