spTimer (version 0.02)

spT.forecast: Forecast for the spatio-temporal models.

Description

This function is used to get the K-step forecast values using MCMC samples.

Usage

spT.forecast(nBurn, K=1, fore.data, fore.coords, 
     posteriors, pred.samples.ar=NULL, Summary=TRUE)

Arguments

nBurn

Number of burn-in. Initial MCMC samples to discard before making inference.

K

Number of K-step (time points) ahead forecast.

fore.data

The data set for the covariate values for forecasts. This data should have the same space-time structure as the original data frame.

fore.coords

The coordinates for the forecast sites. The locations are in similar format to coords.

posteriors

Posterior MCMC samples obtained from function spT.Gibbs.

pred.samples.ar

The prediction output of the function spT.prediction, if forecasts are in the prediction locations. Only used for the model: "AR".

Summary

Logical, if TRUE then provide MCMC summary statistics for forecast.

Value

• forecast.samples
  Forecast MCMC samples.
• n.fore.sites
  Total number of sites for forecasts.
• Mean
  Average of the MCMC predictions, if Summary=TRUE.
• Median
  Median of the MCMC predictions, if Summary=TRUE.
• SD
  Standard deviation of the MCMC predictions, if Summary=TRUE.
• Low
  Lower limit for the 95 percent CI of the MCMC predictions, if Summary=TRUE.
• Up
  Upper limit for the 95 percent CI of the MCMC predictions, if Summary=TRUE.
• computation.time
  The computation time.
• model
  The model method used for prediction.

See Also

spT.Gibbs, spT.prediction.

Examples

##

###########################
## The GP models:
###########################

##
##  Foreecasts: GP models
##

# define forecast coordinates and variables
# forecast in the validation sites (hold-out sites)

fore.coords<-as.matrix(val.site[,2:3])

# one-step ahead forecast, i.e., in day 62
fore.gp <- spT.forecast(nBurn=100, K=1, fore.data=val.fore, 
           fore.coords=fore.coords, posteriors=post.gp, Summary=TRUE)

names(fore.gp)

# forecast validations for the validation sites
spT.validation(val.fore$o8hrmax,c(fore.gp$Mean)) # 
spT.pCOVER(val.fore$o8hrmax,c(fore.gp$Up),c(fore.gp$Low)) # 
spT.segment.plot(val.fore$o8hrmax,c(fore.gp$Mean),
      c(fore.gp$Up),c(fore.gp$Low))
abline(a=0,b=1)


# forecast in the fitted sites 
fore.coords<-as.matrix(fit.site[,2:3])

fore.gp <- spT.forecast(nBurn=100, K=1, fore.data=fit.fore, 
           fore.coords=fore.coords, posteriors=post.gp, Summary=TRUE)
names(fore.gp)

# forecast validations for the fitted sites
spT.validation(fit.fore$o8hrmax,c(fore.gp$Mean)) # 
spT.pCOVER(fit.fore$o8hrmax,c(fore.gp$Up),c(fore.gp$Low)) # 
spT.segment.plot(fit.fore$o8hrmax,c(fore.gp$Mean),
      c(fore.gp$Up),c(fore.gp$Low))
abline(a=0,b=1)


###########################
## The AR models:
###########################


##
## Forecasts: AR models
##

# define forecast coordinates and variables
# forecast in the validation sites (hold-out sites)

fore.coords<-as.matrix(val.site[,2:3])

# one-step ahead forecast, i.e., in day 62
fore.ar <- spT.forecast(nBurn=100, K=1, fore.data=val.fore, 
           fore.coords, pred.samples.ar=pred.ar,
           posteriors=post.ar, Summary=TRUE)

names(fore.ar)

# forecast validations for the validation sites
spT.validation(val.fore$o8hrmax,c(fore.ar$Mean)) # 
spT.pCOVER(val.fore$o8hrmax,c(fore.ar$Up),c(fore.ar$Low)) # 
spT.segment.plot(val.fore$o8hrmax,c(fore.ar$Mean),
      c(fore.ar$Up),c(fore.ar$Low))
abline(a=0,b=1)


# forecast in the fitted sites 
fore.coords<-as.matrix(fit.site[,2:3])

fore.ar <- spT.forecast(nBurn=100, K=1, fore.data=fit.fore, 
           fore.coords, pred.samples.ar=NULL,
           posteriors=post.ar, Summary=TRUE)
names(fore.ar)


# forecast validations for the fitted sites
spT.validation(fit.fore$o8hrmax,c(fore.ar$Mean)) # 
spT.pCOVER(fit.fore$o8hrmax,c(fore.ar$Up),c(fore.ar$Low)) # 
spT.segment.plot(fit.fore$o8hrmax,c(fore.ar$Mean),
      c(fore.ar$Up),c(fore.ar$Low))
abline(a=0,b=1)


###########################
## Models with GPP approximations:
###########################


##
## Forecasts: GPP based models
##

# define forecast coordinates and variables
# forecast in the validation sites (hold-out sites)
fore.coords<-as.matrix(val.site[,2:3])

# one-step ahead forecast, i.e., in day 62
fore.gpp <- spT.forecast(nBurn=100, K=1, fore.data=val.fore, 
           fore.coords=fore.coords, posteriors=post.gpp, 
           Summary=TRUE)

names(fore.gpp)

# forecast validations for the validation sites
spT.validation(val.fore$o8hrmax,c(fore.gpp$Mean)) #
spT.pCOVER(val.fore$o8hrmax,c(fore.gpp$Up),c(fore.gpp$Low)) # 
spT.segment.plot(val.fore$o8hrmax,c(fore.gpp$Mean),
      c(fore.gpp$Up),c(fore.gpp$Low))
abline(a=0,b=1)


# forecast in the fitted sites 
forecast.coords<-as.matrix(fit.site[,2:3])

fore.gpp <- spT.forecast(nBurn=100, K=1, fore.data=fit.fore, 
           fore.coords=forecast.coords, posteriors=post.gpp, 
           Summary=TRUE)
names(fore.gpp)

# forecast validations for the fitted sites
spT.validation(fit.fore$o8hrmax,c(fore.gpp$Mean)) # 
spT.pCOVER(fit.fore$o8hrmax,c(fore.gpp$Up),c(fore.gpp$Low)) # 
spT.segment.plot(fit.fore$o8hrmax,c(fore.gpp$Mean),
      c(fore.gpp$Up),c(fore.gpp$Low))
abline(a=0,b=1)


##