ESSDEarth System Science DataESSDEarth Syst. Sci. Data1866-3516Copernicus PublicationsGöttingen, Germany10.5194/essd-9-471-2017Using ERA-Interim reanalysis for creating datasets of energy-relevant
climate variablesJonesPhilip D.p.jones@uea.ac.ukhttps://orcid.org/0000-0001-5032-5493HarphamColinTroccoliAlbertoGschwindBenoitRanchinThierryWaldLucienhttps://orcid.org/0000-0002-2916-2391GoodessClare M.DorlingStephenClimatic Research Unit (CRU), School of Environmental Sciences,University of East Anglia, Norwich, NR4 7TJ, UKSchool of Environmental Sciences, University of East Anglia, Norwich, NR4 7TJ, UKMINES ParisTech, PSL Research University, O.I.E. – Centre Observation, Impacts, Energy, 06904 Sophia Antipolis, FranceCenter of Excellence for Climate Change Research, Department of Meteorology, King Abdulaziz University, Jeddah, Saudi ArabiaWorld Energy & Meteorology Council (WEMC), Norwich, NR4 7TJ, UKPhilip D. Jones (p.jones@uea.ac.uk)21July20179247149516December20166January201711May201725May2017This work is licensed under the Creative Commons Attribution 3.0 Unported License. To view a copy of this licence, visit https://creativecommons.org/licenses/by/3.0/This article is available from https://essd.copernicus.org/articles/9/471/2017/essd-9-471-2017.htmlThe full text article is available as a PDF file from https://essd.copernicus.org/articles/9/471/2017/essd-9-471-2017.pdf
The construction of a bias-adjusted dataset of climate variables at the near
surface using ERA-Interim reanalysis is presented. A number of different,
variable-dependent, bias-adjustment approaches have been proposed. Here we
modify the parameters of different distributions (depending on the variable),
adjusting ERA-Interim based on gridded station or direct station
observations. The variables are air temperature, dewpoint temperature,
precipitation (daily only), solar radiation, wind speed, and relative
humidity. These are available on either 3 or 6 h timescales over the period
1979–2016. The resulting bias-adjusted dataset is available through the
Climate Data Store (CDS) of the Copernicus Climate Change Data Store (C3S)
and can be accessed at present from ftp://ecem.climate.copernicus.eu.
The benefit of performing bias adjustment is demonstrated by comparing
initial and bias-adjusted ERA-Interim data against gridded observational
fields.
Introduction
Climate/weather information has been widely used in a number of
climate-related impact sectors (e.g. agriculture, water, and energy) for
decades. Increasingly, users are moving beyond the use of station
observations to the use of gridded products, especially meteorological
reanalysis datasets. These are reconstructions of past climate produced
through the blending of observations with physical/numerical models which
have been developed explicitly for climate monitoring and research (Compo et
al., 2011; Dee et al., 2011; Hersbach et al., 2015). How good ERA-Interim is
for climate monitoring has been extensively addressed recently by Simmons et
al. (2017). This study shows excellent agreement for global- and
continental-scale trends in surface air temperatures over land with
conventional station-based datasets (see Jones, 2016, for details of these
datasets).
Reanalyses have the specific advantage of being spatially and temporally
complete through the process of physical/dynamic representation of the
climate system which provides internally consistent fields across most
surface atmospheric variables as well as in the atmospheric column up to the
stratosphere (Compo et al., 2011). The present paper deals with the use of
reanalysis for the production of datasets of climate variables relevant to
the energy sector. The work took place within the European Climatic Energy
Mixes (ECEM) project, a Sectoral Information Service (SIS) part of the
Copernicus Climate Change Service (C3S). This project is primarily focused on
users in the energy sector who are interested in sub-daily (e.g. 6 h)
and daily variability for the following variables at the near surface: air
temperature, dewpoint temperature, precipitation, solar radiation, wind
speed, and relative humidity. Despite this choice of variables being of primary
relevance to the energy sector, it is likely that the results will also be of
use to other sectors (particularly water and agriculture).
Because reanalyses are computed on a model grid, inevitably there will be
differences when they are compared to station observations. Differences are not solely
related to scales: reanalyses are dependent on the underlying
weather-forecast model and the amount of observational data entering the
assimilation system used to produce them (see example fields given in Dee et
al., 2011, and we expand on this in Sect. 2.6). Many users of reanalysis
products attempt to adjust them to observational distributions through a
process that is referred to using different terminology: bias adjustment and
calibration being the most commonly used terms (Maraun et al., 2010). Here,
the term bias adjustment is used.
The principal reason for performing a bias adjustment is that reanalyses are
potentially biased compared to direct station observations (even when the
station observations are gridded to a comparable spatial resolution), more
so for some variables than others (e.g. precipitation compared to
temperature), and the bias may also vary in value, space, and time; i.e. the
bias may be larger for more extreme values or it might be larger for regions
or time periods of sparse station coverage. The importance of the bias
depends to a large extent on how the data will be used. For some variables,
the monthly average/totals will be important, but many other users require
that extremes of the distribution be well simulated. With time, the
complexity of approaches to bias adjustment has developed from getting the
monthly averages correct to the present attempts to adjust the whole
distribution and to even account for the multivariate relationships between
some variables (see, e.g., Vrac and Friederichs, 2015). These advances reflect
not only the greater expectations with each generation of reanalysis but
also the greater number of users in a greater number of sectors.
A widely used bias-adjustment dataset was developed in the WATCH project
(Weedon et al., 2010, 2011, 2014). The methodology applied to ERA-40
reanalysis data to create the WATCH Forcing Data (Weedon et al., 2010, 2011)
was used later with ERA-Interim data to produce the WFDEI dataset (WATCH
Forcing Data methodology applied to ERA-Interim data; Weedon et al., 2014).
Bias adjustment in WFDEI was undertaken on the monthly average scale for a
number of hydrological variables necessary to calculate evapotranspiration,
soil moisture, and runoff (so including air temperature, rainfall, snowfall,
long-wave and short-wave solar radiation, wind speed, specific humidity, and
surface pressure) and for the period of analysis 1958–2001 (1979–2014)
based on the ERA-40 (ERA-Interim) reanalysis. The dataset was developed for
forcing land surface models using meteorological data (bias-adjusted
reanalysis) and the WATCH project built on earlier work (Cosgrove et
al., 2003; Sheffield et al., 2006), which also developed forcing datasets.
The spatial coverage for WFDEI is all land areas north of latitude
60∘ S. ECEM is less spatially extensive covering the European Domain
(27–72∘ N, 22∘ W–45∘ E). The current period of
study is 1979–2016 based on the ERA-Interim reanalysis with sub-daily and
daily timescales.
The aim of this paper is to present the construction of a sub-daily
bias-adjusted dataset of the climate variables listed above, by using
ERA-Interim reanalysis. The ECEM dataset is freely available through the Climate Data Store
(CDS) of C3S (currently ftp://ecem.climate.copernicus.eu). The benefit of
performing bias adjustment is demonstrated by comparing initial and
bias-adjusted data against station observations and gridded observation
products. The ERA-Interim reanalysis and the gridded and station
observation-based datasets used for bias adjustment are described in Sect. 2.
Section 3 provides more information on the methods for bias adjustment on the
daily and sub-daily timescales with a focus on the specific context of the
energy sector. The selected techniques are discussed in Sect. 4. Section 5
discusses issues related to whether our bias adjustment is applicable to
other sectors. Different sectors have different user demands relating to the variables required, timescales, and the length of historical reanalysis data
needed. Section 6 gives details of dataset access.
Data
This section provides details of ERA-Interim and the various gridded and
station observation datasets used to assess the quality of this reanalysis.
