This thesis presents an investigation of ensemble Gaussianity, the effect of non- Gaussianity on covariance structures, storm-centered data assimilation techniques, and the relationship between commonly used data assimilation variables and the underlying dynamics for the case of Hurricane Humberto. Using an Ensemble Kalman Filter (EnKF), a comparison of data assimilation results in Storm-centered and Eulerian coordinate systems is made. In addition, the extent of the non-Gaussianity of the model ensemble is investigated and quantified. The effect of this non-Gaussianity on covariance structures, which play an integral role in the EnKF data assimilation scheme, is then explored. Finally, the correlation structures calculated from a Weather Research Forecast (WRF) ensemble forecast of several state variables are investigated in order to better understand the dynamics of this rapidly intensifying cyclone. Hurricane Humberto rapidly intensified in the northwestern Gulf of Mexico from a tropical disturbance to a strong category one hurricane with 90 mph winds in 24 hours. Numerical models did not capture the intensification of Humberto well. This could be due in large part to initial condition error, which can be addressed by data assimilation schemes. Because the EnKF scheme is a linear theory developed on the assumption of the normality of the ensemble distribution, non-Gaussianity in the ensemble distribution used could affect the EnKF update. It is shown that multiple state variables do indeed show significant non-Gaussianity through an inspection of statistical moments. In addition, storm-centered data assimilation schemes present an alternative to traditional Eulerian schemes by emphasizing the centrality of the cyclone to the assimilation window. This allows for an update that is most effective in the vicinity of the storm center, which is of most concern in mesoscale events such as Humberto. Finally, the effect of non-Gaussian distributions on covariance structures is examined through data transformations of normal distributions. Various standard transformations of two Gaussian distributions are made. Skewness, kurtosis, and correlation between the two distributions are taken before and after the transformations. It can be seen that there is a relationship between a change in skewness and kurtosis and the correlation between the distributions. These effects are then taken into consideration as the dynamics contributing to the rapid intensification of Humberto are explored through correlation structures.

The overall goals of this dissertation are to advance the field of chemical data assimilation, and to develop efficient computational tools that allow the atmospheric science community benefit from state of the art assimilation methodologies. Data assimilation is the procedure to combine data from observations with model predictions to obtain a more accurate representation of the state of the atmosphere. As models become more complex, determining the relationships between pollutants and their sources and sinks becomes computationally more challenging. The construction of an adjoint model ( capable of efficiently computing sensitivities of a few model outputs with respect to many input parameters ) is a difficult, labor intensive, and error prone task. This work develops adjoint systems for two of the most widely used chemical transport models: Harvard's GEOS-Chem global model and for Environmental Protection Agency's regional CMAQ regional air quality model. Both GEOS-Chem and CMAQ adjoint models are now used by the atmospheric science community to perform sensitivity analysis and data assimilation studies. Despite the continuous increase in capabilities, models remain imperfect and models alone cannot provide accurate long term forecasts. Observations of the atmospheric composition are now routinely taken from sondes, ground stations, aircraft, and satellites, etc. This work develops three and four dimensional variational data assimilation capabilities for GEOS-Chem and CMAQ which allow to estimate chemical states that best fit the observed reality. Most data assimilation systems to date use diagonal approximations of the background covariance matrix which ignore error correlations and may lead to inaccurate estimates. This dissertation develops computationally efficient representations of covariance matrices that allow to capture spatial error correlations in data assimilation. Not all observations used in data assimilation are of equal importance. Erroneous and redundant observations not only affect the quality of an estimate but also add unnecessary computational expense to the assimilation system. This work proposes techniques to quantify the information content of observations used in assimilation; information-theoretic metrics are used. The four dimensional variational approach to data assimilation provides accurate estimates but requires an adjoint construction, and uses considerable computational resources. This work studies versions of the four dimensional variational methods (Quasi 4D-Var) that use approximate gradients and are less expensive to develop and run. Variational and Kalman filter approaches are both used in data assimilation, but their relative merits and disadvantages in the context of chemical data assimilation have not been assessed. This work provides a careful comparison on a chemical assimilation problem with real data sets. The assimilation experiments performed here demonstrate for the first time the benefit of using satellite data to improve estimates of tropospheric ozone. / Ph. D.

