Understanding spatial organization during precipitation-induced convective cloud transitions

Abstract

Parameterizations of boundary layer clouds and convection remain one of the largest causes for uncertainty in large-scale models for numerical weather and climate prediction. Recent research has highlighted the important role played by the spatial organization of convective cloud fields in Earth’s climate system, suggesting that the onset of precipitation could play a key role. How such organization affects the transition from shallow to deep convection is an unanswered research question.

The aim of this proposal is to find a better understanding of the role of precipitation in the spatial organization of boundary layer clouds over land, in particular in the transition from shallow to deeper convection. The proposed research aims to prove or disprove the following main hypotheses:

1. The transition between boundary layer convection to deeper convection over land is characterized by distinct patterns in the spatial structure of cumulus cloud populations.
2. Cloud size distribution analyses and pattern recognition algorithms from computer vision are effective methods for detecting and classifying such spatial patterns.
3. The time evolution of these spatial patterns is controlled by the way precipitation spatially interacts with the turbulent dynamics, radiation, and impacts of surface heterogeneity.
4. Using machine learning, population-dynamical models can be trained against spatially-aware datasets to reproduce key organizational metrics and memory effects of convective regime transitions.

A comprehensive mix of the observations from the United States Department of Energy’s Atmospheric Radiation Measurement (ARM) facility, fine-scale modeling, and conceptual population modeling is used to better characterize, understand, and model the organization of convection. The prime focus is the feedback loop between precipitation and spatial organization. However, new insights will be interpreted in the context of other key processes such as radiation and land surface heterogeneity. We use several spatially aware observational datasets, including scanning radar data, hemispheric camera data, and satellite data. Three ARM observatories are considered, including at the Southern Great Plains site, the GO-AMAZON campaign in Brazil, and the CACTI campaign in Argentina. As a result, many transition cases are covered under a broad range of meteorological conditions. To achieve our research goals a variety of tools is adopted, combining machine learning algorithms with more traditional statistics expressing spatial structure. This includes size distributions of cloud area, rain and neighbor spacing, as well as various metrics expressing the degree of organization.