biasCorrect: Biascorrect the input timeseries or hyfo dataset

In hyfo: Hydrology and Climate Forecasting

Biascorrect the input timeseries or hyfo dataset

Description

Biascorrect the input time series or dataset, the input time series or dataset should consist of observation, hindcast, and forecast. observation and hindcast should belong to the same period, in order to calibrate. Then the modified forecast will be returned. If the input is a time series, first column should be date column and rest columns should be the value column. If the input is a hyfo dataset, the dataset should be the result of loadNcdf, or a list file with the same format.

Usage

biasCorrect(
  frc,
  hindcast,
  obs,
  method = "scaling",
  scaleType = "multi",
  preci = FALSE,
  prThreshold = 0,
  extrapolate = "no"
)

## S4 method for signature 'data.frame,data.frame,data.frame'
biasCorrect(
  frc,
  hindcast,
  obs,
  method = "scaling",
  scaleType = "multi",
  preci = FALSE,
  prThreshold = 0,
  extrapolate = "no"
)

## S4 method for signature 'list,list,list'
biasCorrect(
  frc,
  hindcast,
  obs,
  method = "scaling",
  scaleType = "multi",
  preci = FALSE,
  prThreshold = 0,
  extrapolate = "no"
)

Arguments

frc	a hyfo grid data output or a dataframe (time series) consists of Date column and one or more value columns, representing the forecast to be calibrated.
hindcast	a hyfo grid data output or a dataframe(time series) consists of Date column and one or more value columns, representing the hindcast data. This data will be used in the calibration of the forecast, so it's better to have the same date period as observation data. Check details for more information.
obs	a hyfo grid data output or a dataframe (time series) consists of Date column and one or more value columns, representing the observation data.
method	bias correct method, including 'delta', 'scaling'..., default is 'scaling'
scaleType	only when the method "scaling" is chosen, scaleType will be available. Two different types of scaling method, 'add' and 'multi', which means additive and multiplicative scaling method. More info check details. Default scaleType is 'multi'.
preci	If the precipitation is biascorrected, then you have to assign preci = TRUE. Since for precipitation, some biascorrect methods may not apply to, or some methods are specially for precipitation. Default is FALSE, refer to details.
prThreshold	The minimum value that is considered as a non-zero precipitation. Default to 0 (assuming mm). If you want to use precipitation biascorrect, you should consider carefully how to set this threshold, usually is 1. But you can try with different numbers to see the results.
extrapolate	When use 'eqm' method, and 'no' is set, modified frc is bounded by the range of obs. If 'constant' is set, modified frc is not bounded by the range of obs. Default is 'no'.

Details

Since climate forecast is based on global condition, when downscaling to different regions, it may include some bias, biascorrection is used then to fix the bias.

Hindcast

In order to bias correct, we need to pick up some data from the forecast to train with the observation, which is called hindcast in this function. Using hindcast and observation, the program can analyze the bias and correct the bias in the forecast.

Hindcast should have EVERY attributes that forecast has.

Hindcast is also called re-forecast, is the forecast of the past. E.g. you have a forecast from year 2000-2010, assuming now you are in 2005. So from 2000-2005, this period is the hindcast period, and 2005-2010, this period is the forecast period.

Hindcast can be the same as forecast, i.e., you can use forecast itself as hindcast to train the bias correction.

How it works

Forecast product has to be calibrated, usually the system is doing forecast in real time. So, e.g., if the forecast starts from year 2000, assuming you are in year 2003, then you will have 3 years' hindcast data (year 2000-2003), which can be used to calibrate. And your forecast period is (2003-2004)

E.g. you have observation from 2001-2002, this is your input obs. Then you can take the same period (2001-2002) from the forecast, which is the hindcast period. For forecast, you can take any period. The program will evaluate the obs and hindcast, to get the modification of the forecast, and then add the modification to the forecast data.

The more categorized input, the more accurate result you will get. E.g., if you want to bias correct a forecast for winter season. So you'd better to extract all the winter period in the hindcast and observation to train. extractPeriod can be used for this purpose.

method

Different methods used in the bias correction. Among which, delta, scaling can be applied to different kinds of parameters, with no need to set preci; eqm has two conditions for rainfall data and other data, it needs user to input preci = TRUE/FALSE to point to different conditions; gqm is designed for rainfall data, so preci = TRUE needs to be set.

delta

This method consists on adding to the observations the mean change signal (delta method). This method is applicable to any kind of variable but it is preferable to avoid it for bounded variables (e.g. precipitation, wind speed, etc.) because values out of the variable range could be obtained (e.g. negative wind speeds...)

scaling

This method consists on scaling the simulation with the difference (additive) or quotient (multiplicative) between the observed and simulated means in the train period. The additive or multiplicative correction is defined by parameter scaling.type (default is additive). The additive version is preferably applicable to unbounded variables (e.g. temperature) and the multiplicative to variables with a lower bound (e.g. precipitation, because it also preserves the frequency).

eqm

Empirical Quantile Mapping. This is a very extended bias correction method which consists on calibrating the simulated Cumulative Distribution Function (CDF) by adding to the observed quantiles both the mean delta change and the individual delta changes in the corresponding quantiles. This method is applicable to any kind of variable.

