AI-based retrieval scheme for warm and cold precipitation for use with satellite data: global survey compared with climate simulation

MotivationPrecipitation is critical for climate and climate change. It determines the net latent heating influencing mesoscale and large-scale circulations.

It also affects the Earth’s radiation budget because its removal of water substance determines the atmospheric humidity and extent of clouds.There is uncertainty about the physical origins of precipitation in clouds.

Two mechanisms are known, involving the ice and liquid phases of clouds, causing ‘cold’ and ‘warm’ precipitation respectively.

Remarkably, there have been no rigorous studies of the ultimate microphysical origin of precipitation falling from clouds worldwide.

Ours in 2023 was the first study to track the components of microphysical species aloft from both mechanisms of precipitation in an atmospheric model. We simulated three storms.

Yet it is unclear if our results apply to the globe.Consequently, the proposed project seeks to create a neural network algorithm that can learn from our cloud model how to retrieve the occurrence of cold and warm precipitation from the data of an upcoming satellite.

This will have the advantage of blending observations and detailed modeling in a computationally cheap retrieval scheme that can provide real-time observations of this precipitation property.It is vital to fill this gap in knowledge about the physical origins of precipitation globally because any model, whether for weather or climate, is merely a reflection of the science.

Hypothetically, if our neural network were somehow to show that most surface precipitation globally is, say, ‘warm’, then model development would be motivated to focus on issues affecting coalescence like ascent variability, in-cloud activation, and turbulent enhancement of collisions.

Equally, more acuity of validation of models in the community will be possible with our proposed product, since it will be easy to code tagging tracers into any model for warm and cold precipitation.

Hence, our AI-based retrieval scheme will ensure weather and climate models predict precipitation for the right reasons, benefitting model development generally.

Planned research activities and objectivesThe proposed project aims to elucidate how, in the present-day climate, the balance between warm and cold surface precipitation depends on geographic location, season and cloud-type. This will be done by creating a new product for the upcoming EarthCARE satellite to be launched this month.

The approach involves development of a neural network that predicts the ratio of intensities of warm surface precipitation relative to the total (warm plus cold).

Inputs to the neural network will be products that are already planned by EarthCARE, such as cloud-top and -base temperatures, cloud-top phase, vertical profiles of rain and ice water contents, and the ambient aerosol conditions of chemistry and loading, especially regarding mineral dust and solute aerosols such as sulfate.The sequence of objectives is:- (1) construct a preliminary version of an AI-retrieval scheme that is a neural network to infer from satellite data (EarthCARE) the ‘warm’ and ‘cold’ fractions of surface precipitation, training it with our cloud simulations at three locations; (2) develop the capability for two global models, one conventional and the other advanced, to predict with tagging tracers the warm and cold components of precipitation; (3) validate both global models, especially for their predictions of precipitation occurrence at the three locations; (4) train the neural network (from (1)) using the advanced global model’s simulation (from (2) and (3)); and (5) chart the global distribution of the ‘warm’ fraction of surface precipitation from the AI-retrieval scheme for all EarthCARE satellite observations, analysing it and comparing with conventional global simulation.A deliverable will be a satellite retrieval scheme given to the EarthCARE scientific community, for observations of warm and cold precipitation globally in real-time.