| Abstrakt: | This thesis deals with improving temperature forecasts created by meteorological model
Aladin, that is used by Slovak Hydrometeorological Institute. The goal is to further
improve results achieved in the diploma thesis written by Michal Hajdin in 2016 and
experiment with models which include their previous errors as new features. We present
analysis of data we worked with, new approaches that further improved the results,
and the results reached after experimenting with various learning algorithms and using
previous error terms. Finally, we present improvement obtained for all the meteoro-
logical stations, whose data were available for our thesis. Our final model decreased
absolute error of Aladin predictions from 18.95% up to 73.61%.
|
|---|