Assimilation of Doppler Lidar-Derived Planetary Boundary Layer Heights Using the Ensemble Kalman Filter During the PECAN Field Campaign

The assimilation of Planetary Boundary Layer (PBL) height (PBLH) data into numerical weather prediction (NWP) models is still in its early stages of development. This study explores various strategies for integrating lidar-derived PBLH data into a convective-permitting configuration of the Weather Research and Forecasting (WRF) model, using the ensemble-based Data Assimilation Research Testbed (DART). Building on a previous study on PBLH data assimilation, we conducted a series of experiments to assess the impact of different parameters and approximations on DA performance. Specifically, these experiments examined: 1) The magnitude of initial perturbations required to generate a physically consistent initial ensemble; 2) The use of multi-physics to represent uncertainties in PBLH estimates and their associated sensitivities to the state vector; 3) Two variants of the ensemble Kalman filter; 4) Vertical error covariance localization; 5) Magnitude of observation errors; and 6) Error covariance inflation. We ran multiple WRF/DART model forecasts and analysis during the Plain Elevated Convection at Night (PECAN) campaign periods (June to mid-July 2015). Doppler lidar-retrieved PBLH from two PECAN sites (FP1, FP2), obtained using the σ²_w threshold and a composite fuzzy logic method, respectively, were assimilated, and the resulting analyses were compared with radiosonde measurements of potential temperature (θ), water vapor mixing ratio (q), wind (U, V) and Atmospheric Emitted Radiance Interferometer (AERIoe) measurements of θ and q at two sites.

Our findings highlight the strengths and limitations of PBLH data assimilation in NWP models as follows: a) The use of multi-physics better captures the sensitivities between state variables and PBLH; b) Initial inflation strongly influences the analysis increments of the state variables; c) The types of vertical localization significantly affect the location and magnitude of analysis increments for state variables; d) The data assimilation process tends to overcorrect, leading to underprediction or overprediction of wind speed near the top of the PBL. The most effective DA setup among those tested, based on fit to radiosonde atmospheric profiles, includes: i) Using only small perturbations of the outer domain and a 6-hour spinup period; ii) A boxcar-ramp vertical localization with a 1km cutoff; iii) Initial inflation set to 1.5; iv) Observation error set to 0.2*PBLH. This setup resulted in a decrease in errors for θ, q, and wind speed by 0.8 K, 0.5 g/kg, and 4.0 m/s, respectively, averaged over the vertical profile below 700mb, when compared with radiosonde measurements. The RMSE of PBLH decreased from 60% of the ensemble mean value to less than 20% after data assimilation which is within the bounds of the observation error. Our results highlight the importance of further investigating the optimal assimilation of PBLH data, particularly in updating the associated state vector, which can sometimes degrade the analysis of certain thermodynamic variables. We aim to report these findings to provide valuable insights for future studies on PBLH data assimilation.