 |
| Titel |
Downscaling of GCM parameter outputs to RCM spatial scale using an artificial neural network |
| VerfasserIn |
Robin Chadwick, Erika Coppola, Filippo Giorgi |
| Konferenz |
EGU General Assembly 2010
|
| Medientyp |
Artikel
|
| Sprache |
Englisch
|
| Digitales Dokument |
PDF |
| Erschienen |
In: GRA - Volume 12 (2010) |
| Datensatznummer |
250033346
|
|
|
|
|
|
| Zusammenfassung |
| An artificial neural network (ANN) approach was used to downscale temperature and rainfall fields from the ECHAM-5 global climate model to the spatial scale of the RegCM3 regional climate model over Europe. Inputs to the ANN include the GCM temperature/rainfall field, the GCM and RCM orography fields, and the distance between GCM and RCM gridpoints.
The ANN was trained with 20 years of RegCM3 and ECHAM-5 data, then ANN downscaled estimates were assessed against RegCM3 outputs for several different time periods within a 120 year model run. A comparison was also performed of the ANN method against a simple lapse-rate downscaling method for the ECHAM-5 temperature field. It was found that the ANN was able to accurately reproduce RegCM3 parameter fields for a validation time period near to the training time period, but not for time periods far from the training time period. For validation periods near to the training time-period, the ANN approach outperformed the lapse-rate method.
Work is ongoing into a ‘timeslice’ ANN training method, using data from three distinct 10 year slices within a RegCM3 model run for training, and results from this will be presented. |
|
|
|
|
|
|