When radionuclides are released into the atmosphere, the assessment of the consequences depends on the evaluation of the magnitude and temporal evolution of the release, which can be highly variable as in the case of Fukushima Daiichi. Here, we propose Bayesian inverse modelling methods and the reversible-jump Markov chain Monte Carlo technique, which allows one to evaluate the temporal variability of the release and to integrate different types of information in the source reconstruction.

Iterative optimization techniques, the state of the art in data assimilation, have largely focused on extending forecast accuracy to moderate- to long-range forecast systems. However, current methodology may not be cost-effective in reducing forecast errors in online, short-range forecast systems. We propose a novel optimization of these techniques for online, short-range forecast cycles, simultaneously providing an improvement in forecast accuracy and a reduction in the computational cost.

The assessment of the environmental consequences of a radionuclide release depends on the estimation of its source. This paper aims to develop inverse Bayesian methods which combine transport models with measurements, in order to reconstruct the ensemble of possible sources. Three methods to quantify uncertainties based on the definition of probability distributions and the physical models are proposed and evaluated for the case of ¹⁰⁶Ru releases over Europe in 2017.

All scales of a dynamical physical process cannot be resolved accurately in a multiscale, geophysical model. The behavior of unresolved scales of motion are often parametrized by a random process to emulate their effects on the dynamically resolved variables, and this results in a random–dynamical model. We study how the choice of a numerical discretization of such a system affects the model forecast and estimation statistics, when the random–dynamical model is unbiased in its parametrization.

A small-size ensemble of mesoscale simulations has been filtered to characterize the spatial structures of transport errors in atmospheric CO₂ mixing ratios. The extracted error structures in in situ and column CO₂ show similar length scales compared to other meteorological variables, including seasonality, which could be used as proxies in regional inversion systems.

This paper describes an innovative way to use data assimilation to infer the dynamics of a physical system from its observation only. The method can operate with noisy and partial observation of the physical system. It acts as a deep learning technique specialised to dynamical models without the need for machine learning tools. The method is successfully tested on chaotic dynamical systems: the Lorenz-63, Lorenz-96, and Kuramoto–Sivashinski models and a two-scale Lorenz model.

We explore the possibility of combining data assimilation with machine learning. We introduce a new hybrid method for a two-fold scope: (i) emulating hidden, possibly chaotic, dynamics and (ii) predicting its future states. Numerical experiments have been carried out using the chaotic Lorenz 96 model, proving both the convergence of the hybrid method and its statistical skills including short-term forecasting and emulation of the long-term dynamics.

We demonstrate that transport model errors, one of the main contributors to the uncertainty in regional CO₂ inversions, can be represented by a small-size ensemble carefully calibrated with meteorological data. Our results also confirm transport model errors represent a significant fraction of the model–data mismatch in CO₂ mole fractions and hence in regional inverse CO₂ fluxes.

Data assimilation looks for an optimal way to learn from observations of a dynamical system to improve the quality of its predictions. The goal is to filter out the noise (both observation and model noise) to retrieve the true signal. Among all possible methods, particle filters are promising; the method is fast and elegant, and it allows for a Bayesian analysis. In this review paper, we discuss implementation techniques for (local) particle filters in high-dimensional systems.

Using the framework Lyapunov vectors, we analyze the asymptotic properties of ensemble based Kalman filters and how these are influenced by dynamical chaos, especially in the context of random model errors and small ensemble sizes. Particularly, we show a novel derivation of the evolution of forecast uncertainty for ensemble-based Kalman filters with weakly-nonlinear error growth, and discuss its impact for filter design in geophysical models.

The forecast of weather prediction uncertainty is a real challenge and is crucial for risk management. However, uncertainty prediction is beyond the capacity of supercomputers, and improvements of the technology may not solve this issue. A new uncertainty prediction method is introduced which takes advantage of fluid equations to predict simple quantities which approximate real uncertainty but at a low numerical cost. A proof of concept is shown by an academic model derived from fluid dynamics.

This study generalizes a paper by Pires et al. (1996) to state-of-the-art data assimilation techniques, such as the iterative ensemble Kalman smoother (IEnKS). We show that the longer the time window over which observations are assimilated, the better the accuracy of the IEnKS. Beyond a critical time length that we estimate, we show that this accuracy finally degrades. We show that the use of the quasi-static minimizations but generalized to the IEnKS yields a significantly improved accuracy.

The focus is on the development of low-order models of atmospheric transport and chemistry and their use for data assimilation purposes. A new low-order coupled chemistry meteorology model is developed. It consists of the Lorenz40-variable model used as a wind field coupled with a simple ozone photochemistry module. Advanced ensemble variational methods are applied to this model to obtain insights on the use of data assimilation with coupled models, in an offline mode or in an online mode.

The popular data assimilation technique known as the ensemble Kalman filter (EnKF) suffers from sampling errors due to the limited size of the ensemble. This deficiency is usually cured by inflating the sampled error covariances and by using localization. This paper further develops and discusses the finite-size EnKF, or EnKF-N, a variant of the EnKF that does not require inflation. It expands the use of the EnKF-N to a wider range of dynamical regimes.

Data assimilation is used in atmospheric chemistry models to improve air quality forecasts, construct re-analyses of concentrations, and perform inverse modeling. Coupled chemistry meteorology models (CCMM) are atmospheric chemistry models that simulate meteorological processes and chemical transformations jointly. We review here the current status of data assimilation in atmospheric chemistry models, with a particular focus on future prospects for data assimilation in CCMM.