84th AMS Annual Meeting

Tuesday, 13 January 2004: 2:30 PM
Application of the Error Subspace Statistical Estimation (ESSE) system to real-time error forecasting, data assimilation and adaptive sampling off the Central California Coast during AOSN-II
Room 6A
Pierre F. J. Lermusiaux, Harvard University, Cambridge, MA; and W. G. Leslie, C. Evangelinos, P. J. Haley, O. Logoutov, P. Moreno, A. R. Robinson, G. Cossarini, X. S. Liang, and S. J. Majumdar
During the August-September 2003 AOSN-II experiment, the Error Subspace Statistical Estimation (ESSE) system was utilized in real-time to forecast physical fields and uncertainties, assimilate a varied set of data streams (ships, gliders, aircraft), and provide suggestions for adaptive sampling and guide dynamical investigations. ESSE is a physical-biogeochemical-acoustical estimation scheme which aims to capture, forecast and reduce the dominant uncertainties, i.e the error subspace. It is currently based on a singular value decomposition of the minimum error variance update and on an adaptive ensemble scheme for the forecast of the largest errors. Each ensemble member was a sample path of the stochastic primitive equation model of the Harvard Ocean Prediction System (HOPS). The initial conditions were set in accord with posterior error estimates. The stochastic forcings represented various model errors. The data assimilation and adaptive sampling schemes were consistent.

Over the period 4 August to 3 September, 10 sets of ESSE error forecasts, data assimilation products, adaptive sampling recommendations were issued in real-time, using a total of 4323 ensemble members. Operational products and dynamical interpretations were posted on the web. Scientific and technical results will be presented, including the: dynamical findings in Monterey Bay and the California Current System; convergence of the ensemble; ensemble mean and most probable forecast (mpf); forecast skill estimates; variances, covariances and singular vectors of the ensemble; data assimilation and the modeling of data and model errors; adaptive sampling, either subjective based on error forecasts or quantitative based on nonlinear field and data forecasts.

Supplementary URL: