Technical ReportPDF Available

Bias Adjusting Climate Model Projections

Authors:
  • The Climate Data Factory

Abstract and Figures

This document is an overview of the different data sets provided by the climate data factory on its site theclimatedatafactory.com. It is intended for users who wish to apply these data in climate change impact studies, from local to global scale. This document describes essential information about data origin, processing methods, metadata information and assumption and limitations.
No caption available
… 
No caption available
… 
Content may be subject to copyright.
T E C H N I C A L REPORT
BI A S A D J U S T I N G C L I M A T E
M OD E L P R O J E C T I O N S
| FEBRUARY 2018
Technical Report
Bias Adjusting Climate Model Projections
Florian Cochard - Harilaos Loukos - Thomas Noël
Version 1.0
February 2018
Technical Report: Bias Adjusting Climate Model Projections | 1
Cover photo: Mountains and Clouds by rachel_thecat / CC BY-SA 2.0
Please cite this report as:
the climate data factory ”2018), Technical report: bias adjusting climate model
projections.
Technical Report: Bias Adjusting Climate Model Projections | 2
Table of contents
Table of contents 2
Intent of this Document 4
Data Field Descriptions 5
Data Origin and Methods 7
Introduction 7
Methods 8
Datasets 8
CMIP5 8
CORDEX 8
Preprocessing 9
Remapping 9
Bias Adjustment 10
Delta methods 10
Quantile matching methods 10
Standardization 11
Quality Control 12
QA-DKRZ 12
In-House 12
Spatial Extraction 13
Country-level 13
City-level 14
Considerations and Recommended Use 15
Recommended Use 15
Assumptions and Limitations 15
References 16
Dataset and Document Revision History 20
Annex 21
Technical Report: Bias Adjusting Climate Model Projections | 3
 
Technical Report: Bias Adjusting Climate Model Projections | 4
1. Intent of this Document
This document is an overview of the different data sets provided by the climate data
factory on its site theclimatedatafactory.com. It is intended for users who wish to
apply these data in climate change impact studies, from local to global scale. This
document describes essential information about data origin, processing methods,
metadata information and assumption and limitations. References and
supplementary information are provided at the end of this document.
The data sets on the climate data factory are meant to support users involved in
climate change adaptation topics, such as impact researchers, adaptation
practitioners, urban planners or energy professionals, conducting local, regional or
global studies. Each data set is remapped on a reference grid, bias-adjusted for direct
use in impacts studies and quality controlled to comply with climate communitys
standards, data consistency and metadata. Dont hesitate to contact us to share
insights and comments at: support@theclimatedatafactory.com.
Website URL:
https://theclimatedatafactory.com
Technical point of contact:
Thomas Noel thomas[at]theclimatedatafactory.com 
Technical Report: Bias Adjusting Climate Model Projections | 5
2. Data Field Descriptions
Variable name, units
tas
Daily Near-Surface Air Temperature
Degrees Kelvin
Spatial resolution
0.50 degrees x 0.50 degrees ”CMIP5)
0.11 degrees x 0.11 degrees ”EURO-CORDEX)
Temporal resolution
Daily from 1951-01-01 to 2100-12-31
Units are in days since a reference date ”e.g., 1850-01-01). The
reference date varies by model and experiment, and is based on
the reference date used in the corresponding CMIP5 GCM or
CORDEX RCM experiment.
Coverage
Country to city level
Variable name, units
tasmin
Daily Minimum Near-Surface Air Temperature
Degrees Kelvin
Spatial resolution
0.50 degrees x 0.50 degrees ”CMIP5)
0.11 degrees x 0.11 degrees ”EURO-CORDEX)
Temporal resolution
Daily from 1951-01-01 to 2100-12-31
Units are in days since a reference date ”e.g., 1850-01-01). The
reference date varies by model and experiment, and is based on
the reference date used in the corresponding CMIP5 GCM or
EURO-CORDEX RCM experiment.
Coverage
Country to city level
Variable name, units
tasmax
Daily Maximum Near-Surface Air Temperature
Degrees Kelvin
Spatial resolution
0.50 degrees x 0.50 degrees ”CMIP5)
0.11 degrees x 0.11 degrees ”EURO-CORDEX)
Temporal resolution
Daily from 1951-01-01 to 2100-12-31
Units are in days since a reference date ”e.g., 1850-01-01). The
reference date varies by model and experiment, and is based on
the reference date used in the corresponding CMIP5 GCM or
CORDEX RCM experiment.
Coverage
Country to city level
Technical Report: Bias Adjusting Climate Model Projections | 6
Variable name, units
pr
Precipitation ”mean of the daily precipitation rate)
kg m-2 s-1
Spatial resolution
0.50 degrees x 0.50 degrees ”CMIP5)
0.11 degrees x 0.11 degrees ”EURO-CORDEX)
Temporal resolution
Daily from 1951-01-01 to 2100-12-31
Units are in days since a reference date ”e.g., 1850-01-01). The
reference date varies by model and experiment, and is based on
the reference date used in the corresponding CMIP5 GCM or
CORDEX RCM experiment.
Coverage
Country to city level
rsds
Daily Surface Downwelling Shortwave Radiation
W m-2
0.50 degrees x 0.50 degrees ”CMIP5)
0.44degrees x 0.44 degrees ”EURO-CORDEX)
Daily from 1951-01-01 to 2100-12-31
Units are in days since a reference date ”e.g., 1850-01-01). The
reference date varies by model and experiment, and is based on
the reference date used in the corresponding CMIP5 GCM or
CORDEX RCM experiment.
Country to city level
Variable name, units
sfcWind
Daily Near-Surface Wind Speed
m s-1
Spatial resolution
0.50 degrees x 0.50 degrees ”CMIP5)
0.11 degrees x 0.11 degrees ”EURO-CORDEX)
Temporal resolution
Daily from 1951-01-01 to 2100-12-31
Units are in days since a reference date ”e.g., 1850-01-01). The
reference date varies by model and experiment, and is based on
the reference date used in the corresponding CMIP5 GCM or
CORDEX RCM experiment.
Coverage
Country to city level
Technical Report: Bias Adjusting Climate Model Projections | 7
3. Data Origin and Methods
3.1. Introduction
The data sets on the climate data factory include remapped, bias-adjusted and quality
controlled climate scenarios for 66 countries and more than 4,300 cities worldwide.
They are derived from the General Circulation Model ”GCM) and Regional Climate
Model ”RCM) runs conducted under 2 model intercomparison projects: the Coupled
Model Intercomparison Project Phase 5 ”CMIP5) ”Taylor et al. 2012) and the Coordinated
Regional Downscaling Intercomparison project ”CORDEX) ”Giorgi et al. 2009), and forced
under 2 greenhouse gas emissions scenarios known as Representative Concentration
Pathways ”RCPs) ”Moss et al. 2010). The CMIP5 climate projections were used for the
Fifth Assessment Report of the Intergovernmental Panel on Climate Change ”IPCC AR5).
Raw CMIP5 and CORDEX archives are extracted from the Earth System Grid
Federation ”ESGF) data portal with Synda, a software developed by the Institut Pierre
Simon Laplace.
A climate model is an approximate representation of the real world climate drivers.
This simplification is due to incomplete understanding of climate physics and is
required for computational purpose. This inevitably introduces random models errors
in models simulations when their statistical properties ”e.g., mean, variance) are
compared to climatological observations, thus limiting the use of raw models data in
impact studies.
The Cumulative Distribution Function transform ”CDF-t) method ”Michelangeli et al.,
2009, Vrac et al., 2016, Famien et al., 2017) used in generating data sets on the climate
data factory is a bias-adjustment method we co-developed with academics to address
climate models limitations. CDF-t is a variant of the quantile mapping ”QM) method
which consists in comparing the cumulative distribution function ”CDF) of a climate
variable ”e.g., temperature) at large scale ”e.g., from GCM) to the CDF of the same
variable at a local scale. CDF-t method has been extensively used in the literature and
validated for many variables ”e.g., Kallache et al., 2011; Vrac et al., 2012; Lavaysse et
al., 2012; Vautard et al., 2013; Vrac and Friederichs, 2015).
We have applied the CDF-t method to create a data archive of bias-adjusted CMIP5
and EURO-CORDEX climate projections. The purpose of these data sets lies in bridging
the gap between raw models data and climate change impact studies. CFD-t method
was applied on 20 CMIP5 GCMs ”Table 1 in Annex) and 18 EURO-CORDEX RCMs ”Table
2 in Annex) for the periods from 2006 to 2100 ”Climate Projections) under 2 RCP
scenarios ”RCP4.5 and RCP8.5) and from 1951 through 2005 ”Historical) for the
historical experiment. The observation-based reference dataset for CMIP5 is the
WATCH Forcing Data methodology applied to ERA Interim reanalysis data ”WFDEI;
Technical Report: Bias Adjusting Climate Model Projections | 8
Weedon et al., 2014) from 1979 to 2012 on a 0.5° x 0.5° grid. The observation-based
reference dataset for EURO-CORDEX is the Mesoscale Analysis System ”MESAN;
Landelius et al. 2016) from 1989 to 2012 on a 0.11° x 0.11° grid. The final result is a
data archive size of more than 5 TB.
This document provides a description of the data origin and the CDF-t method as
applied in the bias-adjustment of the CMIP5 GCMs and EURO-CORDEX RCMs data. The
code for CDF-t method is freely available as an R package ”link:
https://www.rdocumentation.org/packages/CDF-t/versions/1.0.1/topics/CDF-t).
Additional technical details may also be found in Michelangeli et al., 2009 and Vrac et
al. 2016.
3.2. Methods
3.2.1. Datasets
CMIP5
Climate Model Data: we compiled 39 climate simulations from 22 CMIP5 GCMs and 2
RCP scenarios ”RCP4.5 and RCP8.5; see Table 1 in Annex) . Each of the climate
1
simulations include daily near-surface temperature, daily maximum near-surface
temperature, daily minimum near-surface temperature, daily precipitation, daily
near-surface wind speed and daily surface downwelling shortwave radiation for the
periods from 1951 through 2005 ”Historical) and from 2006 to 2100 ”Climate
Projections). Unless specified, all 39 climate simulations are bias-adjusted through the
same procedures.
