DSensemble: Downscale ensemble runs

In metno/esd: Climate analysis and empirical-statistical downscaling (ESD) package for monthly and daily data

Downscale ensemble runs

Description

Downscales an ensemble of climate model runs, e.g. CMIP5, taking the results to be seasonal climate statistics. For temperature, the result hold the seasonal mean and standard deviation, whereas for precipitation, the results hold the wet-day mean, the wet-day frequency, and the wet/dry-spell statistics. The call assumes that netCDF files containing the climate model ensemble runs are stores in a file structure, linked to the path argument and the rcp argument.

Usage

DSensemble(y, ...)

Arguments

y	A station object.
...	additional arguments
plot	Plot intermediate results if TRUE.
path	The path where the GCM results are stored.
rcp	Which (RCP) scenario
biascorrect	TRUE, apply a bias adjustment using biasfix
predictor	The predictor, a field or EOF object
non.stationarity.check	If TRUE perform stationarity test - work in progress
ip	Which EOFs to include in the step-wise multiple regression.
rmtrend	TRUE: detrend before calibrating the regression model.
lon	Longitude range for predictor
lat	Latitude range for predictor
rel.cord	TRUE: use the range relative to predictand; FALSE use absolute range
it	Used to extract months or a time period. See subset.
select	GCMs to select, e.g .subsample the ensemble (1:3 selects the three first GCMs)
FUN	Function for aggregating the predictand (daily), e.g. 'mean', 'wetmean'
threshold	Used together with FUN for some functions ('wetmean').
nmin	Minimum number of day used in annual used for aggregating the predictand/predictor
FUNX	Function for transforming the predictor, e.g. 'C.C.eq' to estimate the saturation water vapout
type	Type of netCDF used in retrieve for reading GCM data.
pattern	File name pattern for GCM data.
verbose	TRUE for checking and debugging the functions.
file.ds	Name of file saving the results.
path.ds	Path of file saving the results.
xfuns	Names of functions which do not work in annual(x,FUN=f). These functions are used using the following code annual(f(x),FUN="mean")
mask	TRUE mask out land
ds.interval	Default set to NULL, otherwise set the time period for downscaling, e.g. c(1950,2100)

Details

These methods are based on DS, and DSensemble is designed to make a number of checks and evaluations in addition to performing the DS on an ensemble of models. It is based on a similar philosophy as the old R-package 'clim.pact', but there is a new default way of handling the predictors. In order to attempt to ensure a degree of consistency between the downscaled results and those of the GCMs, a fist covariate is introduced before the principal components (PCs) describing the EOFs.

DSensemble.pca is used to downscale a predictor represented in terms of PCA, and can reduce the computation time significantly. See Benestad et al. (2015) http://dx.doi.org/10.3402/tellusa.v67.28326.

The argument non.stationarity.check is used to conduct an additional test, taking the GCM results as 'pseudo-reality' where the predictand is replaced by GCM results interpolated to the same location as the provided predictand. The time series with interpolated values are then used as predictor in calibrating the model, and used to predict future values. This set of prediction is then compared with the interpolated value itself to see if the dependency between the large and small scales in the model world is non-stationary.

Other chekch include cross-validation (crossval) and diagnostics comparing the sample of ensemble results with the observations: number of observations outside the predicted 90-percent conf. int and comparing trends for the past.

The 'bias correction' is described in Imbert and Benestad (2005), Theor. Appl. Clim. http://dx.doi.org/10.1007/s00704-005-0133-4.

Value

A 'dsensembele' object - a list object holding DS-results.

Examples

## Not run: 
# Import historical temperature data from Oslo
data(Oslo)

## Download NorESM1-M from 'climexp.knmi.nl' in default directory
## (home directory for linux/mac users)
url <-"http://climexp.knmi.nl/CMIP5/monthly/tas"
## Download NorESM1-ME for the emission scenario RCP4.5
noresm <- "tas_Amon_NorESM1-M_rcp45_000.nc"
if (!file.exists(noresm)) {
  download.file(url=file.path(url,noresm), destfile=noresm,
                method="auto", quiet=FALSE, mode="w", cacheOK=TRUE)
}

## Download FIO-ESM for the emission scenario RCP4.5
fioesm <- "tas_Amon_FIO-ESM_rcp45_000.nc"
if (!file.exists(fioesm)) {
  download.file(url=file.path(url,fioesm), destfile=fioesm,
                method="auto", quiet=FALSE, mode="w", cacheOK=TRUE)
}

## Downscale the predictor (ERA-interim reanalysis 2m temperature)
predictor <- "air.2m.mon.mean.nc"
if (!file.exists(predictor)) {
  url <-"http://climexp.knmi.nl/ERA-interim/erai_t2m.nc"
  download.file(url=url, destfile=predictor,
                method="auto", quiet=FALSE, mode="w", cacheOK=TRUE)
}

# Downscale the temperature in Oslo
rcp4.5 <- DSensemble.t2m(Oslo, path='~', rcp='', pattern="tas_Amon_",
                         biascorrect=TRUE, predictor = predictor,
                         plot=TRUE, verbose=TRUE)

## Evaluation: 
## (1) combare the past trend with downscaled trends for same
## interval by ranking and by fitting a Gaussian to the model ensemble;
## (2) estimate the probability for the counts outside the 90
## percent confidence interval according to a binomial distribution.
diagnose(rcp4.5, plot = TRUE, type = "target")

## End(Not run)

metno/esd documentation built on March 9, 2024, 11:21 a.m.