|
|
||||||
   School of Computational Sciences CAMP Home People Research Simulation gallery ├ Hurricane predictions ├ Higher-order upwind schemes on unstructured meshes ├ Urban scale simulation └ Advanced statistical methods (neural networks) Publications Resources for students Data archive Annual conference Computing resources Related links About our webpage Contact us |
 |
Advanced statistical methods (neural networks)The accuracy of surface temperature forecast is important for a mesoscale atmospheric model, since it affects the boundary conditions along the surface. An improvement in the model temperature forecast can be achieved through the implementation of neural network (NN) to post-process output data. The temperature forecasts from a mesoscale model are post-processed by a neural network. For this study, we have intentionally chosen two runs with predicted temperature showing large biases; 1) warm bias case and 2) cold bias case. The eight input variables considered for the neural network are: forecast time, top soil temperature, deep soil temperature, soil moisture, surface latent heat flux, net radiation flux and horizontal surface wind components. The targeted variable is observed surface temperature at a given station. The results show that the model temperature forecasts are improved significantly; a) comparison of time series for the warm bias case, b) comparison of time series for the cold bias case, c) bias and variance for the warm bias case, d) bias and variance for the cold bias case.Click on the images for enlarged views (sizes: 16KB, 16KB, 14KB and 17KB, respectively).
See the
contact page for detailed
information on
contacting us. |
||||