With gridded datasets, the spatial resolutions may vary, so it is often
necessary to regrid data onto a common resolution (in this study a grid of
0.5∘×0.5∘ latitude×longitude).
ERA-Interim
The development of ERA-Interim is described by Dee et al. (2011). Surface air
temperature, precipitation, wind speed at 10 m, surface downwelling
solar irradiance, and relative humidity data were extracted from ERA-Interim
on its reduced Gaussian grid. The period is 1979–2016, and the temporal
resolution is either 3 h (forecast) or 6 h (analysis),
depending on the variable (see Dee et al., 2011, for details). These five are
Essential Climate Variables (ECVs) defined by the Global Climate Observing
System (Bojinski et al., 2014). After extraction, the variables have been
regridded onto a latitude×longitude grid of 0.5∘×0.5∘
for the ECEM domain using a bilinear interpolation technique. There are two
principal reasons for this regridding: (i) some of the observation datasets
for the assessment of ERA-Interim are available on this regular
latitude×longitude grid (e.g. E-OBS; see next section), and (ii) potential
users of the datasets developed here requested regular latitude×longitude
grids with cells size of 0.5∘ for practical reasons (in particular
for aggregation to the country scale). It is also preferable to regrid a
dataset without missing values, as opposed to an observation-based gridded
product as these can contain missing values when some station data were not
available.
Gridded observation datasets
Among the available gridded products for air temperature and precipitation,
we used
E-OBS for both variables (http://www.ecad.eu/,
Haylock et al., 2008),
CRU TS for both variables (CRU TS 3.23,
https://crudata.uea.ac.uk/cru/data/hrg/, Harris et al., 2014) and
GPCC (Global Precipitation Climatology Centre) for precipitation
(https://www.dwd.de/EN/ourservices/gpcc/gpcc.html, Becker et
al., 2013).
E-OBS, CRU TS, and GPCC data were downloaded for the ECEM grid. All three
datasets only cover land regions, so any bias adjustment using these datasets
will not include marine areas. E-OBS covers the period from 1951 to 2016, so
fully encompassing the 1979–2016 period of ERA-Interim. CRU TS and GPCC
cover the period from 1901 to 2016 (up to 2013 for GPCCv5) but are both
monthly averages/totals, so can only provide an assessment on this timescale.
These additional two monthly gridded datasets are included as they are used
by the WATCH/WFDEI (Weedon et al., 2011, 2014) dataset, and we will compare
our bias-adjusted ERA-Interim dataset with this dataset in Sect. 4.5.
HadISD
No gridded observed product is available for wind speed and dewpoint
temperature. Dewpoint temperature is necessary as it can be combined with air
temperature to calculate relative humidity, which is needed for energy
calculations, such as demand. Station data for wind speed at 10 m height and
dewpoint temperature were extracted from HadISD
(http://www.metoffice.gov.uk/hadobs/hadisd/) for approximately
1500 stations across Europe. Station data were extracted every 6 h at
the SYNOP hours 00, 06, 12, and 18 for the period 1979–2014 (Smith et
al., 2011; Dunn et al., 2012). We additionally extracted air temperature data
from HadISD, so we could use this with the concurrent dewpoint temperatures
to calculate dewpoint depression (see later in Sect. 4.2). HadISD has
additionally been assessed for long-term homogeneity by Dunn et al. (2014).
Variations in station coverage within HadISD are considerably greater than
the coverage achieved for air temperature from E-OBS and precipitation from
E-OBS and GPCC. This indicates that it would be unwise to attempt spatial
interpolation to a 0.5∘×0.5∘ grid using the
HadISD
stations. Instead each station series will be compared with that from the
nearest ERA-Interim grid-box series.
Surface solar irradiance from the World Radiation Data Center
and the Baseline Surface Radiation Network
National meteorological services (NMSs) usually measure surface solar
irradiance at a limited number of sites. Data are sent to the World Radiation
Data Center (WRDC), a laboratory of the Voeikov Main Geophysical Observatory
in Saint-Petersburg, Russia, under the control of the World Meteorological
Organization (WMO). There, the data are archived and published
(http://wrdc.mgo.rssi.ru). Most of the data are daily irradiation;
hourly (or higher-frequency) irradiation is available at very few sites. All
data are scrutinized at WRDC and quality-flagged before entering archives.
