TITLE:
Classified Early Warning and Forecast of Severe Convective Weather Based on LightGBM Algorithm
AUTHORS:
Xinwei Liu, Haixia Duan, Wubin Huang, Runxia Guo, Bolong Duan
KEYWORDS:
Severe Convective Weather, Machine Learning, LightGBM, Early Warning and Forecast
JOURNAL NAME:
Atmospheric and Climate Sciences,
Vol.11 No.2,
April
14,
2021
ABSTRACT: Severe convective weather
can lead to a variety of disasters, but they are still difficult to be
pre-warned and forecasted in the meteorological operation. This study generates
a model based on the light gradient boosting machine (LightGBM) algorithm using
C-band radar echo products and ground observations, to identify and classify
three major types of severe convective weather (i.e., hail, short-term heavy rain (STHR), convective gust (CG)).
The model evaluations show the LightGBM model performs well in the training set
(2011-2017) and the testing set (2018) with the overall false identification
ratio (FIR) of only 4.9% and 7.0%, respectively. Furthermore, the average
probability of detection (POD), critical success index (CSI) and false alarm
ratio (FAR) for the three types of severe convective weather in two sample sets
are over 85%, 65% and lower than 30%, respectively. The LightGBM model and the
storm cell identification and tracking (SCIT) product are then used to forecast
the severe convective weather 15 - 60 minutes in advance. The average POD, CSI
and FAR for the forecasts of the three types of severe convective weather are
57.4%, 54.7% and 38.4%, respectively, which are significantly higher than those
of the manual work. Among the three types of severe convective weather, the
STHR has the highest POD and CSI and the lowest FAR, while the skill scores for
the hail and CG are similar. Therefore, the LightGBM model constructed in this
paper is able to identify, classify and forecast the three major types of
severe convective weather automatically with relatively high accuracy, and has
a broad application prospect in the future automatic meteorological operation.