Summary

  • CNN-LSTM model demonstrates the ability to automatically extract features in both learning tasks.
  • Search procedure is necessary in order to define suitable network configuration and architecture.
  • Deep learning is able to learn mappings from GCM outputs to local scale observations in an end to end manner.
  • Although differences between locations were apparent.
  • Model does exhibit higher uncertainty in different seasons (summer for single point and winter for 2d grid models).
  • Small decreases in available radiation at the surface predicted under scenario RCP85.

Creative Commons License
Downscaling Global Climate Models with Convolutional and Long-Short-Term Memory Networks for Solar Energy Applications by C.P. Davey is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.