It's really a shame that they've chosen not to open source their algorithm. They use data from the 4km (hi-res) NAM model, which can be found here[0]. The GRIB files can be read and exported to CSV using Panopoly [1]. They mention using RH at different heights along with low, medium, and high cloud cover (LCDC, MCDC, HCDC). I wanted to make a simple model and train it on Instagram data for #sunset tag frequency for a given location, but Instagram just closed access to their global data. If someone else has access and wants to run with this, it could be fun...<p>[0] <a href="http://nomads.ncep.noaa.gov/cgi-bin/filter_hiresconus.pl" rel="nofollow">http://nomads.ncep.noaa.gov/cgi-bin/filter_hiresconus.pl</a><p>[1] <a href="http://www.giss.nasa.gov/tools/panoply/" rel="nofollow">http://www.giss.nasa.gov/tools/panoply/</a>
More at <a href="https://news.ycombinator.com/item?id=10623436" rel="nofollow">https://news.ycombinator.com/item?id=10623436</a>.