WG: Weather generators for conditioned on simulated climate...

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

Weather generators for conditioned on simulated climate aggregated statistics.

Description

Weather generators for conditional simulation of daily temperature and/or precipitation, given mean and/or standard deviation. The family of WG functions produce stochastic time series with similar characteristics as the station series provided (if none if provided, it will use either ferder or bjornholt provided by the esd-package). Here characteristics means similar mean value, standard deviation, (frequency and wet-day mean precipitation for precipitation), and spectral properties. FTscramble takes the Fourier components (doing a Fourier Transform - FT) of a series and reassigns random phase to each frequency and then returns a new series through an inverse FT. The FT scrambling is used for temperature, but not for daily precipitation that is non-Gaussian and involves sporadic events with rain. Instead, FTscramble is used for randomly scrambling annual wet-day frequency and wet-day mean precipitation, however, and for precipitation, the annual wet-day frequency and the annual wet-day mean precipitation are used to randomly generate exponentially distributed numbers to provide similar aggregated annual statistics as the station or predicted though downscaling. The precipitation WG can also take into account the number of consecutive number-of-wet-days statistics using a geometric distribution.

Usage

WG(x, ...)

Arguments

x	station object
...	additional arguments
option	Define the type of WG
amean	annual mean values. If NULL, use those estimated from x; if NA, estimate using DSensemble.t2m, or if provided, assume a 'dsensemble' object.
asd	annual standard deviation. If NULL, use those estimated from x; if NA, estimate using DSensemble.t2m, or if provided, assume a 'dsensemble' object.
t	Time axis. If null, use the same as x or the last interval of same length as x from downscaled results.
ip	passed on to DSensemble.t2m
select	passed on to DSensemble.t2m
lon	passed on to DSensemble.t2m
lat	passed on to DSensemble.t2m
plot	if TRUE, plot results
biascorrect	passed on to DSensemble.t2m
verbose	passed on to DSensemble.t2m
mu	annual wet-mean values. If NULL, use those estimated from x; if NA, estimate using DSensemble.t2m, or if provided, assume a 'dsensemble' object.
fw	annual wet-day frequency. If NULL, use those estimated from x; if NA, estimate using DSensemble.t2m, or if provided, assume a 'dsensemble' object.
ndd	annual mean dry spell length. If NULL, use those estimated from x; if NA, estimate using DSensemble.t2m, or if provided, assume a 'dsensemble' object.
threshold	Definition of a rainy day.
method	Assume a gemoetric or a poisson distribution. Can also define ownth methods.
t2m	station object with temperature
precip	station object with precipitation.
ndbr	Number of
n.spells.year	= c('fw','spell') if 'fw' then estimate number of spells according to 365.25 else estimate number of events from spell.
alpha.scaling	TRUE scale the low-probability events according to alpha in DOI:10.1088/1748-9326/ab2bb2
alpha	values for alpha-scaling
ensure.fw	TRUE then WG tries to ensure that fw of simulations match those of observations or prescribed by adding or subtracting wet days.
w.fw.ac	weighting to balance how the wet day occurrences follows seasonal cycle or randomness. 0 - no seasonal cycle; 1000 - mainly determined by climatology (default=30).
w.mu.ac	same as above, but for wet-day mean precipitation (default=10).

Details

The weather generator produces a series with similar length as the provided sample data, but with shifted dates according to specified scenarios for annual mean mean/standard deviation/wet-day mean/wet-day frequency.

WG.FT.day.t2m generates daily temperature from seasonal means and standard deviations. It is given a sample station series, and uses FTscramble is based on a Fourier Transform which generates a new series with random phase but nevertheless similar (or predicted - in the future) spectral characteristics as the original series. It then uses a quantile transform to prescribe predicted mean and standard deviation, assuming the distributions are normal, which usually is OK for seasonall/annually aggregated statistics (e.g. annual mean, annual wet-day frequency, or annual wet-day mean precipitation). The temporal structure (power spectrum) of the random series is therefore similar as the sample provided.

