The dataset files provided here are produced through random sampling and splitting in train and test sets from an existing large dataset produced by Sarazin (2018).

Description of Sarazin (2018) dataset:
- The dataset contains 17,000,000 simulations (data points) of a hydrological model and 34 input variables which characterize climate properties and land cover types. 
- There are 4 classes in the dataset, namely, C1 (<20 mm/yr), C2 (20 – 100 mm/yr), C3 (100 – 300 mm/yr) and C4 (>300 mm/yr) which correspond to different recharge values.

The input variables and the classes remain the same for the sample dataset with the only difference the amount of data points it contains.

Reference:
Understanding the sensitivity of karst groundwater recharge to climate and land cover changes at a large-scale, Sarrazin, F. (Author). 25 Sep 2018
