Clouds Decoded: High Resolution Cloud Property Retrievals in Sentinel-2

Observational data for cloud processes are captured at a huge range of spatial resolutions, from particle processes resolved in micrometres, to mesoscale systems measured with horizontal sampling distances of kilometres. However, between these resolution ranges, at the scale of tens of metres, few observational constraints exist. Nevertheless, there is an increasing understanding that processes (e.g. phase heterogeneity) are occurring at scales which fall between the observable length-scales of current sensing paradigms, and can drastically alter cloud evolution and their resulting radiative effects.

Motivated by this relative lack of observational data, we have developed and deployed a suite of physical property retrieval tools for Sentinel-2 imagery, a 10 m/pixel multispectral satellite, as part of the Clouds Decoded project (funded by the UK’s Advanced Research and Invention Agency). In this presentation, we will provide a tour of the algorithms and techniques developed---and released open-source---for cloud type classification, cloud height, optical depth, ice/liquid phase, and particle effective radius. These involve a mix of radiative transfer modelling and inversion, computer vision techniques, and machine learning, and are validated against ground-based measurements from ACTRIS sites. In addition to describing the methods themselves, we will also provide an overview of the large, open dataset we have produced, which comes as both individual products and in a regridded, parameterised format, and which will also be made open to the community. We will highlight the potential uses of this data and hope to encourage the community to adopt it as a source of high-resolution information about clouds that can complement and enhance existing data sources and modelling efforts.