Data for the manuscript "DEUCE v1.0: A neural network for probabilistic precipitation nowcasting with aleatoric and epistemic uncertainties" by Harnist et al. (2023)


File descriptions: and

Contain folders of raw PGM composites, containing needed input data for verification experiments, that is for the creation of baseline nowcast as well as observation HDF5 files.

Contains HDF5 archives of the composite crops used for the training of DEUCE, derived from raw PGM composites using the script given.

Contains the two intermediate pytorch lightning checkpoints of DEUCE, as the final model checkpoint used for making predictions.

Contains the netcdf files of verification metric values computed from DEUCE and baseline model predictions compared to observations.

Metadata Access
Creator Harnist, Bent; Pulkkinen, Seppo; Mäkinen, Terhi
Publisher Finnish Meteorological Institute
Publication Year 2023
Rights CC-BY; info:eu-repo/semantics/openAccess
OpenAccess true
Contact bent.harnist(at)
Language English
Resource Type Dataset
Format zip; py
Size 7.5 GB; 6 files
Discipline Environmental science