Additionally, six stations were added from the Baseline Surface Radiation
Network (BSRN, http://bsrn.awi.de/). Altogether, 55 stations with
high-quality daily irradiation data were kept, for which mean daily
irradiance was computed.
HelioClim-3v5 (HC3v5)
Boilley and Wald (2015) have shown the need to correct ERA-Interim estimates
of solar irradiance. As only a limited number of stations are available for
solar irradiance over Europe (but also globally), it was decided to exploit
the satellite-derived HelioClim-3v5 (HC3v5) dataset to correct ERA-Interim.
HC3v5 originates from the daily processing of images acquired by the series
of satellites Meteosat-MSG by the Heliosat-2 method (Blanc et al., 2011;
Rigollier et al., 2004). In version 5 of HelioClim-3, a correcting table was
developed in 2015 between 15 min estimates made by HelioClim-3 and
data from the six BSRN stations. It has been established by merging all data;
i.e. it is a global correction and not a local one. Inputs to the correcting
table are the solar zenith angle and the HelioClim-3 estimates; there is no
local input. In this respect, HC3v5 is independent of the surface station
data in BSRN and also data from WRDC.
HC3v5 does not cover the northern part of the ECEM domain. The first
estimates began on 1 February 2004, and these have been compared
satisfactorily with measurements taken at ground stations (Eissa et
al., 2015; Thomas et al., 2016a, b; Marchand et al., 2017). HC3v5 data were
downloaded from the SoDa (Solar Radiation Data) Service website
(www.soda-pro.com) from which one may select the timescale, here the
daily mean of irradiance I. The HC3v5 product comprises the irradiance at
the top of atmosphere E0, from which one may compute the clearness index
KT:
KT=IE0.KT is a good indicator of the optical state of the atmosphere
with a dependency on the position of the sun that is much less pronounced than in
I. KT greater than 0.7 signifies a clear sky, while
KT less than 0.2 signifies an overcast sky. The advantage of
HC3v5 over the station series is that it provides gridded data for the ECEM
domain up to 60∘ N. The independent station data from WRDC and BSRN
will be used in Sect. 4.4 for assessing the performance of the bias
adjustment for solar irradiance.
Independence of the station/gridded observation series versus
ERA-Interim
In this study, we propose bias-adjusting ERA-Interim for the five ECVs (wind
speed, air temperature, dewpoint temperature, precipitation, and irradiance).
As stated in Sect. 2.1, ERA-Interim assimilates many different climate
datasets: surface station data are just one set of several; satellite and
radiosonde data are also assimilated. In this section, we discuss how
independent the station observations and gridded products are that are used
in this bias adjustment compared to the surface station data assimilated into
ERA-Interim. Precipitation and irradiance data are totally independent as
these data are not assimilated. These variables are forecast outputs from
ERA-Interim (see Dee et al., 2011). Of the other three variables,
near-surface air and dewpoint temperatures are assimilated. For wind, the u
and v components of the 10 m wind speeds are assimilated. It is important
to understand what is assimilated and what importance may be given to these
variables. The output for these three variables used is their value in the
analysis (referred to as an analysis variable), produced every 6 h.
ERA-Interim does not provide details of all the specific station data (and
additional satellite and radiosonde data) that are assimilated. Dee et
al. (2011) give details of what datasets are available for assimilation.
ERA-Interim provides a dynamically consistent estimate of the climate state
at each 6 h time step, but it does not specifically give any details of
which potential information was used to produce the analysis variables.
