Weather radar observation dataset for machine learning

<p>This file contains a NetCDF4-format dataset of precipitation radar data extracted from the MeteoSwiss radar composite of the year 2018. There are a total of 180000 samples, each containing a series of 8 radar images spaced 10 minutes apart, of 128x128 pixels each. Some of the samples overlap with each other. The spatial resolution is approximately 0.5 km.</p>

<p>The dataset is made available to support the code found here: <a href="https://github.com/jleinonen/geogan">https://github.com/jleinonen/geogan</a></p>

<p>The dataset has been split into three parts due to file size limitations. To use the data, merge the files e.g. as follows:
<br><code>cat samples-2018-128x128.nc-a samples-2018-128x128.nc-b samples-2018-128x128.nc-c > samples-2018-128x128.nc</code><br>
The data in the "sequences" variable is given as 8-bit unsigned integers, which are decoded into rain rates using the conversion table found in the scale_rzc.txt file. Values of 0 are decoded to the value on the first line of this file, values of 1 to the value on the second line, and so on.</p>

We thank MeteoSwiss for providing the data and permitting us to release it in this form.