tiger: Calculate temporal dynamics of model performance
In tiger: TIme series of Grouped ERrors

Description Usage Arguments Details Value Author(s) References See Also Examples

About fifty performance measures are calculated for a gliding window, comparing two time series. The resulting matrix is clustered, such that each time window can be assigned to an error type cluster. The mean performance measures for each cluster can be used to give meaning to each cluster. Additionally, synthetic peaks are used to better characterize the clusters.

tiger(modelled, measured, window.size, step.size = 1,
                 use.som = TRUE, som.dim = c(20, 20), som.init =
                 "sample", som.topol = "hexa", maxc = 15,
                 synthetic.errors = NA)
tiger.peaks(result, synthetic.errors)

`modelled`	Time series of modelled data
`measured`	Time series of measured data
`window.size`	Size of the moving window
`maxc`	Maximum number of clusters to be tested
`synthetic.errors`	Matrix returned from `synth.peak.error`
`result`	object returned from tiger
`use.som`	boolean, indicating whether to use SOM before applying fuzzy clustering
`som.dim`	Dimension of the Self Organizing Map (SOM) c(x,y)
`som.init`	Method to initialize the SOM
`som.topol`	Topology of the SOM
`step.size`	Size of the steps defining the number of scores to be calculating along the time series. For example, with a value of 5 every fifth value is included

See the package vignette.

`maxc`	see input parameter
`window.size`	see input parameter
`modelled`	see input parameter
`measured`	see input parameter
`synthetic.errors`	see input parameter
`measures.synthetic.peaks`	matrix of performance measures for synthetic errors
`measures`	matrix of performance measures for the gliding time window
`na.rows`	vector of boolean, indicating which time windows contain NA values
`names`	names of the perfomance measures
`measures.uniform`	measures, transformed to uniform distribution
`measures.uniform.synthetic.peaks`	measures for synthetic errors, transformed with the corresponding transformation from previous item
`error.names`	names of the synthetic error types
`best.value.location`	list, indicating what the value for "no error" for each performance measure is
`validityMeasure`	vector with validty index for solutions with 2:maxc clusters
`cluster.assignment`	list of 2:maxc objects returned from `cmeans`

Dominik Reusser

Reusser, D. E., Blume, T., Schaefli, B., and Zehe, E.: Analysing the temporal dynamics of model performance for hydrological models, Hydrol. Earth Syst. Sci. Discuss., 5, 3169-3211, 2008.

The package vignette

data(tiger.example)
modelled <- tiger.single$modelled
measured <- tiger.single$measured
peaks <- synth.peak.error(rise.factor=2, recession.const=0.02, rise.factor2=1.5)
## Not run: result2 <- tiger(modelled=modelled, measured=measured, window.size=240, synthetic.errors=peaks)
errors.in.time(d.dates, result2, solution=6, show.months=TRUE)
## End(Not run)

peaks2 <- synth.peak.error(rise.factor=2, recession.const=0.02,
     rise.factor2=1.5, err1.factor=c(1.3,1.5,2.0),
     err2.factor = c(0.02,0.03,0.06), 
     err3.factor=c(2,4,10), 
     err4.factor = c(9,22,40), 
     err5.factor = c(0.2,0.3,0.5),
     err6.factor =c(2,3,5),
     err9.factor=c(1.5,3,6)
   )

## Not run: result3 <- tiger.peaks(result2, peaks2)

   peaks.in.clusters(result2, solution=6)
   x11()
   peaks.in.clusters(result3, solution=6)

## End(Not run)