Climatology of Linear Mesoscale Convective System Morphology in the United States based on Random Forests Method

Wenjun Cui,Xiquan Dong,Baike Xi,Zhe Feng
DOI: https://doi.org/10.1175/jcli-d-20-0862.1
IF: 5.38
2021-07-01
Journal of Climate
Abstract:Abstract This study uses machine learning methods, specifically the random forest (RF), on a radar-based mesoscale convective system (MCS) tracking dataset to classify the five types of linear MCS morphology in the contiguous United States during the period 2004-2016. The algorithm is trained using radar- and satellite-derived spatial and morphological parameters, and reanalysis environmental information from 5-yr manually identified nonlinear and five linear MCS modes. The algorithm is then used to automate the classification of linear MCSs over 8 years with high accuracy, providing a systematic, long-term climatology of linear MCSs. Results reveal that nearly 40% of MCSs are classified as linear MCSs, in which half of the linear events belong to the type of system having a leading convective line. The occurrence of linear MCSs shows large annual and seasonal variations. On average, 113 linear MCSs occur annually during the warm season (through March to October), with most of these events clustered from May through August in the central eastern Great Plains. MCS characteristics, including duration, propagation speed, orientation, and system cloud size, have large variability among the different linear modes. The systems having a trailing convective line and the systems having a back-building area of convection typically move more slowly and have higher precipitation rate, and thus have higher potential in producing extreme rainfall and flash flooding. Analysis of the environmental conditions associated with linear MCSs show that the storm-relative flow is of most importance in determining the organization mode of linear MCSs.
meteorology & atmospheric sciences
What problem does this paper attempt to address?