SpatioTemporal (version 1.1.7)

updateTrend.STdata: Update Trend in STdata or STmodel Object

Description

Updates/sets the temporal trend for STdata or STmodel objects. It also checks that the spatio-temporal covariate exists for all dates in the trend, mainly an issue if extra.dates!=NULL adds additional times at which to do predictions.

Usage

# S3 method for STdata
updateTrend (object, n.basis = 0,
    fnc = NULL, extra.dates = NULL, ...)
  # S3 method for STmodel
updateTrend (object, n.basis = 0,
    fnc = NULL, ...)
  updateTrend(object, n.basis = 0, fnc = NULL, ...)
  updateSTdataTrend(object, n.basis = 0,
    extra.dates = NULL, fnc = NULL, ...)

Arguments

object

A STdata or STmodel object, see mesa.data.raw.

n.basis

number of basis functions for the temporal trend

extra.dates

Additional dates for which smooth trends should be computed (otherwise only those in object$obs$date are used); only for STdata.

fnc

Function that defines the trend, see Details and Example.

...

Additional parameters passed to calcSmoothTrends.

Value

Returns a modfied version of the input, with an added/altered trend.

Details

If n.basis is given this will use calcSmoothTrends to compute smoothed SVDs of data for use as temporal trends. If fnc is given, n.basis is ignored and fnc should be a function that, given a vector of dates, returns an object that can be coerced to a data.frame with numeric temporal trends; recall that an intercept is always added.

For a STmodel object the new trend must have no more components than the existing trend; if a function is given colnames of the new trend must match those of the existing trend. In both cases the returned STdata or STmodel object will have both a $trend and $trend.fnc field.

Function updateSTdataTrend is deprecated and will be removed in future versions of the package.

See Also

Other STdata functions: c.STmodel, createDataMatrix, createSTdata, createSTmodel, detrendSTdata, estimateBetaFields, removeSTcovarMean

Other STmodel functions: createCV, createDataMatrix, createSTmodel, dropObservations, estimateBetaFields, loglikeST, loglikeSTdim, loglikeSTnaive, predictNaive, processLocation, processLUR, processST, updateCovf

Other SVD for missing data: boxplot.SVDcv, calcSmoothTrends, plot.SVDcv, print.SVDcv, summary.SVDcv, SVDmiss, SVDsmooth, SVDsmoothCV

Examples

# NOT RUN {
##load data
data(mesa.model)

##default data and time trend for one location
par(mfrow=c(3,1), mar=c(2.5,2.5,3,1))
plot(mesa.model)

##let's try with no trend
mesa.model.0 <- updateTrend(mesa.model, n.basis=0)
plot(mesa.model.0)

##...and two basis functions, based on only AQS sites and much less smooth
subset <- mesa.model$locations$ID[mesa.model$locations$type=="AQS"]
mesa.model.2 <- updateTrend(mesa.model, n.basis=2, subset=subset, df=100)
plot(mesa.model.2)

##Compute trends based on only 10 sites (and compute the cross-validated
##trends leaving each of the sites out
smooth.trend <- calcSmoothTrends(mesa.model, n.basis=2, cv=TRUE,
                                 subset=subset[1:10])

##update trends using the function definition
mesa.model <- updateTrend(mesa.model, fnc=smooth.trend$trend.fnc)

##and create objects with each of the trends.
mesa.model.cv <- vector("list", length(smooth.trend$trend.fnc.cv))
for(i in 1:length(mesa.model.cv)){
  suppressMessages(mesa.model.cv[[i]] <- updateTrend(mesa.model, 
                   fnc=smooth.trend$trend.fnc.cv[[i]]))
}

##plot
par(mfrow=c(1,1),mar=c(2.5,2.5,3,1))
plot(mesa.model)
for(i in 1:length(mesa.model.cv)){
  plot(mesa.model.cv[[i]], add=TRUE, col=i, pch=NA, lty=c(NA,2))
}
# }