tiger (version 0.2.2)

tiger: Calculate temporal dynamics of model performance

Description

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.

Usage

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)

Arguments

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

Value

• 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

Details

See the package vignette.

References

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.

See Also

The package vignette

Examples

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)
result2 <- tiger(modelled=modelled, measured=measured, window.size=240, synthetic.errors=peaks)
errors.in.time(d.dates, result2, solution=6, show.months=TRUE)

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)
   )

result3 <- tiger.peaks(result2, peaks2)

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