Adding targeted observations to the array of observations routinely available to numerical weather prediction is generally found to give small impact, although studies have often focused on a very limited number of cases in field experiments. It is expected that observation targeting in sensitive areas may have more impact on forecasts ‘busts’. A forecast ‘bust’ in a given verification region is defined as a doubling of the mean monthly error for the day‐two forecast. This does not occur very often in the operational ECMWF system. The primary purpose of this study is to produce three reference data‐assimilation experiments: the first assimilation experiment uses all available observations (as they are used in the operational ECMWF analysis system). The second assimilation experiment systematically excludes all observations in the North Pacific oceanic region. The third assimilation experiment systematically excludes all observations in the North Atlantic oceanic region. These three experiments are then used to evaluate the potential of targeting strategies which are described in two separate papers (parts II and III of this study).Results, based on a summer and a winter season, indicate that on average: Oceanic data are important to reduce day‐2 forecast errors over the downstream land areas, the downstream impact of observations over the Pacific Ocean being larger than the downstream impact of observations over the Atlantic Ocean. The removal of observations taken over the Pacific Ocean has generally a small impact on medium‐range forecasts verified over Europe, but results strongly depend on the data‐assimilation system used to assimilate the observations. Forecasts started from 3D‐Var analyses are worse than forecasts started from 4D‐Var analyses and, given the current observing system, the relative impact of ocean observations is higher if a 3D‐Var system is used instead of a 4D‐Var system. In other words, 4D‐Var is better capable of propagating information from data‐rich to data‐void areas, and is therefore less vulnerable to data gaps in a system which otherwise includes a large number of satellite observations. Copyright © 2007 Royal Meteorological Society

In this third and final paper of a series, we assess the performance of the three‐dimensional variational data assimilation scheme, in the light of the results from the extensive pre‐operational programme of numerical experimentation. Its performance is compared with that of the previous operational scheme at the European Centre for Medium‐Range Weather Forecasts, which was based on Optimal Interpolation. The main features of the new scheme are illustrated, in particular the effects of non‐separable structure functions and the improved data usage. TIROS‐N Operational Vertical Sounder cloud‐cleared radiances, for example, are used directly without a separate retrieval step. Scatterometer data are assimilated in the form of ambiguous winds with the ambiguity removal taking place within the analysis itself. Problems encountered during the tests are discussed and the solutions implemented are explained. The overall impact on forecast accuracy in the troposphere of the northern hemisphere extratropics is neutral for geopotential and positive for wind and temperature. The impact is neutral in the tropics, and significantly positive in the southern hemisphere. Analyses and forecasts for the stratosphere have improved in all regions. Other positive results include a clear improvement in analyses of near‐surface winds over oceans, particularly in the vicinity of tropical storms. This is predominantly because of the assimilation of scatterometer wind‐data.

This study examines atmospheric motion vectors derived by tracking features in image sequences from Meteosat for anomalies during the second satellite eclipse period of 2001. During eclipse periods (March–April; September–October), data from geostationary sensors are prone to anomalies caused by solar stray light entering the radiometer around local midnight. As a result, imagery from geostationary satellites can exhibit local anomalies for certain time slots, and these anomalies can change considerably from one time slot to the next. The effect of these anomalies on atmospheric motion vectors is characterized in this study by investigating the temporal consistency of the data and by monitoring the winds against short-term forecasts from the European Centre for Medium-Range Weather Forecasts (ECMWF) global model. Atmospheric motion vectors from the Meteosat water vapor channel exhibit spuriously fast winds for certain time slots around local midnight for extended periods around the satellite's eclipse. The anomalies are caused primarily through tracking problems, but height assignment is also affected. Less severe anomalies are found for infrared winds, in agreement with less severe image anomalies for the infrared channel. The European Organization for the Exploitation of Meteorological Satellites (EUMETSAT) and ECMWF quality control are capable of reducing the problem, but anomalous characteristics can still be found in the quality-controlled subsamples. As a result of these findings, atmospheric motion vectors from Meteosat are now excluded from the operational assimilation at ECMWF for certain time slots around the eclipse periods.