Observational Climate Data: we used the WATCH Forcing Data methodology applied
to ERA Interim reanalysis data ”WFDEI; Weedon et al., 2014) as the observation-based
reference dataset. This dataset is based on the European Centre for Medium-range
Weather Forecasts ”ECMWF) ERA-Interim reanalysis. It includes eight meteorological
variables at 3-hourly time steps, and as daily averages from 1979 to 2012, for the
global land surface at 0.5° x 0.5° resolution, including Antarctica ”Weedon et al., 2014).
We used the 0.5° x 0.5° resolution and historical data from 1979 to 2012 for daily
near-surface temperature, daily maximum near-surface temperature, daily minimum
near-surface temperature, daily precipitation, daily near-surface wind speed and daily
surface downwelling shortwave radiation.
CORDEX
Climate Model Data: we compiled 18 climate simulations from 4 CORDEX RCMs forced
by 5 GCMs and 2 RCP scenarios ”RCP4.5 and RCP8.5; see Table 2 in Annex). Each of the
climate simulations include daily temperature, maximum temperature, minimum
1 17 GCMs for RCP4.5 projection and 22 GCMs for RCP8.5 projection.
Technical Report: Bias Adjusting Climate Model Projections | 9
temperature, precipitation, wind speed and solar radiation for the periods from 1951
through 2005 ”Historical) and from 2006 to 2100 ”Climate Projections). Unless
specified, all 18 climate simulations are bias-adjusted and quality controlled with the
same procedures.
Observational Climate Data: we used the Mesoscale Analysis System ”MESAN;
Häggmark et al., 2000) as the observation-based reference dataset. MESAN is a system
for operational mesoscale univariate analysis of selected meteorological parameters
”see Landelius et al. 2016 for more information). We used the 0.11° x 0.11° resolution
and historical data from 1989 to 2012 for daily near-surface temperature, daily
maximum near-surface temperature, daily minimum near-surface temperature, daily
precipitation and daily near-surface wind speed. Daily surface downwelling shortwave
radiation was missing from MESAN, so we used a 0.44° x 0.44° resolution grid based
on WFDEI data.
3.2.2. Preprocessing
Raw CMIP5 and CORDEX archives are extracted from the Earth System Grid
Federation ”ESGF) data portal with Synda, a software developed by the Institut Pierre
Simon Laplace. Models data are first checked to make sure there are neither technical
nor numerical bugs, and to validate metadata integrity.
3.2.3. Remapping
Models intercomparison, bias adjustment or comparison of output model simulations
with observations require spatial interpolation of models data on a reference grid. The
remapping step consists in spatially interpolating raw daily models data to a finer
reference-grid resolution. We use the Climate Data Operators ”CDO) ”CDO, 2016)
software from the Max Planck Institute which gathers various algorithms for
interpolation used by the scientific community.
Only CMIP5 GCMs are remapped on the 0.5° x 0.5° grid of WFDEI. Indeed, all raw
CORDEX variables are already available on the 0.11° x 0.11° grid from MESAN, except
rsds variable which is missing from MESAN observations. Hence, the rsds variable from
CORDEX 0.11° x 0.11° is remapped on the CORDEX grid at 0.44° x 0.44° resolution ”see
Figure 1).
From one variable to another, different interpolation methods are used, depending on
the trend ”linear, non-linear) or distribution ”gaussian, non gaussian, etc.) of the CMIP5
variable:
tas, tasmin, tasman and sfcWind are interpolated with a bicubic method
pr and rsds are interpolated with a conservative method ”first and second
order)
Technical Report: Bias Adjusting Climate Model Projections | 10
3.2.4. Bias Adjustment
Models have skills in simulating future climate but show systematic biases when
statistically compared to climatological observations. Bias-adjustment methods are
used to calibrate model simulations to ensure their statistical properties are similar
to those of the corresponding observed values ”climate4impacts.com glossary). There
are in the litterature two types of approaches to adjust climate model outputs:
Delta methods are the simplest and consist in adjusting the average simulation
outputs
Quantile matching methods are most advanced and commonly found in the
litterature. Rather than focusing solely on the average outputs, they aim at
correcting the complete statistical distribution of model variables.
Delta methods
Delta methods ”Hay et al., 2000; Lenderink et al., 2007; Gudmundsson et al., 2012)
consist in disturbing time-series of simulated variables through constant addition or
multiplication of an adjustment coefficient. These methods can whether adjust the
mean and/or the standard deviation of a simulated variable. However, this basic
approach can only be used with time-series of observations, and arent appropriate to
adjust the simulated climate variability.
Quantile matching methods
Quantile mapping ”QM) ”Wood et al., 2004; Maurer et al., 2007; Déqué et al., 2007;
Christensen et al., 2008; Lopez et al., 2009) are more advanced adjustment methods
and consist in comparing the cumulative distribution function ”CDF) of a variable ”e.g.,
temperature) at a global scale ”e.g., from GCM) to the CDF of the variable at a
local-scale ”e.g., station). Different variations of the QM approach are referenced in
the literature of which Quantile Delta Mapping ”Cannon et al., 2015), Scaled
Distribution Mapping ”Switanek et al., 2016) and Linear Regression Quantile Mapping
”Passow et al., 2017). For a review of methods see Galmarini et al 2018).
