"An lightweight CNN-CBAM model for identifying remnant low systems from global reanalysis"

"This dataset includes samples of Normal Tropical Cyclones, Remnant Lows\uff08RL\uff09, and Non-Tropical-Cyclone (No-TC) conditions. The labels for Normal Tropical Cyclones and Residual Systems are derived from the China Meteorological Administration (CMA) best-track dataset, with lifecycle stage definitions referenced from IBTrACS.The No-TC samples are generated based on the CMA best-track records over coastal regions of China. Specifically, the coastal area is divided into three sub-regions: Region A (Guangdong, Guangxi, Hainan, Hong Kong, Macao, and adjacent seas), Region B (Fujian Province and the Xiamen offshore area), and Region C (Jiangsu, Zhejiang, Shanghai, and adjacent seas). For each month from 2000 to 2022, ten time points are randomly selected as candidate No-TC samples.The selection of No-TC samples follows two constraints. First, if a tropical cyclone occurs in Region A during a given period while no cyclone activity is detected in Region B, samples may be selected from Region B. Second, sampling is excluded within three days before the onset and three days after the termination of any tropical cyclone lifecycle. For example, if a cyclone occurs in Region A on 2 August 2001, no samples are generated from that region within the \u00b13-day window, and sampling is shifted to Region B or Region C.ERA5 reanalysis grid data are then retrieved for all selected samples. Six key meteorological variables representing the dynamical and thermodynamical structures of tropical cyclones are extracted: mean sea level pressure (MSL), maximum wind speed (i10fg), mean wind speed (10fg), 850-hPa relative vorticity (Vo850), 300-hPa temperature (t300), and 500-hPa temperature (t500). These variables are merged in the above order and stored in NumPy (.npy) format.The resulting dataset is suitable for studies on residual system identification, tropical cyclone lifecycle classification, and data-driven prediction of remnant low systems using reanalysis data."