Through dynamical consistency, information from satellites, radiosondes, and
other surface variables (e.g. pressure) are also used. Essentially, the
quantity of surface station data for Europe is similar to that available in
the HadISD database, which we know is about 1500 series, but only about 800
are relatively complete over the 1979–2014 period (Dunn et al., 2012, 2014).
Thus, for air temperature, the ∼ 2000 additional daily Tx and Tn
observations are not assimilated. So our E-OBS dataset for air temperature
contains a much greater volume of additional temperature series than
assimilated within ERA-Interim. The wind speed and dewpoint temperature from
HadISD should have been available for assimilation, but the importance given
to these observations is not as great as the importance given to the station
pressure observations.
The production of a reanalysis has occasionally been referred to as dynamic
infilling, which is quite different from the statistical spatial infilling
techniques that are used to produce the E-OBS, CRU TS, and GPCC datasets.
Spatial infilling techniques use a variety of statistical procedures (e.g.
inverse distance weighting and kriging) and are generally applied for each
variable independently of other variables. In data-sparse regions,
statistical-infilling techniques will likely spread information from the few
available stations across the unobserved areas. The effects of this are
generally evident as reduced variance in the generated fields. In contrast, a
reanalysis will make use of additional information (e.g. the large-scale
circulation and satellite information), potentially not placing great
emphasis on a specific observed variable (e.g. wind observations). In
addition, balances of mass, wind, and energy fields mean that consistency
between different variables is ensured, though this is particularly the case
for forecast variables at a few to several hours lead time. At analysis time,
such balances might be not guaranteed, but this depends on the specific data
assimilation scheme used and whether the scheme enforces physical/dynamical
balances.
Bias-adjustment approaches
Bias adjustment and bias correction are widely used terms for the assessment
of climate model output (from both global and regional climate models, GCMs
and RCMs; see, e.g., Maraun et al., 2010; Maraun, 2012) generally through
comparison with station observational data. In this context, the biases
compared to observations, are often much larger than differences with recent
reanalysis products. There are a number of studies where GCMs and RCMs are
bias adjusted against reanalyses, so the assumption is made there that
reanalyses are a true representation of the real climate. This happens more
in regions where observational datasets are sparse and/or hard to access
(Oyerinde et al., 2017, use the MERRA reanalyses, Rienecker et al., 2011, for air temperature, when
bias-adjusting RCM simulations for western Africa). Bias adjustment of
reanalyses has been undertaken for a number of years, though. An extensive
exercise was carried out by the WATCH project
(http://www.eu-watch.org/, see Weedon et al., 2011, 2014). This used
the CRU TS and GPCC datasets as the basis for adjusting ERA-40 and
ERA-Interim, and the adjustments are based on average monthly differences
treating each variable independently of each other, as we will do.
Numerous and more complex (than Weedon et al., 2014) methods for bias
adjusting climate variables derived from climate models have been proposed. A
number of review papers have been published (e.g. Maraun et al., 2010;
Maraun, 2013; Vrac and Friedrichs, 2015). Among the various possibilities are
the cumulative distribution function (CDF) transform method of Vrac et
al. (2012), the distribution-based scaling (DBS) method of Yang et
al. (2010), empirical quantile mapping (Themeßl et al., 2011, 2012;
Wilcke et al., 2013), and using the R package “qmap” used by MetNorway
(Gudmundsson et al., 2012). Unlike the bias adjustment within the WATCH
project, the latest examples from the literature attempt to address the
issues of spatial dependence of the bias (any bias in ERA-Interim for a
variable is expected to be relatively smooth) and temporal dependence (biases
may be greater for certain types of weather, which has led to the approaches
improving the fit between the distributions), and some attempt to adjust
climate variables in a multivariate way (e.g. Vrac and Friedrichs, 2015).
Research in the literature has tended to emphasize precipitation (where bias
adjustment can also be classed as a form of downscaling). In ECEM,
precipitation is less important, with instead a greater emphasis on wind
speed and solar irradiance as well as temperature. As stated earlier, how
good bias adjustment has to be depends on how the adjusted data will be used.