It can keep the extreme value, if you choose constant extrapolation method. But then you will face the risk that the extreme value is an error.

gqm

Gamma Quantile Mapping. This method is described in Piani et al. 2010 and is applicable only to precipitation. It is based on the initial assumption that both observed and simulated intensity distributions are well approximated by the gamma distribution, therefore is a parametric q-q map that uses the theorical instead of the empirical distribution.

It can somehow filter some extreme values caused by errors, while keep the extreme value. Seems more reasonable. Better have a long period of training, and the if the forecast system is relatively stable.

It is a generic function, if in your case you need to debug, please see ?debug() for how to debug S4 method.

Author(s)

Yuanchao Xu [e-mail]

References

Bias correction methods come from biasCorrection from dowscaleR

• Santander Meteorology Group (2015). downscaleR: Climate data manipulation and statistical downscaling. R package version 0.6-0. https://github.com/SantanderMetGroup/downscaleR/wiki

• R.A.I. Wilcke, T. Mendlik and A. Gobiet (2013) Multi-variable error correction of regional climate models. Climatic Change, 120, 871-887

• A. Amengual, V. Homar, R. Romero, S. Alonso, and C. Ramis (2012) A Statistical Adjustment of Regional Climate Model Outputs to Local Scales: Application to Platja de Palma, Spain. J. Clim., 25, 939-957

• C. Piani, J. O. Haerter and E. Coppola (2009) Statistical bias correction for daily precipitation in regional climate models over Europe, Theoretical and Applied Climatology, 99, 187-192

• O. Gutjahr and G. Heinemann (2013) Comparing precipitation bias correction methods for high-resolution regional climate simulations using COSMO-CLM, Theoretical and Applied Climatology, 114, 511-529

Examples

######## hyfo grid file biascorrection
########

# If your input is obtained by \code{loadNcdf}, you can also directly biascorrect
# the file.

# First load ncdf file.
filePath <- system.file("extdata", "tnc.nc", package = "hyfo")
varname <- getNcdfVar(filePath)    
nc <- loadNcdf(filePath, varname)

data(tgridData)
# Since the example data, has some NA values, the process will include some warning #message, 
# which can be ignored in this case.




# Then we will use nc data as forecasting data, and use itself as hindcast data,
# use tgridData as observation.
newFrc <- biasCorrect(nc, nc, tgridData)  
newFrc <- biasCorrect(nc, nc, tgridData, scaleType = 'add')   
newFrc <- biasCorrect(nc, nc, tgridData, method = 'eqm', extrapolate = 'constant', 
preci = TRUE) 
newFrc <- biasCorrect(nc, nc, tgridData, method = 'gqm', preci = TRUE) 


######## Time series biascorrection
########

# Use the time series from testdl as an example, we take frc, hindcast and obs from testdl.
data(testdl)

# common period has to be extracted in order to better train the forecast.

datalist <- extractPeriod(testdl, startDate = '1994-1-1', endDate = '1995-10-1')

frc <- datalist[[1]]
hindcast <- datalist[[2]]
obs <- datalist[[3]]


# The data used here is just for example, so there could be negative data.

# default method is scaling, with 'multi' scaleType
frc_new <- biasCorrect(frc, hindcast, obs)

# for precipitation data, extra process needs to be executed, so you have to tell
# the program that it is a precipitation data.

frc_new1 <- biasCorrect(frc, hindcast, obs, preci = TRUE)

# You can use other scaling methods to biascorrect.
frc_new2 <- biasCorrect(frc, hindcast, obs, scaleType = 'add')

# 
frc_new3 <- biasCorrect(frc, hindcast, obs, method = 'eqm', preci = TRUE)
frc_new4 <- biasCorrect(frc, hindcast, obs, method = 'gqm', preci = TRUE)

plotTS(obs, frc, frc_new, frc_new1, frc_new2, frc_new3, frc_new4, plot = 'cum')

# You can also give name to this input list.
TSlist <- list(obs, frc, frc_new, frc_new1, frc_new2, frc_new3, frc_new4)
names(TSlist) <- c('obs', 'frc', 'delta', 'delta_preci', 'scale', 'eqm', 'gqm')
plotTS(list = TSlist, plot = 'cum')


# If the forecasts you extracted only has incontinuous data for certain months and years, e.g.,
# for seasonal forecasting, forecasts only provide 3-6 months data, so the case can be 
# for example Dec, Jan and Feb of every year from year 1999-2005.
# In such case, you need to extract certain months and years from observed time series.
# extractPeriod() can be then used.
  
  



# More examples can be found in the user manual on https://yuanchao-xu.github.io/hyfo/

hyfo documentation built on Aug. 16, 2023, 5:08 p.m.