WG.fwmu.day.precip has been designed to be used with downscaled results for annual wet-day frequency and annual wet-day mean precipitation. It also tries to simulate the wet-spell duration statistics (number of consecutive wet days) based on provided sample data (argument x). The process can take annual wet-day mean precipitation and the wet-day frequency as input when used to make projections for the future, together with a sample station of daily values, to simulate stochastic numbers of consecutive wet days, based on its annual mean number of consecutive wet days. It also uses the mean annual cycle of wet-day frequency as well as the wet-day mean precipitation to guide the seasonal timing of wet days and amounts, and hence tries to mimic rain seasons. If not specified, it is taken from the sample data after being phase scrambled/shuffled (FTscramble - a bit like a deck of cards). If not specified, the annual wet-day frequency is a phase-scrambled version of annual aggregates from the sample data. The daily amount is taken from stochastic values generated with rexp scaled for the tail according to alpha in (described in DOI: 10.1088/1748-9326/ab2bb2) as in day2IDF. The number of consecutive wet days is approximated by a geometric distribution (rgeom), and the annual number of wet days is either given as input or estimated from the sample series. test.WG.fwmu.day.precip presents diagnostics of tests of WG.fwmu.day.precip.

Author(s)

R.E. Benestad

Examples

## Temperature
data(ferder)
x <- WG(ferder)
## Plot the results
plot(merge(ferder,x), xlab='', ylab=c('Obs T2m','WG T2m'), col='blue', main=paste(loc(x),' Obs/WG'))

## Daily precipitation
data(bjornholt)
z <- WG(bjornholt)
## Plot the results
plot(merge(bjornholt,z), xlab='', ylab=c('Obs precip','WG precip'), col='blue', main=paste(loc(z),' Obs/WG'))
sz <- sort(coredata(z)[index(z) %in% index(bjornholt)])
sy <- sort(coredata(bjornholt)[index(bjornholt) %in% index(z)])
## Use WG to 'simulate' climate change
z2 <- WG(bjornholt, mu=annual(bjornholt, FUN='wetmean') + 2)
sz2 <- sort(coredata(z2)[index(z2) %in% index(bjornholt)])

## Plot the comparison of quantiles
plot(sy, sz, pch=19, cex=0.7, main='QQ-plot', xlab='Observations', ylab='WG')
grid()
lines(c(0, max(sy,sz,na.rm=TRUE)), c(0,max(sy,sz,na.rm=TRUE)), lty=2, col='red')
points(sy, sz2, col='blue', cex=0.7)


## Simple simulation of contnued trends in wet-day mean precipitation and frequency
mu <- annual(bjornholt,FUN='wetmean',nmin=270) # Avoid missing values (NA)
fw <- annual(bjornholt,FUN='wetfreq',nmin=270) # Avoid missing values (NA)
mu.trend <- trend(mu)
fw.trend <- trend(fw)
## Construct precipitation statistics for input to WG
mu2 <- c(mu,zoo(coredata(mu)+coredata(max(mu.trend)-min(mu.trend)),order.by=max(index(mu))+1:length(mu)))
fw2 <- c(fw,zoo(coredata(fw)+coredata(max(fw.trend)-min(fw.trend)),order.by=max(index(fw))+1:length(fw)))
z <- WG(bjornholt,mu=mu2,fw=fw2,verbose=TRUE)
plot(z)

#' ## Test the WG
z2 <- WG(bjornholt,w.mu.ac=1000,plot=TRUE,verbose=TRUE)
plot(aggregate(z2,by=month,FUN='wetmean')); lines(aggregate(bjornholt,by=month,FUN='wetmean'))
z3 <- WG(bjornholt,w.fw.ac=1000,plot=TRUE,verbose=TRUE)
plot(aggregate(z3,by=month,FUN='wetfreq')); lines(aggregate(bjornholt,by=month,FUN='wetfreq'))

## Test-routine for WG
test.WG.fwmu.day.precip()

metno/esd documentation built on Oct. 11, 2024, 11:37 p.m.