ABSTRACT: Standard linear and nonlinear downscaling models are compared using identical atmospheric circulation forcing fields. The target variables chosen were observed daily values of average temperature (TAV), precipitation (PRC), and vapor pressure (HPR) at a Central European station. Being without much sophistication, both models show acceptable performance on this time scale only for TAV and HPR; PRC, which behaves in a predominantly nonlinear fashion, handled very poorly. By considerably refining the evaluation it is nevertheless possible to distinguish significant differences between the 2 models and, with the nonlinear model, to describe specific rainfall conditions. We argue that this difference is caused by the limitations of the linear approach, and discuss how this might affect the downscaling of nonlinear quantities in general.

KEY WORDS: Downscaling · Neural nets · Linear · Nonlinear

Full text in pdf format

Published in CR Vol. 10, No. 2 (1998) on August 14
ISSN: 0936-577X. Copyright © Inter-Research, Oldendorf/Luhe, 1998