Les dernières décennies ont vu croître en quantité et en qualité les données satellites. Au fil des ans, ces observations ont pris de plus en plus d'importance en prévision numérique du temps. Ces données sont aujourd'hui cruciales afin de déterminer de manière optimale l'état du système étudié, et ce, notamment car elles fournissent des informations denses et de qualité dansdes zones peu observées par les moyens conventionnels. Cependant, le potentiel de ces séquences d'images est encore largement sous–exploitée en assimilation de données : ces dernières sont sévèrement sous–échantillonnées, et ce, en partie afin de ne pas avoir à tenir compte des corrélations d'erreurs d'observation.Dans ce manuscrit nous abordons le problème d'extraction, à partir de séquences d'images satellites, d'information sur la dynamique du système durant le processus d'assimilation variationnelle de données. Cette étude est menée dans un cadre idéalisé afin de déterminer l'impact d'un bruit d'observations et/ou d'occultations sur l'analyse effectuée.Lorsque le bruit est corrélé en espace, tenir compte des corrélations en analysant les images au niveau du pixel n'est pas chose aisée : il est nécessaire d'inverser la matrice de covariance d'erreur d'observation (qui se révèle être une matrice de grande taille) ou de faire des approximationsaisément inversibles de cette dernière. En changeant d'espace d'analyse, la prise en compte d'une partie des corrélations peut être rendue plus aisée. Dans ces travaux, nous proposons d'effectuer cette analyse dans des bases d'ondelettes ou des trames de curvelettes. En effet, un bruit corréléen espace n'impacte pas de la même manière les différents éléments composants ces familles. En travaillant dans ces espaces, il est alors plus aisé de tenir compte d'une partie des corrélations présentes au sein du champ d'erreur. La pertinence de l'approche proposée est présentée sur différents cas tests.Lorsque les données sont partiellement occultées, il est cependant nécessaire de savoir comment adapter la représentation des corrélations. Ceci n'est pas chose aisée : travailler avec un espace d'observation changeant au cours du temps rend difficile l'utilisation d'approximations aisément inversibles de la matrice de covariance d'erreur d'observation. Dans ces travaux uneméthode permettant d'adapter, à moindre coût, la représentations des corrélations (dans des bases d'ondelettes) aux données présentes dans chaque image est proposée. L'intérêt de cette approche est présenté dans un cas idéalisé. / Recent decades have seen an increase in quantity and quality of satellite observations . Over the years , those observations has become increasingly important in numerical weather forecasting. Nowadays, these datas are crucial in order to determine optimally the state of the studied system. In particular, satellites can provide dense observations in areas poorly observed by conventionnal networks. However, the potential of such observations is clearly under--used in data assimilation : in order to avoid the management of observation errors, thinning methods are employed in association to variance inflation.In this thesis, we adress the problem of extracting information on the system dynamic from satellites images data during the variationnal assimilation process. This study is carried out in an academic context in order to quantify the influence of observation noise and of clouds on the performed analysis.When the noise is spatially correlated, it is hard to take into account such correlations by working in the pixel space. Indeed, it is necessary to invert the observation error covariance matrix (which turns out to be very huge) or make an approximation easily invertible of such a matrix. Analysing the information in an other space can make the job easier. In this manuscript, we propose to perform the analysis step in a wavelet basis or a curvelet frame. Indeed, in those structured spaces, a correlated noise does not affect in the same way the differents structures. It is then easier to take into account part of errors correlations : a suitable approximation of the covariance matrix is made by considering only how each kind of element is affected by a correlated noise. The benefit of this approach is demonstrated on different academic tests cases.However, when some data are missing one has to address the problem of adapting the way correlations are taken into account. This work is not an easy one : working in a different observation space for each image makes the use of easily invertible approximate covariance matrix very tricky. In this work a way to adapt the diagonal hypothesis of the covariance matrix in a wavelet basis, in order to take into account that images are partially hidden, is proposed. The interest of such an approach is presented in an idealised case.

MODEL OF ERROR COVARIANCES FOR THE ASSIMILATION OF RADAR REFLECTIVITY INTO NWP MODEL Predicting events with a severe convection is not easy due to the small spatial scale and rapid development of this phenomenon. But being able to predict such events is important in view of the dangerous phenomena that accompany these events, such as flash floods, strong winds, hailstorms or atmospheric electricity. Improved forecast can be achieved by more precisely defined initial conditions that enter the model. These data must match the scale of the studied phenomenon. Therefore, radar data is used in this case. Although the NWP model should describe real processes due to the simplifications and approximations the model's behavior does not entirely correspond the reality. Therefore, if we want the model to generate precipitation, we must ensure that the values of the model variables and their relationship are such that the process is started. To find out these relationships, we want to use a covariant model. In this paper, we focused on the correlation analysis of the model variables in the regions of convection between radar reflection, its conversion to the intensity of precipitation and other model variables. The COSMO data with a horizontal resolution of 2.8 km were used, which were describing approximately...

Ensemble-based methods have gained widespread popularity in the field of data assimilation. An ensemble of model realizations encapsulates information about the error correlations driven by the physics and the dynamics of the numerical model. This information can be used to obtain improved estimates of the state of non-linear dynamical systems such as the atmosphere and/or the ocean. This work develops efficient ensemble-based methods for data assimilation. A major bottleneck in ensemble Kalman filter (EnKF) implementations is the solution of a linear system at each analysis step. To alleviate it an EnKF implementation based on an iterative Sherman Morrison formula is proposed. The rank deficiency of the ensemble covariance matrix is exploited in order to efficiently compute the analysis increments during the assimilation process. The computational effort of the proposed method is comparable to those of the best EnKF implementations found in the current literature. The stability analysis of the new algorithm is theoretically proven based on the positiveness of the data error covariance matrix. In order to improve the background error covariance matrices in ensemble-based data assimilation we explore the use of shrinkage covariance matrix estimators from ensembles. The resulting filter has attractive features in terms of both memory usage and computational complexity. Numerical results show that it performs better that traditional EnKF formulations. In geophysical applications the correlations between errors corresponding to distant model components decreases rapidly with the distance. We propose a new and efficient implementation of the EnKF based on a modified Cholesky decomposition for inverse covariance matrix estimation. This approach exploits the conditional independence of background errors between distant model components with regard to a predefined radius of influence. Consequently, sparse estimators of the inverse background error covariance matrix can be obtained. This implies huge memory savings during the assimilation process under realistic weather forecast scenarios. Rigorous error bounds for the resulting estimator in the context of data assimilation are theoretically proved. The conclusion is that the resulting estimator converges to the true inverse background error covariance matrix when the ensemble size is of the order of the logarithm of the number of model components. We explore high-performance implementations of the proposed EnKF algorithms. When the observational operator can be locally approximated for different regions of the domain, efficient parallel implementations of the EnKF formulations presented in this dissertation can be obtained. The parallel computation of the analysis increments is performed making use of domain decomposition. Local analysis increments are computed on (possibly) different processors. Once all local analysis increments have been computed they are mapped back onto the global domain to recover the global analysis. Tests performed with an atmospheric general circulation model at a T-63 resolution, and varying the number of processors from 96 to 2,048, reveal that the assimilation time can be decreased multiple fold for all the proposed EnKF formulations.Ensemble-based methods can be used to reformulate strong constraint four dimensional variational data assimilation such as to avoid the construction of adjoint models, which can be complicated for operational models. We propose a trust region approach based on ensembles in which the analysis increments are computed onto the space of an ensemble of snapshots. The quality of the resulting increments in the ensemble space is compared against the gains in the full space. Decisions on whether accept or reject solutions rely on trust region updating formulas. Results based on a atmospheric general circulation model with a T-42 resolution reveal that this methodology can improve the analysis accuracy. / Ph. D.