We use a variant of QM method called Cumulative Distribution Function transform
”CDF-t). While standard QM approach only projects the CDF of a simulated large-scale
variable ”i.e., from GCM) onto the CDF of the historical to compute and match
quantiles, CDF-t accounts for the evolution of the large-scale CDF from historical to
future time period ”Michelangeli et al., 2009, Vrac et al., 2012). As a result, the adjusted
climate projections have the same CDF as the observations data, and potential biases
in the statistical structure of the raw climate model simulations are removed.
CDF-t was first used to adjust wind variable ”Michelangeli et al., 2009) and is now
referenced in more than 100 peer-reviewed publications to adjust different sets of
variables such as temperature, precipitation or solar radiation ”e.g., Oettli et al., 2011 ;
Technical Report: Bias Adjusting Climate Model Projections | 11
Vrac et al., 2012 ; Lavaysse et al., 2012 ; Colette et al., 2012 ; Tisseuil et al., 2012 ; Vrac
et al., 2016).
We applied CDF-t to 5 variables ”i.e., tas, tasmin, tasmax, rsds, sfcWind) over the
period 1951-2100 ”historical, RCP4.5, RCP8.5). The precipitation variable ”i.e., pr) was
adjusted with an updated version of CDF-t referred to as Singularity Stochastic
Removal ”SSR) which considers rainfall occurrence and intensity issues ”see Vrac et
al., 2016 for more details).
Figure 1. Processing chain from raw to adjusted climate models data
3.2.5. Standardization
Standardization consists in rewriting output data files and related metadata to comply
with the climate communitys standards ”e.g., the Climate and Forecast metadata
convention and the Data Reference Syntax). We use the Climate Model Output
Rewriter 2 ”CMOR 2) library.
Technical Report: Bias Adjusting Climate Model Projections | 12
3.2.6. Quality Control
For each bias-adjusted variable, we check data compliance with climate community“s
standards, data consistency and metadata. Doing quality control is crucial in the data
publication process and data re-use. We use the QA-DKRZ methodology combined
with an additional in-house quality control that checks values of adjusted and
standardized variables data.
QA-DKRZ
During the Quality Assurance process of the DKRZ, the following criteria are checked:
1. Number of data sets is correct and > 0
2. Size of every data set is > 0
3. The data sets and corresponding metadata are accessible
4. The data sizes are controlled and correct
5. The spatial-temporal coverage description ”metadata) is consistent to the data,
time steps are correct and the time coordinate is continuous
6. The format is correct
7. Variable description and data are consistent
In-House
In addition, an In-house quality control is built upon CDO and NCO tools and consists
twofold in:
Analyzing the difference between adjusted model and observation values on
the reference period
Analyzing the time evolution difference between adjusted and non-adjusted
model.
Difference between adjusted model and observations
First, we estimate two quantities:
average for the months of the year on the reference period for observations
average for the months of the year on the reference period for adjusted model
Then, we estimate the difference between these two quantities and get 12 files in
output ”one per month). For each month ”i.e., for each file), we take the 10th and 90th
quantile which gives 12 values for each quantile.
Finally, we control that these 12 values are comprised in the following ranges:
temperature between [ -1 ; 1 ] in K
precipitation between [-0.5 ; 0.5] in mm.day-1
solar radiation between [-5 ; 5] in W.m-2
surface wind between [-0.5 ; 0.5] in m.s-1
Technical Report: Bias Adjusting Climate Model Projections | 13
If values are outside the range, the script raises an error with a 1 status.
Difference between adjusted model and non-adjusted model
First, we estimate four quantities:
Average for the seasons on the reference period for the non-adjusted model
Average for the seasons on the reference period for the adjusted model
Average for the seasons on the future period ”2071-2100) for the non-adjusted
model
Average for the seasons on the future period ”2071-2100) for the adjusted
model
Then, we compute the evolution between future and reference periods for the
non-adjusted and adjusted model. We estimate the difference between them and get
4 files in output ”one per season). For each season ”i.e., for each file), we take the 10th
and 90th quantile of the difference which gives 4 values for each quantile.
Finally, we control these 4 values are comprised in the following range:
temperature between [-2 ; 2] in K
precipitation between [-1 ; 1]in mm.day-1
solar radiation between [-10 ; 10] in W.m-2
surface wind between [-1 ; 1] in m.s-1
If values are outside the range, the quality control raises an error with a 1 status.
3.2.7. Spatial Extraction
Raw ESGF files are available as global ”CMIP5) or continental ”CORDEX) domains ”Asia,
Europe, etc.), so we extract country-level and city-level information to help users
focusing on their area of interest.
Country-level
Country-level extraction method consists in identifying the border grid points for a
country and drawing a rectangle around them. The drawback is that neighboring
country points can be included in this rectangle. In the next version, well create a
mask per country to only consider country points.
Technical Report: Bias Adjusting Climate Model Projections | 14
City-level
Model grid points are spaced approximately every 100 km to 50 km for CMIP5 models
and 15 to 10 km for CORDEX models. To extract city-level information, we consider the
nearest grid point for a city ”we only consider continental points). City level data
correspond to a single grid point. They give a trend but do not account for local
phenomena like the urban heat island effect that modulates small scale changes and
requires higher resolution ”typically 100m) and specific modeling to be resolved.
Technical Report: Bias Adjusting Climate Model Projections | 15
4. Considerations and Recommended Use
4.1. Recommended Use
The data sets on the climate data factory are meant to support users involved in
climate change adaptation topics, of which impact researchers, adaptation
practitioners, urban planners or energy sector. Data are intended for use in scientific
research and impact studies from local to global scales. Extensive metadata
information are provided in netCDF files.