Within ECEM, techniques were selected to be fit for purpose, and that purpose
is energy sector applications. Even though users in the energy sector are a
diverse group, they are mainly interested in only one or two variables, and
our initial determination of their needs indicated that univariate bias
adjustment will be sufficient.
Bias adjustment and results
In the present work, the same univariate approach as Weedon et al. (2014)
was followed, and ERA-Interim was compared against the gridded observational
products on the monthly timescale. The bias was computed as the mean of the
differences (model minus observations). For both temperature and
precipitation (not shown), differences are generally greater (but variable
in sign) over mountainous regions and some coastal areas (the Norwegian
coast for temperature and most west-facing coasts for precipitation). Users
in the energy sector are much more interested in the extremes of the
distribution, so the approach moved to adjusting the whole ERA-Interim
distribution on the daily and sub-daily timescales, using a different
statistical distribution for each variable. The following sections begin
with wind speed, then move to air and dewpoint temperature, then
precipitation, and finally a new approach entirely for solar radiation.
Wind speed at 10 m
In this section, results from the univariate bias adjustment are presented
starting with wind speed at 10 m. For use in the energy sector, wind
speeds at hub heights (80–120 m) are potentially more useful, but
assessing ERA-Interim wind speeds from these heights is only possible at a
limited number of masts (Harpham et al., 2016). Assessment over the whole
domain is only possible using surface station measurements which measure wind
speeds at 10 m. The two-parameter Weibull distribution is the
most-used probability distribution for representing wind speeds and is of
strong relevance in the energy sector. The Weibull distribution, with scale
parameter α>0 and shape parameter β>0, has a cumulative
distribution function for x>0 given by
PrX≤x=Fx;α,β=1-exp[-xαβ].
The scale parameter α relates to the mean wind speed, and β
characterizes the skewness of the distribution; typical values of β
range between 1 (highly variable wind speed) and 3 (fairly constant wind
speed). The 2-parameter Weibull distribution was fitted to 6 h wind speed
data from ERA-Interim on a monthly basis, i.e. a separate fit was made for
each month of the year, for each grid box using all the 6 h data for
1981–2010, irrespective of the wind direction. The same approach was applied
for the wind data from 803 stations in the HadISD dataset that have at least
66.6 % data completeness for this 30-year period. The scale and shape
parameters (α, β) for the 803 stations were compared with the
same parameters from the nearest ERA-Interim grid box. Figure 1 shows
differences (ERA-Interim minus observations) between the scale and shape
parameters for January across the European domain. The maps indicate
generally good agreement for January, i.e. the values for the two parameters
are generally within ±1 of each other. Exceptions may be found in some
mountainous regions and around west-facing coasts but this is very dependent
on the month (larger differences when wind speeds are stronger). The
similarity of the two distributions in terms of their scale and shape
parameters indicates that bias adjustment could be achieved by replacing the
ERA-Interim scale and shape parameters with those inferred from the HadISD
stations.
Differences in scale and shape parameters of the Weibull
distribution between ERA-Interim and HadISD station observations for wind
speed at 10 m. Based on all 6-hourly data for January for
1981–2010.
Comparison of statistical distributions of wind speed at
10 m for Kirkwall, Scotland, for observations (black), ERA-Interim
(orange), and bias-adjusted ERA-Interim (green), based on all 6-hourly data
for the 1981–2010 period.
Comparison of statistical distributions of wind speed at
10 m for Maribor, Slovenia, for observations (black), ERA-Interim
(orange), and bias-adjusted ERA-Interim (green), based on all 6-hourly data
for the 1981–2010 period.
Differences in scale and shape parameters of the Weibull
distribution between bias-adjusted ERA-Interim and HadISD station
observations for wind speed at 10 m. Based on all 6-hourly data for
January for 1981–2010.
Differences in means and standard deviations (SDs) between
ERA-Interim and E-OBS for mean surface air temperature (Tmean). Based on
daily data for April for 1981–2010.
