Marine Boundary Layer Cloud Mesoscale Organization

Download or read book Marine Boundary Layer Cloud Mesoscale Organization written by Johannes Karel Christiaan Mohrmann and published by . This book was released on 2020 with total page 104 pages. Available in PDF, EPUB and Kindle. Book excerpt: Marine low clouds are an important feature of the climate system, cooling the planet due to their high albedo and warm temperatures. They display a variety of different mesoscale organizations, which are tied to the varying environmental conditions in which they occur. This dissertation explores the drivers of marine low cloud variability using an observational perspective that draws upon aircraft, satellite, and reanalysis data, and uses the application of a number of machine learning techniques. The focus is largely though not exclusively on mesoscale organization. In the first part of this work, data from the Cloud System Evolution over the Trades (CSET) campaign over the Pacific stratocumulus-to-cumulus transition are organized into 18 Lagrangian cases suitable for study and future modeling, made possible by the use of a track-and-resample flight strategy. Analysis of these cases shows that 2-day Lagrangian coherence of long-lived species (CO and O3) is high (r=0.93 and 0.73, respectively), but that of subcloud aerosol, MBL depth, and cloud properties is limited. Although they span a wide range in meteorological conditions, most sampled air masses show a clear transition when considering 2-day changes in cloudiness (-31%averaged over all cases), MBL depth (1560 m), estimated inversion strength (EIS; 22.2K), and decoupling, agreeing with previous satellite studies and theory. Changes in precipitation and droplet number were less consistent. The aircraft-based analysis is augmented by geostationary satellite retrievals and reanalysis data along Lagrangian trajectories between aircraft sampling times, documenting the evolution of cloud fraction, cloud droplet number concentration, EIS, and MBL depth. An expanded trajectory set spanning the summer of 2015 is used to show that the CSET-sampled air masses were representative of the season, with respect to EIS and cloud fraction. Two Lagrangian case studies attractive for future modeling are presented with aircraft and satellite data. The first features a clear Sc-Cu transition involving MBL deepening and decoupling with decreasing cloud fraction, and the second undergoes a much slower cloud evolution despite a greater initial depth and decoupling state. Potential causes for the differences in evolution are explored, including free-tropospheric humidity, subsidence, surface fluxes, and microphysics. The remaining work focuses on the mesoscale organization of marine low clouds. A convolutional neural network (CNN) model is trained to classify 128 km by 128 km scenes of marine low clouds into six categories: stratus, open-cellular mesoscale cellular convection (MCC), closed-cellular MCC, disorganized MCC, clustered cumulus, and suppressed cumulus. Overall model test accuracy was approximately 90%. This model is applied to three years of data in the southeast Pacific, as well as the 2015 northeast Pacific summer for comparison with the CSET campaign. Meteorological variables related to marine low cloud processes are composited by mesoscale cloud type, allowing for the identification of distinct meteorological regimes. Presentation of MCC is largely consistent with previous literature, both in terms of geographic distribution boundary layer structure, and cloud-controlling factors. The two more novel types, clustered and suppressed cumulus, are examined in more detail. The patterns in precipitation, circulation, column water vapor, and cloudiness are consistent with the presentation of marine shallow mesoscale convective self-aggregation found in previous large eddy simulations of the boundary layer. Although they occur under similar large-scale conditions, the suppressed and clustered low cloud regimes are found to be well-separated by variables associated with a low-level mesoscale circulation, with surface wind divergence being the clearest discriminator between them, whether reanalysis or satellite observations are used. Divergence is consistent with near-surface inflow into clustered regimes and outflow from suppressed regimes. To further understand the dependencies of mesoscale cloud type on environmental factors, a second classification model is built. This uses a random forest of decision trees to predict cloud type, but instead of using an image of a cloud scene, mesoscale averages of meteorological variables are used as inputs. The model uses the three-year dataset output from the CNN model for training, and overall accuracy is approximately 50%. Rotated principal component analysis of the meteorological variables is used to create a set of decorrelated features on which to train the model, allowing for the application of certain statistical analyses which rely on uncorrelated data. Permutation feature importance is used to quantify which variables are most important for correct prediction of cloud mesoscale organization. Overall, temperature and stability are approximately equally important; for correctly distinguishing between open-MCC and closed-MCC, stability is the most important feature, and for correctly distinguishing between suppressed and clustered cumulus, surface divergence is the most important variable. Partial dependence analysis is used to show the relationship between each input variable and the likelihood of observing each cloud type, and 2-dimensional partial dependence analysis shows bimodal distributions of MCC types, consistent with their subtropical and midlatitude incarnations. The random forest model is able to reproduce the geographic distributions of cloud type occurrences.