Applications of the LETKF to adaptive observations, analysis sensitivity, observation impact and the assimilation of moisture
Publication or External Link
In this thesis we explore four new applications of the Local Ensemble Transform Kalman Filter (LETKF), namely adaptive observations, analysis sensitivity, observation impact, and multivariate humidity assimilation. In each of these applications we have obtained promising results.
In the adaptive observation studies, we found that ensemble spread strategy, where adaptive observations are selected among the points with largest ensemble spread (with the constraint that observations cannot be contiguous in order to avoid clusters of adaptive observations) is very effective and close to optimal sampling. The application on simulated Doppler Wind Lidar (DWL) adaptive observation studies shows that 3D-Var is as effective as LETKF with 10% adaptive observations sampled with the ensemble spread strategy. With 2% adaptive observations, 3D-Var is not as effective as the LETKF.
In the analysis sensitivity study, we proposed to calculate this quantity within the LETKF with low additional computational time. Unlike in 4D-Var (Cardinali et al., 2004), in the LETKF, the computation is exact and satisfies the theoretical value limits (between 0 and 1). The results from simulated experiments show that the trace of analysis sensitivity qualitatively reflects the observation impact obtained from independently computed data addition or data denial OSSE experiments.
In the observation impact study, we derived a formula to estimate the impact of observations on short-range forecasts as in Langland and Baker (2004), but without using an adjoint model. Both methods estimate more than 90% accuracy the actual observation impact on the short-range forecast error improvement. Like the adjoint method, the method we proposed detects observations that have either large random error or unaccounted bias. This method can be easily calculated within the LETKF, and provides a powerful tool to estimate the quality of observations.
Finally, for the first time, we assimilate humidity observations multivariately in both perfect model experiments and real data assimilation. We found that multivariate assimilation is better than univariate assimilation. The assimilation of pseudo-RH (Dee and da Silva, 2003) is better than the choice of specific humidity and relative humidity. The multivariate assimilation of AIRS specific humidity retrievals on NCEP GFS system shows positive impact on the winds analysis.