![]() Furthermore we have improved accuracy and efficiency of the GP solution by implementing constant variation and optimization as genetic operators. We have found this to be a simple task to the GP system. ![]() the hydrostatic equation, from our training data. At first we have tested the GP system to retrieve the known physical rule for downscaling surface pressure, i.e. We are currently working with GPLAB, a Genetic Programming toolbox for Matlab. To cope with the vast number of possible solutions, we use genetic programming, a method from machine learning, which is based on the principles of natural evolution. To accomplish that, we broaden the search by allowing for interdependencies between different atmospheric parameters, non-linear relations, non-local and time-lagged relations. precipitation or near surface specifc humidity. Aim of our work is to improve those rules and to find deterministic rules for the remaining variables, which require downscaling, e.g. Up to now deterministic rules are available for downscaling surface pressure and partially, depending on the prevailing weather conditions, for near surface temperature and radiation. ![]() The deterministic rules in this scheme are partly based on known physical relations and partly determined by an automated search for linear relationships between the high resolution fields of the atmospheric model output and high resolution data on surface characteristics. For the development of the scheme, training and validation data sets have been created by carrying out high-resolution runs of the atmospheric model. The current downscaling scheme combines a bi-quadratic spline interpolation, deterministic rules and autoregressive noise. Subject of our work is the downscaling scheme used to derive high resolution forcing data for land-surface and subsurface models from coarser atmospheric model output. Coupling models for the different components of the Soil-Vegetation-Atmosphere-System requires up-and downscaling procedures.
0 Comments
Leave a Reply. |
Details
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |