Non‐linearity in statistical downscaling: does it bring an improvement for daily temperature in Europe?

Abstract

Several linear and non-linear statistical downscaling methods are compared for winter daily temperature at eight European stations. The linear methods include linear regression of gridpoint values (pointwise regression) and of predictors' principal components (PC regression). The non-linear methods are represented by artificial neural networks. The non-linearity is also achieved by a stratification of data by classification of circulation patterns and a linear regression conducted separately within each class. As predictors, gridded 500 hPa heights and 850 hPa temperature are used. The verification is conducted in the cross-validation framework. The downscaling methods are evaluated according to four criteria: (1) fit to observations (quantified by the correlation coefficient), (2) shape of the statistical distribution, namely its skewness and kurtosis, (3) temporal autocorrelations with 1 day lag, and (4) interstation correlations. Considering all the criteria together, the pointwise linear regression appears to be the best method. It achieves the best fit with the observations and possesses the best temporal structure. The deviations of statistical distributions from normality are only captured by the neural networks, while the classification methods yield the best spatial correlations.

Spatial Downscaling of TRMM Precipitation Product Using a Combined Multifractal and Regression Approach: Demonstration for South China

“…Comparison studies show both methods could have superiority to others while dealing with different study areas, predictands, predictors, and statistical indicators used for judgement [48,49]. Thus, because there is no a priori supposition of the relationship, we would try both multi-linear and ANN models in our study for regression.…”

Downscaling of surface moisture flux and precipitation in the Ebro Valley (Spain) using analogues and analogues followed by random forests and multiple linear regression

“…Among nonlinear methods is worth mentioning the use of different types of NN which are reported to outperform linear models (Weichert and Bürger, 1998;Miksovsky and Raidl, 2005;Davy et al, 2010). Other works suggest that for statistical downscaling of temperature, linear models cannot be beaten by NN (Huth et al, 2008).…”

Reassessing Statistical Downscaling Techniques for Their Robust Application under Climate Change Conditions

“…1; note that, hereafter, Zi stands for a specific zone. Over this region we considered a number of atmospheric variables (see Table 1) typically used as predictors in temperature downscaling studies (Benestad 2002;Huth 2002;Hanssen-Bauer et al 2005;Huth et al 2008). It has been recently shown that these variables -considering anomalies-are suitable predictors for climate change studies, since their distribution is skillfully reproduced by Global Climate Models (GCMs) in the area under study (see Brands et al 2011a).…”