Comparison of statistical distributions of surface air temperature
for northern Scotland (58.25∘ N, 3.75∘ W), for
observations (black), ERA-Interim (orange), and bias-adjusted ERA-Interim
(green), based on daily data for the 1981–2010 period.
Comparison of statistical distributions of surface air temperature
for the Maribor grid box (46.25∘ N, 15.75∘ E), for
observations (black), ERA-Interim (orange), and bias-adjusted ERA-Interim
(green), based on daily data for the 1981–2010 period.
Differences in means and standard deviations (SDs) between
bias-adjusted ERA-Interim and E-OBS for mean surface air temperature (Tmean).
Based on all data for April for 1981–2010.
Differences in means and standard deviations (SDs) between
ERA-Interim and HadISD for dewpoint temperature (∘C). Based on daily
data for July for 1981–2010.
Comparison of statistical distributions of DPD for Kirkwall, for HadISD observations (black), ERA-Interim (orange), and
bias-adjusted ERA-Interim (green), based on daily data for the 1981–2010
period.
Comparison of statistical distributions of DPD for Maribor, for HadISD observations (black), ERA-Interim (orange), and
bias-adjusted ERA-Interim (green), based on daily data for the 1981–2010
period.
Differences in means and standard deviations (SDs) between
bias-adjusted ERA-Interim and HadISD for dewpoint temperature (∘C).
Based on daily data for July for 1981–2010.
Equation (3) of Tye et al. (2014) provides a means to adjust the original
variable X into a variable X* having scale and shape parameters
α* and β* by the following power-law transfer function:
X*=α*Xαββ*.
Where stations are available, α* and β* are those of the
stations. The scale and shape parameters computed at stations were
interpolated to each ERA-Interim grid box with the bilinear INTERP function
within the R Akima software package. A bias-adjusted dataset of wind speeds
for ERA-Interim is obtained by applying Eq. (3). Figures 2 and 3 exhibit the
smoothed distributions using the HadISD observations, original ERA-Interim, and bias-adjusted ERA-Interim for the 12 calendar months for the stations
Kirkwall, Scotland, and Maribor, Slovenia. These two locations were chosen as
one is maritime and the other more continental. The other 801 distributional
fits are shown on the ftp site with the unadjusted and adjusted ERA-Interim
grids (ftp://ecem.climate.copernicus.eu). The smoothed representations
of the distributions are curtailed below zero wind speed. It is clear from
these two examples that the distributional fit for the stations has moved the
adjusted ERA-Interim data series towards the observational distribution, more
so for Kirkwall, which shows a much greater improvement than for Maribor,
where the distribution moves are a clear improvement in winter months but
less so for spring and early summer months. Bias adjustment is less
successful than in the examples shown for a few stations located in coastal
areas and a few sites in mountainous regions (see the full set of
distributional fits on the ftp site). Some observed distributions are a
little erratic due to some years in the observed data having wind speeds
rounded to integer values. Similarly to Fig. 1, but for bias-adjusted
ERA-Interim minus observations, Fig. 4 shows differences between the scale
and shape parameters for January. Almost all stations across Europe exhibit
similar shape and scale parameters between the stations and adjusted
ERA-Interim. However, a few stations in coastal areas and at high-elevation
mountain locations still show differences in parameters. The fit is not
perfect as the estimation of the shape and scale parameters for the
ERA-Interim grid boxes from HadISD is influenced by the station distribution.
In addition, the number of stations in some parts of Europe is less dense and so
involves greater extrapolation from stations more distant from the grid
boxes.
