Adaptive observations during FASTEX: A systematic survey of upstream flights

Abstract The effectiveness of the recent concept of adaptive observations is tested on 20 cases for which Fronts and Atlantic Storm‐Track EXperiment (FASTEX) upstream dropsondes were available. In the first stage, the influence of the targeted data is tested without taking into account the real evolution of the atmosphere. On average, the inclusion of these targeted data significantly influences the forecasts (mean impact of about 9 hPa at 36 h). However, a large spread of impacts is observed and at 36 h (close to the mean verifying time) the mean impact is about 7.8 hPa for National Center for Atmospheric Research Learjet 36 (L36) flights and 10.8 hPa for National Oceanic and Atmospheric Administration Gulfstream IV (GIV) flights, with different behaviours. These tests also indicate that the targeted observations may indirectly affect the mediumrange forecast of the subsequent lows. the redundancy of targeted data with respect to conventional no‐FASTEX data (defined as the difference in impact between forecasts with and without conventional data), is about 40% for L36 flights and 14% for GIV flights, at 36 h. These differences between L36 and GIV flights (for the temporal characteristics of the impact signals and the redundancy) might be due either to the location of the data (L36 data were closer to the observationally dense North American continent), or to the difference between targeting techniques used to guide the planes. It seems also that adaptive observations are clearly constrained by the conventional observing network, and the sampling strategy used is also a crucial problem to solve. In the second stage, the skill of forecasts with targeted dropsondes is evaluated. the inclusion of these targeted data leads to a mixture of improvement and degradation cases, the mean improvement is very weak and has the same order of magnitude as the verifying‐analysis errors. However, a study of cases suggests that the adaptive‐observation effectiveness is high when the predictability is weak. Otherwise, the assimilation system seems unable to reduce the errors in the most unstable directions, and these deficiencies blur the impact of the targeted data. This suggests that the success of adaptive observations also depends on the assimilation system used.