| KGE {greenbrown} | R Documentation |
This function is an implementation of the Kling-Gupta efficiency (KGE) (Gupta et al. 2009) for model evaluation. It was originally developed to compare modelled and observed time series. The KGE is a model evluation criterion that can be decomposed in the contribution of mean, variance and correlation on model performance. In this implemenation, the Kling-Gupta effciency is defined as following: KGE = 1 - eTotal eTotal is the euclidean distance of the actual effects of mean, variance, correlation and trend (optional) on the time series: eTotal = sqrt(eMean + eVar + eCor + eTrend) eTotal can be between 0 (perfect fit) and infinite (worst fit).
KGE(x, ref, trend = FALSE, mosum.pval = 0.05, h = 0.15, breaks = 0,
eTrend.ifsignif = TRUE, ...)
x |
time series from model result or factorial model run |
ref |
reference time series (observation or standard model run) |
trend |
Include the effect of trend in the calculation? (default: FALSE) |
mosum.pval |
(only used if trend=TRUE) See |
h |
(only used if trend=TRUE) See |
breaks |
(only used if trend=TRUE) See |
eTrend.ifsignif |
compute effect on trend only if trend in reference series is significant, if FALSE compute always effect on trend (if trend = TRUE) |
... |
further arguments for the function |
The function returns a vector with the following components:
KGE Kling-Gupta effciency = 1 - eTotal
eTotal total effect, i.e. euclidean distance
fMean fraction of mean to the total effect
fVar fraction of variance to the total effect
fCor fraction of correlation to the total effect
fTrend fraction of trend to the total effect (only if trend=TRUE)
eMean effect of mean
eVar effect of variance
eCor effect of correlation
eTrend effect of trend (only if trend=TRUE)
Matthias Forkel <matthias.forkel@geo.tuwien.ac.at> [aut, cre]
Gupta, H. V., H. Kling, K. K. Yilmaz, G. F. Martinez (2009): Decomposition of the mean squared error and NSE performance criteria: Implications for improving hydrological modelling. Journal of Hydrology 377, 80-91.
# load a time series of NDVI (normalized difference vegetation index) data(ndvi) plot(ndvi) # change the variance and compute effect x <- ndvi + rnorm(length(ndvi), 0, 0.01) plot(x, ndvi); abline(0,1) KGE(x, ndvi, trend=FALSE) # change mean and variance and compute effect x <- ndvi + rnorm(length(ndvi), 0.02, 0.01) plot(x, ndvi); abline(0,1) KGE(x, ndvi, trend=FALSE) # be careful when using trends and breakpoints # using trends is howver not part of the original implementation # of the Kling-Gupta efficiency KGE(x, ndvi, trend=TRUE, breaks=0) KGE(x, ndvi, trend=TRUE, breaks=1)