Quantification of individual observation influence in a campaign re-analysis using Partial Analysis Increments

Assimilating additional observations into an existing data assimilation system influences the analysis states, that serve as initial conditions for subsequent numerical weather predictions (NWP). Consequently, differences in the analysis can propagate through the forecast and impact forecast skill. The influence of individual observations on the analysis state is commonly assessed through single-observation experiments, which are computationally expensive. As an alternative, the so-called Partial Analysis Increments (PAI) diagnostic (Diefenbach et al., 2023) enables an efficient approximation of the contributions of individual observation to the resulting analysis, without requiring dedicated single-observation experiments. Instead, PAIs can be estimated for all assimilated observations using standard output obtained from the data assimilation system used by Deutscher Wetterdienst, which is based on the Local Ensemble Transform Kalman Filter (LETKF). The calculation of PAIs requires, among others, the full analysis ensemble in model and observation space as well as the first-guess departures.

We apply the PAI diagnostic to a campaign re-analysis dataset that incorporates a wide range of non-operational field campaign observations which have been assimilated in addition to observations from the operational measurement network. Specifically, the campaign observations include a network of Doppler wind lidars (DWLs) deployed across southwestern Germany during the ‘Swabian MOSES 2023’ field campaign. Based on this re-analysis, we quantify the contribution of individual observations to the total analysis increments and examine the horizontal and vertical footprints of the respective PAIs, with particular emphasis on assimilated vertical profiles of the horizontal wind retrieved from the DWLs. Moreover, we compare the characteristics of assimilating DWL measurements to that of other observation types providing direct information about the wind field, such as radar radial velocities, radiosonde profiles, and airborne in-situ measurements. Our preliminary results indicate that DWLs contribute substantially to the total analysis increments, and that their spatial influence pattern is distinct from those of radar and airborne observations.

 

Diefenbach, T., Craig, G., Keil, C., Scheck, L. and Weissmann, M. (2023): Partial analysis increments as diagnostic for LETKF data assimilation systems. Q. J. R. Meteorol. Soc., 149, 740-756, https://doi.org/10.1002/qj.4419