4.2. Assumptions and Limitations
Bias-adjustment consists in calibrating model simulations to ensure their statistical
properties are similar to those of the corresponding observed values ”as from
climat4impact.eu). Some authors claim that bias-adjustment techniques introduce
another level of uncertainty making evaluation of projections uncertainty even harder
”e.g., Ehret et al. 2012, Maraun et al., 2016). There are still differences of opinion
regarding whether direct or bias-adjusted climate model simulations should be used
in impact modeling and assessment. On the one hand, the use of direct climate model
simulations ensures spatial and temporal consistency across variables, on the other
hand the substantial biases of raw variables renders direct climate model simulations
unrealistic and ultimately unsuitable for climate change impact modeling. While the
climate modeling community continues to improve climate models, statistical bias
adjustment is currently necessary to make climate projections fit for purpose in
impact modeling and assessment ”Ficklin et al 2016, climate4impact.eu).
As described in Section 3.2.4, The CDF-t bias-adjustment method preserves long-term
trend in climate models data. CDF-t performances are not sensitive to the climate
model performance but to the variability and trend of its driving large-scale fields
”reanalysis or GCMs/RCMs control runs) which can perform better or worse depending
on the variable but also on the season ”Vrac et al., 2012). However, to represent a
correct CDF under historical and/or present climate conditions does not guarantee to
correctly represent the evolution of the CDF in a climate change context. In addition,
CDF-t method is a univariate adjustment method which is applied location by location,
and is not designed to reproduce multi-dimensional properties ”e.g., variable
covariance and spatial correlations). Development of a multivariate and spatial
version of the CDF-t method is underway ”Vrac 2018). The aforementioned limitations
are not specific to the CDF-t method but common to any univariate Quantile-Quantile
method.
Technical Report: Bias Adjusting Climate Model Projections | 16
5. References
Climate Data Operators, Max Planck Institute, 2016.
https://code.zmaw.de/projects/cdo
Christensen, J., F. Boberg, O. Christensen, and P. Lucas-Picher ”2008), On the need for
bias correction of regional climate change projections of temperature and
precipitation, Geophys. Res. Lett., 35, L20709, doi:10.1029/2008GL035694.
Cannon, Alex J., Stephen R. Sobie, and Trevor Q. Murdock. "Bias correction of GCM
precipitation by quantile mapping: How well do methods preserve changes in
quantiles and extremes?." Journal of Climate 28.17 ”2015): 6938-6959.
Colette, A., Vautard, R., & Vrac, M. ”2012). Regional climate downscaling with prior
statistical correction of the global climate forcing. Geophysical Research Letters,
39”13).
Déqué, M. ”2007), Frequency of precipitation and temperature extremes over France
in an anthropogenic scenario: Model results and statistical correction according to
observed values, Global Planet. Change, 57, 16–26.
Ehret U. et al. ”2012) Should we apply bias correction to global and regional climate
model data?, Hydrol. Earth Syst. Sci., 16, 3391–3404.
Famien, A. M., Janicot, S., Ochou, A. D., Vrac, M., Defrance, D., Sultan, B., and Noël, T.: A
bias-corrected CMIP5 dataset for Africa using CDF-t method. A contribution to
agricultural impact studies, Earth Syst. Dynam. Discuss.,
https://doi.org/10.5194/esd-2017-111, in review, 2017.
Ficklin D. et al ”2016) The Influence of Climate Model Biases on Projections of Aridity
and Drought, Journal of Climate, 1269.
Galmarini, S. et al ”2018) Adjusting Climate Model Bias for Agricultural Impact
Assessment:
how to cut the mustard?, BAMS, submitted.
Giorgi, F., Jones, C., & Asrar, G. R. ”2009). Addressing climate information needs at the
regional level: the CORDEX framework. World Meteorological Organization ”WMO)
Bulletin, 58”3), 175.
Technical Report: Bias Adjusting Climate Model Projections | 17
Gudmundsson, Lukas, et al. "Technical Note: Downscaling RCM precipitation to the
station scale using statistical transformations–a comparison of methods." Hydrology
and Earth System Sciences 16.9 ”2012): 3383-3390.
Hay, L., R. Wilby, and G. Leavesley ”2000), A comparison of delta change and
downscaled GCM scenarios for three mountainous basins in the United States, J. Am.
Water Resour., 36, 387–397.
Häggmark L, Ivarsson KI, Gollvik S, Olofsson PO. 2000. MESAN, an operational
mesoscale analysis system. Tellus52A: 2–20, doi:10.1034/j.1600-0870.2000.520102.x.
Landelius, T., Dahlgren, P., Gollvik, S., Jansson, A., & Olsson, E. ”2016). A high-resolution
regional reanalysis for Europe. Part 2: 2D analysis of surface temperature,
precipitation and wind. Quarterly Journal of the Royal Meteorological Society,
142”698), 2132-2142.
Lavaysse, C., M. Vrac, P. Drobinski, M. Lengaigne, and T. Vischel ”2012), Statistical
downscaling of the French Mediterranean climate: Assessment for present and
projection in an anthropogenic scenario, Nat. Hazards Earth Syst. Sci., 12, 651–670,
doi:10.5194/nhess-12-651-2012.
