estCrossVal {EGRET}

Jack-Knife cross validation of the WRTDS (Weighted Regressions on Time, Discharge, and Season)

Description

This function fits the WRTDS model n times (where n is the number of observations). For each fit, the data value being estimated is eliminated from the record. This gives predictions that do not depend on knowing the actual result for that day. Thus it provides for a more "honest" estimate of model performance than a traditional error analysis that uses all the data.

Usage

estCrossVal(DecLow, DecHigh, Sample, windowY = 7, windowQ = 2,
  windowS = 0.5, minNumObs = 100, minNumUncen = 50, edgeAdjust = TRUE,
  verbose = TRUE)

Arguments

DecLow

number specifying minimum decimal year

DecHigh

number specifying maximum decimal year

Sample

data frame containing the sample values, default is Sample

windowY

numeric specifying the half-window width in the time dimension, in units of years, default is 7

windowQ

numeric specifying the half-window width in the discharge dimension, units are natural log units, default is 2

windowS

numeric specifying the half-window with in the seasonal dimension, in units of years, default is 0.5

minNumObs

numeric specifying the miniumum number of observations required to run the weighted regression, default is 100

minNumUncen

numeric specifying the minimum number of uncensored observations to run the weighted regression, default is 50

edgeAdjust

logical specifying whether to use the modified method for calculating the windows at the edge of the record. The modified method tends to reduce curvature near the start and end of record. Default is TRUE.

verbose

logical specifying whether or not to display progress message

Value

SampleCrossV data frame containing the sample data augmented by the results of the cross-validation exercise

Examples

eList <- Choptank_eList
Sample <- getSample(eList)
Daily <- getDaily(eList)
numDays <- length(Daily$DecYear)
DecLow <- Daily$DecYear[1]
DecHigh <- Daily$DecYear[numDays]

SampleCrossV <- estCrossVal(DecLow,DecHigh,Sample)