We explore the potential of Multilevel Monte Carlo methods to improve ensemble-variational data assimilation without increasing the computational cost. By shifting part of the ensemble generation cost to coarser simulation grids, larger sample sizes and smaller sampling errors become affordable, while keeping the final estimate unbiased. Numerical experiments with a quasi-geostrophic model demonstrate the potential of the approach and highlight the challenges towards operational implementation.

This work investigates the potential value of using fine-scale meteorological ensembles to represent the inherent meteorological uncertainties in atmospheric dispersion model outputs. Probabilistic scores were used to evaluate the probabilistic performance of dispersion ensembles, using an original dataset of new continuous ⁸⁵Kr air concentration measurements and a well-known source term. The results show that the ensemble dispersion simulations perform better than deterministic ones.

The effect of deep convection representation on the jet stream above the cold front of an extratropical cyclone is investigated in the global numerical weather prediction model ARPEGE. Two simulations using different deep convection schemes are compared with (re)analysis datasets and NAWDEX airborne observations. A deeper jet stream is observed with the less active scheme. The diabatic origin of this difference is interpreted by backward Lagrangian trajectories and potential vorticity budgets.

Inacurracies in representing processes occurring at spatial scales smaller than the grid scales of the weather forecast models are important sources of forecast errors. This is the case of deep convection representation in models with 10 km grid spacing. We performed simulations of a real extratropical cyclone using a model with different representations of deep convection. These forecasts lead to different behaviors in the ascending air masses of the cyclone and the jet stream aloft.

This contributes to research on uncertainty prediction, which is important either for determining the weather today or estimating the risk in prediction. The problem is that uncertainty prediction is numerically very expensive. An alternative has been proposed wherein uncertainty is presented in a simplified form with only the dynamics of certain parameters required. This tool allows for the determination of the symbolic equations of these parameter dynamics and their numerical computation.