Lenderink, G., A. Buishand, and W. van Deursen ”2007), Estimates of future discharges
of the River Rhine using two scenario methodologies: Direct versus delta approach,
Hydrol. Earth Syst. Sci., 11, 1145–1159, doi:10.5194/hess-11-1145-2007.
Lopez, Ana, et al. "From climate model ensembles to climate change impacts and
adaptation: A case study of water resource management in the southwest of
England." Water Resources Research 45.8 ”2009).
Maraun, D., 2016. Bias Correcting Climate Change Simulations - a Critical Review.
Current Climate Change Reports 2, 211–220.
Maurer, Edwin P. "Uncertainty in hydrologic impacts of climate change in the Sierra
Nevada, California, under two emissions scenarios." Climatic Change 82.3-4 ”2007):
309-325.
Michelangeli, P. A., Vrac, M., & Loukos, H. ”2009). Probabilistic downscaling
approaches: Application to wind cumulative distribution functions. Geophysical
Research Letters, 36”11).
Moss, R. H., Edmonds, J. A., Hibbard, K. A., Manning, M. R., Rose, S. K., Van Vuuren, D.
P., ... & Meehl, G. A. ”2010). The next generation of scenarios for climate change
research and assessment. Nature, 463”7282), 747-756.
Technical Report: Bias Adjusting Climate Model Projections | 18
Oettli, P., B. Sultan, C. Baron, and M. Vrac ”2011), Are regional climate models relevant
for crop yield prediction in West Africa?, Environ. Res. Lett, 6, 14008,
doi:10.1088/1748-9326/6/1/014008.
Passow, Christian, and Reik Donner. "Linear Regression Quantile Mapping ”RQM)-A
new approach to bias correction with consistent quantile trends." ”2017).
Switanek, M. B., Troch, P. A., Castro, C. L., Leuprecht, A., Chang, H.-I., Mukherjee, R.,
and Demaria, E. M. C.: Scaled distribution mapping: a bias correction method that
preserves raw climate model projected changes, Hydrol. Earth Syst. Sci. Discuss.,
doi:10.5194/hess-2016-435, in review, 2016.
Taylor, K. E., Stouffer, R. J., & Meehl, G. A. ”2012). An overview of CMIP5 and the
experiment design. Bulletin of the American Meteorological Society, 93”4), 485-498.
Tisseuil, C., Vrac, M., Grenouillet, G., Wade, A. J., Gevrey, M., Oberdorff, T., ... & Lek, S.
”2012). Strengthening the link between climate, hydrological and species distribution
modeling to assess the impacts of climate change on freshwater biodiversity. Science
of the total environment, 424, 193-201.
Vrac, M., P. Drobinski, A. Merlo, M. Herrmann, C. Lavaysse, L. Li, and S. Somot, 2012:
Dynamical and statistical downscaling of the French Mediterranean climate:
Uncertainty assessment. Nat. Hazards Earth Syst. Sci., 12, 2769–2784,
doi:https://doi.org/10.5194/nhess-12-2769-2012. Crossref
Vrac, M., T. Noël, and R. Vautard ”2016), Bias correction of precipitation through
Singularity Stochastic Removal: Because occurrences matter. J. Geophys. Res. Atmos.,
121, 5237–5258, doi:10.1002/2015JD024511.
Vrac, M.: Multivariate bias adjustment of high-dimensional climate simulations: The
Rank Resampling for Distributions and Dependences ”R2D2) Bias Correction, Hydrol.
Earth Syst. Sci. Discuss., https://doi.org/10.5194/hess-2017-747, in review, 2018.
Weedon, G. P., Balsamo, G., Bellouin, N., Gomes, S., Best, M. J., & Viterbo, P. ”2014). The
WFDEI meteorological forcing data set: WATCH Forcing Data methodology applied to
ERA-Interim reanalysis data. Water Resources Research, 50”9), 7505-7514.
Wood, Andrew W., et al. "Hydrologic implications of dynamical and statistical
approaches to downscaling climate model outputs." Climatic change 62.1 ”2004):
189-216.
Technical Report: Bias Adjusting Climate Model Projections | 19
 
Technical Report: Bias Adjusting Climate Model Projections | 20
6. Dataset and Document Revision History
Rev 0 - 1 February 2018 - Document created.
 
Technical Report: Bias Adjusting Climate Model Projections | 21
7. Annex
Table 1. CMIP5 experiment Global Circulation Models (GCM) for surface temperature
variable per RCP scenario (RCP4.5, RCP8.5) available on the climate data factory
Technical Report: Bias Adjusting Climate Model Projections | 22
GCM
RCM
RCP4.5
RCP8.5
CNRM-CM5
CLMcom-CCLM4-8-17
x
x
CNRM-CM5
SMHI-RCA4
x
EC-EARTH
CLMcom-CCLM4-8-17
x
x
EC-EARTH
DMI-HIRHAM5
x
x
EC-EARTH
KNMI-RACMO22E
x
x
EC-EARTH
SMHI-RCA4
x
x
HadGEM2-ES
CLMcom-CCLM4-8-17
x
HadGEM2-ES
KNMI-RACMO22E
x
HadGEM2-ES
SMHI-RCA4
x
IPSL-CM5A-MR
SMHI-RCA4
x
MPI-ESM-LR
CLMcom-CCLM4-8-17
x
x
MPI-ESM-LR
SMHI-RCA4
x
Table 2. CORDEX experiment Regional Climate Models (RCM) and their forcing GCM for
surface temperature variable, per RCP scenario (RCP4.5, RCP8.5) available on the climate
data factory
Technical Report: Bias Adjusting Climate Model Projections | 23
Cover photo:
Mountains and Clouds
by rachel_thecat / CC BY-SA 2.0
ResearchGate has not been able to resolve any citations for this publication.
