Data-driven Discovery of Snow Cover Parameterization

<p>All data were derived from SNOTEL, Version 1, and were preprocessed for training symbolic regression models in convenience. Train and test data are range from water years of 2001-2009, and of 2010-2018. All NaN value were removed from data and shape as [Sample, feat]. The dataset contains preprocessed dimensonless features:</p> <p>The name and unit of each feature were listed in sequence as follows:</p> <p>&nbsp; &nbsp; 1. Snow depth (mm)<br>&nbsp; &nbsp; 2. Snow water equivalent (mm)<br>&nbsp; &nbsp; 3. Standard deviation of sub-grid topography (m)<br>&nbsp; &nbsp; 4. Air temperature (K)<br>&nbsp; &nbsp; 5. Precipitation (mm/day)<br>&nbsp; &nbsp; 6. 1/snow density (mm/mm)<br>&nbsp; &nbsp; 7. 1/Standard deviation of sub-grid topography (m^-1)</p> <p>The name and unit of target were listed as follows:</p> <p>&nbsp; &nbsp; &nbsp;1. Snow cover fraction [%]</p> <p>Some own defined constant:</p> <ol> <li>surface roughness (0.1 m)</li> <li>0 degree of temperature (273.16 K)</li> <li>own defined std threshold (200 m)</li> <li>mean SWE (122.3 mm)</li> </ol>