[A-AS05] From Weather Predictability to Controllability

[AAS05-04] A Study on the Meteorological Environment at the Initiation of Mesoscale Convective Systems in Malaysia

*Abdul Aizat Nazmi Bin A Azmi¹, Hironobu Iwabuchi¹ (1.Center for Atmospheric and Oceanic Studies, Graduate School of Science, Tohoku University)

Keywords:mesoscale convective system, initiation, machine learning

The investigation explores how effectively the ingredient-based methodology differentiates prospective initiations of isolated deep convection and mesoscale convective systems (IDC and MCS, respectively) and how much it utilizes moisture, instability, lifting, and vertical wind shear for prediction. The study used the MCS tracking algorithm to identify the initiation of convective system life stages. It uses input data from NASA Global Merged IR brightness temperature and the Global Precipitation Mission (GPM) Integrated Multi-satellitE Retrievals for GPM (IMERG) V06B precipitation data alongside global reanalyses to analyze the moisture, lifting, instability, and vertical wind shear. We use tree-based machine learning models such as Extremely Randomized Trees (ET), Random Forest (RF), eXtreme Gradient Boosting (XGB), and Light Gradient-Boosting Machine (LGBM) to differentiate the prospective initiations of IDC and MCS with a lead time of 6 hours based on the history of the environmental factors for 24 hours. The discussion covers the model's performance skills. The XGB performs the best with a symmetric extreme dependency score of 0.76. The RF performs the second best with 0.74. In addition, we utilize sensitivity analysis to determine to what degree the features can separate the IDC and MCS. The input data of each environmental factor is perturbed only at 7 hours lead time depending on a situation, whether it is increased, decreased, or unchanged, ranging from 0 to 10% at 1 % intervals. The sensitivity analysis indicates a consistently rising trend of MCS initiation events, mainly depending on an increasing trend of moisture and lifting regardless of instability and vertical wind shear direction and magnitude. In this study, machine learning can classify MCS 6 hours before initiating using only ingredient-based methodology, even with imbalanced class distributions.