downscale: Perfect-prog downscaling

In SantanderMetGroup/downscaleR: A climate4R package for statistical downscaling <http://meteo.unican.es/climate4r>

Perfect-prog downscaling

Description

Workhorse function to call the different perfect-prog downscaling methods

Usage

downscale(
  y,
  x,
  newdata = x,
  method = c("analogs", "glm", "lm"),
  simulate = c(FALSE, TRUE),
  n.analogs = 1,
  sel.fun = c("mean", "wmean", "max", "min", "median"),
  wet.threshold = 0.1,
  n.pcs = NULL,
  cross.val = c("none", "loocv", "kfold"),
  folds = NULL
)

Arguments

y	The observations dataset. It should be an object as returned by loadeR.
x	The input grid. It should be an object as returned by loadeR.
newdata	It should be an object as returned by loadeR and consistent with x. Default is newdata = x.
method	Downscaling method. Options are c = ("analogs","glm","lm"). Glm can only be set when downscaling precipitation.
simulate	A logic value indicating whether we want to simulate or not based on the GLM distributional parameters. Only relevant when perdicting with a GLM. Default to FALSE.
n.analogs	Applies only when method="analogs" (otherwise ignored). Integer indicating the number of closest neigbours to retain for analog construction. Default to 1.
sel.fun	Applies only when method="analogs" (otherwise ignored). Criterion for the construction of analogs when several neigbours are chosen. Ignored when n = 1. Current values are "mean" (the default), "wmean", "max", "min" and "median".
wet.threshold	Value below which precipitation amount is considered zero
n.pcs	Integer indicating the number of EOFs to be used as predictors
cross.val	Should cross-validation be performed? methods available are leave-one-out ("loocv") and k-fold ("kfold"). Default to "none", which does not perform cross-validation.
folds	This arguments controls the number of folds, or how these folds are created (ignored if cross.val = "loocv"). Folds are always splitted chronologically. If it is given as a fraction in the range (0-1), it splits the data in two subsets, one for training and one for testing, being the given value the fraction of the data used for training (i.e., 0.75 will split the data so that 75% of the instances are used for training, and the remaining 25% for testing). In case it is an integer value, it sets the number of folds in which the data will be split (e.g., folds = 10 for the classical 10-fold cross validation). Alternatively, this argument can be passed as a list, each element of the list being a vector of years to be included in each fold (See examples).

Details

Scaling and centering When the climate variables are used as predictors instead of the PCs, these are previously centered and scaled using the mean and sigma parameters globally computed for the whole spatial domain.

Value

The prediction structure.

See Also

downscaleTrain for training a downscaling model downscalePredict for prediction for a a test dataset with a trained model for downscaleCV for automatic cross-validation downscaleR Wiki for downscaling seasonal forecasting and climate projections.

Other downscaling.functions: downscaleCV(), downscaleChunk(), downscalePredict(), downscaleTrain()

Examples

require(transformeR)
require(climate4R.datasets)
data("NCEP_Iberia_hus850")
data("NCEP_Iberia_ta850")
x <- makeMultiGrid(NCEP_Iberia_hus850, NCEP_Iberia_ta850)
newdata <- subsetGrid(x, years = 1994:1995)
x <- subsetGrid(x, years = 1985:1993)
# Loading predictands
data("VALUE_Iberia_pr")
y <- VALUE_Iberia_pr
y <- getTemporalIntersection(obs = y,prd = x, "obs" )
x <- getTemporalIntersection(obs = y,prd = x, "prd" )
### Analogs ###
# None
yp <- downscale(y, x, method = "analogs")
yp <- downscale(y, x, newdata, method = "analogs")
# kfold
yp <- downscale(y, x, method = "analogs", n.pcs = 15,
                cross.val = "kfold", folds = list(c(1985, 1986, 1987),
                                                  c(1988, 1989, 1990),
                                                  c(1991, 1992, 1993)))
# Leave-one-year-out
yp <- downscale(y, x, method = "analogs", n.pcs = 15,
                cross.val = "loocv")

### GLM ###
# None
yp <- downscale(y, x, method = "glm", simulate = FALSE, n.pcs = 10,
                wet.threshold = 1)
yp <- downscale(y, x, method = "glm", simulate = TRUE, n.pcs = 10,
                wet.threshold = 1)
# kfold
yp <- downscale(y, x, method = "glm", simulate = FALSE, n.pcs = 10,
                cross.val = "kfold", folds = list(c(1985, 1986, 1987),
                                                  c(1988, 1989, 1990),
                                                  c(1991, 1992, 1993)))

SantanderMetGroup/downscaleR documentation built on July 4, 2023, 4:28 a.m.