Article
Full-text available
Climate simulations often suffer from statistical biases with respect to observations or reanalyses. It is therefore common to correct (or adjust) those simulations before using them as inputs into impact models. However, most bias correction (BC) methods are univariate and so do not account for the statistical dependences linking the different locations and/or physical variables of interest. In addition, they are often deterministic, while stochasticity is frequently needed to investigate climate uncertainty and to add constrained randomness to climate simulations that do not possess a realistic variability. This study presents a multivariate method of rank resampling for distributions and dependences (R²D²) bias correction allowing to adjust not only the univariate distributions, but also their inter-variable and inter-site dependence structures. Moreover, the proposed R²D² method provides some stochasticity since it can generate as many multivariate corrected outputs as the number of statistical dimensions (i.e., number of grid-cells × number of climate variables) of the simulations to be corrected. It is based on an assumption of stability in time of the dependence structure – allowing to deal with a high number of statistical dimensions –, that lets the climate model drive the temporal properties and their changes in time. R²D² is applied on temperature and precipitation reanalyses time series with respect to high-resolution reference data over South-East of France (1506 grid-cells). Bivariate, 1506-dimensional and 3012-dimensional versions of R²D² are tested over a historical period and compared to a univariate BC. How the different BC methods behave in a climate change context is also illustrated with an application to regional climate simulations over the 2071–2100 period. The results indicate that the 1d-BC basically reproduces the climate model multivariate properties, 2d-R²D² is only satisfying in the inter-variable context, 1506d-R²D² strongly improves inter-site properties and 3012d-R²D² is able to account for both. Applications of the proposed R²D² method to various climate datasets are relevant for many impact studies. The perspectives of improvements are numerous, such as introducing stochasticity in the dependence itself, questioning its stability assumption, and accounting for temporal properties adjustment while including more physics in the adjustment procedures.
Preprint
Full-text available
The objective of this paper is to present a new data set of bias-corrected CMIP5 global climate models (GCMs) daily data over Africa. This dataset was obtained in using the Cumulative Distribution Function Transform (CDF-t) method, a method that has been applied on several regions and contexts but never on Africa. Here CDF-t is used over the period 1950–2099 combining historical runs and climate change scenarios on 6 variables, precipitation, mean near-surface air temperature, near-surface maximum air temperature, near-surface minimum air temperature, surface down-welling shortwave radiation, and wind speed, which are critical variables for agricultural purposes. Evaluation of the results is carried out over West Africa on a list of priority users-based metrics that was discussed and selected with stakeholders and on simulated yield using a crop model simulating maize growth. Bias-corrected GCMs data are compared with another available dataset of bias-corrected GCMs, and the impact of three different reference datasets on bias-corrections is also examined in details. CDF-t is very effective in removing the biases and in reducing the high inter-GCMs scattering. Differences with other bias-corrected GCMs data are mainly due to the differences between the reference datasets. This is particular true for surface down-welling shortwave radiation, which has impacts in terms of simulated maize yields. Projections of future yields over West Africa have quite different levels, depending on bias-correction method used, but they all show a similar relative decreasing trend over the 21st century.
Article
Full-text available
Commonly used bias correction methods such as quantile mapping (QM) assume the function of error correction values between modelled and observed distributions are stationary or time-invariant. This article finds that this function of the error correction values cannot be assumed to be stationary. As a result, QM lacks justification to inflate/deflate various moments of the climate change signal. Previous adaptations of QM, most notably quantile delta mapping (QDM), have been developed that do not rely on this assumption of stationarity. Here, we outline a methodology called scaled distribution mapping (SDM), which is conceptually similar to QDM, but more explicitly accounts for the frequency of rain days and the likelihood of individual events. The SDM method is found to outperform QM, QDM and detrended QM in its ability to better preserve raw climate model projected changes to meteorological variables such as temperature and precipitation.
Article
Full-text available
Climate models are our major source of knowledge about climate change. The impacts of climate change are often quantified by impact models. Whereas impact models typically require high resolution unbiased input data, global and regional climate models are in general biased, their resolution is often lower than desired. Thus, many users of climate model data apply some form of bias correction and downscaling. A fundamental assumption of bias correction is that the considered climate model produces skillful input for a bias correction, including a plausible representation of climate change. Current bias correction methods cannot plausibly correct climate change trends, and have limited ability to downscale. Cross validation of marginal aspects is not sufficient to evaluate bias correction and needs to be complemented by further analyses. Future research should address the development of stochastic models for downscaling and approaches to explicitly incorporate process understanding.
