CLGAN: Guizhou ML-AWS precipitation dataset

This repository provides the preprocessed datasets and source codes, which are used in the study 'CLGAN: A GAN-based video prediction model for precipitation nowcasting' by Ji et al. (2022). This allows the user to reproduce the presented results for the precipitation nowcasting task. To build the dataset, minute-level precipitation measurements by rain gauges of Automatic Weather Stations over Guizhou, China is collected and preprocessed. The raw AWS data comprise measurements from 93 national basic stations and 1740 automatic weather stations across Guizhou at a high observation frequency (every minute). The raw data spans from 1st Jan 2015 and 31st Dec 2019 and is provided by the Guizhou Meteorological Bureau. A series of preprocessing are performed to obtain the proposed dataset, including bilinear interpolation, rainy sequence selection, log transformation and normalization. The details of preprocessing can be found in the paper mentioned above. This folder holds five single netCDF-files which respectively record the data from 2015 to 2019. Each netCDF-file contains preprocessed precipitation sequences with a shape of [n_sequence, len_sequence, n_lat, n_lon]. n_sequence is the number of the recorded sequences, len_sequence is the length of each sequence, n_lat and n_lon are the number of grid points in the meridian and zonal directions. In our case, the target domain covers from 24.625°N to 29.50°N and 103.625°E to 109.50°E (40x48 grid points) with a resolution of 0.125° and the length of each sequence is 24.