This paper presents a new way to describe the Arctic sea-ice changes based on the shape of the observed seasonal cycles and using machine learning techniques. We show that the East Siberian and Laptev seas have lost their typical permanent sea-ice seasonal cycle while the Kara and Chukchi seas are experiencing a new typical seasonal cycle, corresponding to a partial winter-freezing.

Old observations are necessary to understand the atmosphere. When direct observations are not available, one can use indirect observations, such as tide gauges, which measure the sea level in port cities. The sea level rises when local air pressure decreases and when wind pushes water towards the coast. Several centuries-long tide gauge records are available. We show that these can be complementary to direct pressure observations for studying storms and anticyclones in the 19th century.

The goal of this paper is to weight several dynamic models in order to improve the representativeness of a system. It is illustrated using a set of versions of an idealized model describing the Atlantic Meridional Overturning Circulation. The low-cost method is based on data-driven forecasts. It enables model performance to be evaluated on their dynamics. Taking into account both model performance and codependency, the derived weights outperform benchmarks in reconstructing a model distribution.

A novel data-driven method is proposed to design an optimal sensor network for the reconstruction of offshore wind resources. Based on unsupervised learning of numerical weather prediction wind data, it provides a simple yet efficient method for the siting of sensors, outperforming state-of-the-art methods for this application. It is applied in the main French offshore wind energy development areas to provide guidelines for the deployment of floating lidars for wind resource assessment.

We present a new approach to validate numerical simulations of the current climate. The method can take advantage of existing climate simulations produced by different centers combining an analog forecasting approach with data assimilation to quantify how well a particular model reproduces a sequence of observed values. The method can be applied with different observations types and is implemented locally in space and time significantly reducing the associated computational cost.