REconstructing Cloud FIelds in 4D (RECFI-4D)

While rain is the most spectacular end to the lifecycle of a cloud, only a small fraction of clouds rain. Most clouds mix with dry air from their surroundings and evaporate (a process called entrainment). These mixing processes determine the amount and height of clouds, as well as the distribution of water clouds leave behind in the atmosphere when they evaporate. These cloud properties are critical for determining global temperatures, as clouds are a key control on the Earth's energy balance.

Adding small atmospheric particles (known as aerosols) to the atmosphere through the burning of fossil fuels has the potential to modify entrainment, enhancing severe storms increasing weather impacts in polluted regions. Aerosol increases may also increase entrainment for cumulus clouds, causing them to evaporate faster thus reducing the amount of low cloud, warming the climate.

With such strong controls on cloud properties, it is no surprise that entrainment (and detrainment, the corresponding mixing of air out of a cloud) have a large impact on global climate models.Given their important role in setting cloud properties and how they respond to human activity, it is very unfortunate that entrainment and mixing processes cannot be easily measured with current observations. This is because entrainment operates at a 'cloud scale' - to properly measure it over a field of clouds, we need to measure processes that happen quickly and at a small scale. Current observing systems are not suited to this problem, typically observing at a high temporal or spatial resolution (but not both). This leaves an 'observation gap', of high resolution, frequent measurements that are able to characterise entire cloud lifecycles across a field of clouds. RECFI-4D will provide new observations that fill this gap, enabling the measurement of these critical atmospheric processes.

RECFI-4D will reconstruct the cloud field over London using an array of cameras placed around the city. In the same way a self-driving car can learn about its environment by looking out, we apply similar algorithms to build a 4D model of cloud fields in near real time. However, a self-driving car usually looks at objects with well defined surfaces (cars, phoneboxes, people). In contrast, clouds have poorly defined edges and highly variable internal structures. In RECFI-4D, we apply a novel new volumetric reconstruction algorithm, which is also able to provide information about the cloud interior structure (like a CT scanner). With this new, state-of-the-art cloud imaging technique, we will characterise cloud field development and measure previously almost inaccessible properties and processes across a cloud field.

Being able to image a cloud field opens up a whole range of new applications that were previously inaccessible due to the ``observations gap''. The machine-learning revolution has lead to proposals to train global climate model cloud parametrisations on high resolution models. Constraining these high resolution models requires high resolution observations of cloud processes, such as those provided in RECFI-4D. Even better, why not use the cloud observations directly to train the parametrisation? In-situ observations as accurate, but lack the context for their observations, the techniques developed in RECFI-4D provide that context for the observation within the cloud lifecycle. Imaging cloud fields in 4D provides a real-time, cost effective system for the detection of severe weather, such as downbursts, of interest to aviation. Improved observations of cloud fields also enable better computer rendering of realistic clouds, important for both the film and videogame industry.

With a novel application of state of the art computer vision algorithms to a problem at the heart of cloud and climate physics, RECFI-4D explores a new frontier in atmospheric science.