Premium
Forecast skill of the ECMWF model using targeted observations during FASTEX
Author(s) -
Montani A.,
Thorpe A. J.,
Buizza R.,
Undén P.
Publication year - 1999
Publication title -
quarterly journal of the royal meteorological society
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.744
H-Index - 143
eISSN - 1477-870X
pISSN - 0035-9009
DOI - 10.1002/qj.49712556106
Subject(s) - cyclogenesis , climatology , forecast skill , range (aeronautics) , meteorology , dropsonde , environmental science , data set , forecast error , storm , forecast verification , statistics , computer science , mathematics , cyclone (programming language) , econometrics , tropical cyclone , geography , geology , materials science , field programmable gate array , computer hardware , composite material
The impact of targeted observations on forecast accuracy is investigated for five case‐studies of cyclogenesis during the Fronts and Atlantic Storm‐Track EXperiment (FASTEX). Calculations of localized singular vectors (SVs) have been made using the European Centre for Medium‐Range Weather Forecasts (ECMWF) forecasting model to find the sensitive regions for each case‐study. Two sets of analyses have been prepared: a ‘control’ set consisting of analyses in which all the FASTEX data have been removed from the operational analysis, and a ‘perturbed’ set obtained by adding to the control the data from the dropsondes deployed in regions highlighted as sensitive by SVs. Analysis and forecast differences have been compared with the structure of the SVs as they evolve during the forecast. Model integrations, starting from the two different analyses, have been performed. the impact of the dropsonde data on the forecast skill has been assessed on the basis of both subjective and objective scores. Re‐runs of the model from perturbed analyses almost always give better objective scores than those starting from the control analyses. the impact is more evident if the verification area coincides with the regions over which the SVs have been optimized. In this case, short‐range prediction errors (up to day 2) are reduced on average by 15%, with a maximum error reduction of about 37%.