Article
Full-text available
Global climate models (GCMs) have biases when simulating historical climate conditions, which in turn have implications for estimating the hydrological impacts of climate change. This study examines the differences in projected changes of aridity [defined as the ratio of precipitation (P) over potential evapotranspiration (PET), or P/PET] and the Palmer drought severity index (PDSI) between raw and bias-corrected GCM output for the continental United States (CONUS). For historical simulations (1950-79) the raw GCM ensemble median has a positive precipitation bias (124%) and negative PET bias (27%) compared to the bias-corrected output when averaged over CONUS with the most acute biases over the interior western United States. While both raw and bias-corrected GCM ensembles project more aridity (lower P/PET) for CONUS in the late twenty-first century (2070-99), relative enhancements in aridity were found for biascorrected data compared to the raw GCM ensemble owing to positive precipitation and negative PET biases in the raw GCM ensemble. However, the bias-corrected GCM ensemble projects less acute decreases in summer PDSI for the southwestern United States compared to the raw GCM ensemble (from 1 to 2 PDSI units higher), stemming from biases in precipitation amount and seasonality in the raw GCM ensemble. Compared to the raw GCM ensemble, bias-corrected GCM inputs not only correct for systematic errors but also can produce high-resolution projections that are useful for impact analyses. Therefore, changes in hydroclimate metrics often appear considerably different in bias-corrected output compared to raw GCM output.
Article
Six approaches for downscaling climate model outputs for use in hydrologic simulation were evaluated, with particular emphasis on each method's ability to produce precipitation and other variables used to drive a macroscale hydrology model applied at much higher spatial resolution than the climate model. Comparisons were made on the basis of a twenty-year retrospective (1975–1995) climate simulation produced by the NCAR-DOE Parallel ClimateModel (PCM), and the implications of the comparison for a future(2040–2060) PCM climate scenario were also explored. The six approaches were made up of three relatively simple statistical downscaling methods – linear interpolation (LI), spatial disaggregation (SD), and bias-correction and spatial disaggregation (BCSD) – each applied to both PCM output directly(at T42 spatial resolution), and after dynamical downscaling via a Regional Climate Model (RCM – at 1/2-degree spatial resolution), for downscaling the climate model outputs to the 1/8-degree spatial resolution of the hydrological model. For the retrospective climate simulation, results were compared to an observed gridded climatology of temperature and precipitation, and gridded hydrologic variables resulting from forcing the hydrologic model with observations. The most significant findings are that the BCSD method was successful in reproducing the main features of the observed hydrometeorology from the retrospective climate simulation, when applied to both PCM and RCM outputs. Linear interpolation produced better results using RCM output than PCM output, but both methods (PCM-LI and RCM-LI) lead to unacceptably biased hydrologic simulations. Spatial disaggregation of the PCM output produced results similar to those achieved with the RCM interpolated output; nonetheless, neither PCM nor RCM output was useful for hydrologic simulation purposes without a bias-correction step. For the future climate scenario, only the BCSD-method (using PCM or RCM) was able to produce hydrologically plausible results. With the BCSD method, the RCM-derived hydrology was more sensitive to climate change than the PCM-derived hydrology.
Article
The set-up and performance of the regional reanalysis for Europe with the HIgh-Resolution Limited-Area Model (HIRLAM) to a 3D grid-mesh with 22 km resolution for the years 1989-2010 have been presented in Part 1. This part describes how the 3D dataset is further downscaled and used as input for an analysis of a number of surface-related parameters: 2 m temperature, minimum and maximum daily temperatures, 10 m wind, and daily precipitation. The analysis is done on a 2D grid-mesh with 5 km grid spacing using the MESoscale ANalysis system (MESAN) for temperature and precipitation and a dynamical adaptation method (DYNAD) for the 10 m wind. Results from MESAN and DYNAD are compared with observations and the HIRLAM 3D-Var reanalysis. A couple of cases with severe weather are studied to illustrate how such events are represented in the analyses. The comparisons show statistically significant added value in comparison to the HIRLAM reanalysis.
Article
Quantile mapping bias correction algorithms are commonly used to correct systematic distributional biases in precipitation outputs from climate models. Although they are effective at removing historical biases relative to observations, it has been found that quantile mapping can artificially corrupt future model-projected trends. Previous studies on the modification of precipitation trends by quantile mapping have focused on mean quantities, with less attention paid to extremes. This article investigates the extent to which quantile mapping algorithms modify global climate model (GCM) trends in mean precipitation and precipitation extremes indices. First, a bias correction algorithm, quantile delta mapping (QDM), that explicitly preserves relative changes in precipitation quantiles is presented. QDM is compared on synthetic data with detrended quantile mapping (DQM), which is designed to preserve trends in the mean, and with standard quantile mapping (QM). Next, methods are applied to phase 5 of the Coupled Model Intercomparison Project (CMIP5) daily precipitation projections over Canada. Performance is assessed based on precipitation extremes indices and results from a generalized extreme value analysis applied to annual precipitation maxima. QM can inflate the magnitude of relative trends in precipitation extremes with respect to the raw GCM, often substantially, as compared to DQM and especially QDM. The degree of corruption in the GCM trends by QM is particularly large for changes in long period return values. By the 2080s, relative changes in excess of +500% with respect to historical conditions are noted at some locations for 20-yr return values, with maximum changes by DQM and QDM nearing +240% and +140%, respectively, whereas raw GCM changes are never projected to exceed +120%.