Surface air temperature, dewpoint temperature, and relative
humidity
Like wind speed, both surface air temperature and dewpoint temperature are
produced from ERA-Interim every 6 h. Unlike 10 m wind, both these
variables have a strong diurnal cycle, which is generally slightly stronger
in the summer. A normal distribution was fitted using daily averages of
temperature, taking the average of the four 6 h data for each day. E-OBS is
the dataset on which ERA-Interim is to be adjusted for air temperature. For
dewpoint we use HadISD, but we combine this with air temperatures from HadISD
to calculate dewpoint depression (DPD), the difference between air and
dewpoint temperature. To calculate DPD, we need to pair off air and dewpoint
temperature measurements taken every 6 h. DPD will always be
≥ 0, so we use a Weibull distribution to ensure that any bias
adjustments always produce a DPD that is ≥ 0. Means and standard
deviations of the daily average of air temperature are calculated for each
month of the year for each 0.5∘ grid cell of ERA-Interim coincident
with the E-OBS grid box. The distributional parameters for DPD are
interpolated as for wind speed; then, based on air temperature, a dewpoint
temperature can be calculated. Data are normalized as in Eq. (4) and
transformed back by Eq. (5) for air temperature.
TERA′=TERA-T‾ERAσERA,T*=TERA′σobs+T‾obs,
where T′ is the normalized ERA-Interim temperature anomaly, T* is the
bias-adjusted ERA-Interim temperature, T‾ is the mean temperature, and
σ is the standard deviation. Bias adjustment works by transforming the
normalized ERA-Interim grid-box time series back to air temperatures using
the means and standard deviations from E-OBS and interpolations from station
data in HadISD for DPD. Once daily averages are adjusted, the difference
between the original ERA-Interim daily mean and the adjusted daily mean is
added to each of the four 6 h temperatures within each day. Therefore, no
alteration is made to each diurnal cycle of air temperature or DPD. This
yields the final set of bias-adjusted 6 h surface air and dewpoint
temperatures (the latter calculated from DPD) consistent with one another.
Figure 5 shows the differences in the mean and standard deviation for air
temperature for April as an example. There is good agreement between
estimates for ERA-Interim and those calculated from E-OBS. As these are both
gridded datasets, the maps shown are fully coloured for each 0.5∘
grid box. Differences in Fig. 5 are likely related to elevational differences
in the 0.5∘ grid boxes between E-OBS and ERA-Interim. The
significance of this is discussed more in Sect. 4.5 when our adjusted
ERA-Interim and the WFDEI datasets are separately compared with E-OBS.
Figures 6 and 7 exhibit the distributional fits of the E-OBS, original
ERA-Interim, and bias-adjusted ERA-Interim for the 12 calendar months for
the nearest land grid boxes that approximate the locations of Kirkwall and
Maribor used for wind speed. For Maribor, this is the 0.5∘ grid box
where the city is located. Kirkwall is on the Orkney Islands, so the nearest
grid box within E-OBS is located further south in northern Scotland. The
distributional fits for the Maribor grid box were good for ERA-Interim, and
bias adjustment brings minor improvement. For the Kirkwall grid box, the
adjustments improve the fits in all months but are less good for the cold
tail of air temperature in winter. The full set of results for the 4621
grid-box comparisons can be viewed on the ftp site. The Kirkwall example and
also most of the northern and eastern Europe comparisons illustrate an issue
with approximating daily air temperature by a normal distribution for winter.
For these regions (see the full set of results), daily winter air
temperatures are often negatively skewed. This is even apparent in monthly
temperatures but is more clearly visible on the daily timescale. Horton et
al. (2001) experimented with an alternate distribution (an inverted gamma
distribution), but this adds an additional parameter to any bias-adjustment
approach and fitting requires an interactive procedure. Figure 8 shows the
differences between the means and standard deviations, but this time after
adjustment, to allow comparison with Fig. 5.
Figure 9 shows the differences in the Weibull distribution parameter plots
for DPD for July. There is good agreement between estimates from ERA-Interim
and those calculated from HadISD using DPD. This plot shows the station
locations in a similar fashion to that for wind speed in Fig. 1. Figures 10
and 11 show the distributional fits of the HadISD, original ERA-Interim, and
bias-adjusted ERA-Interim for DPD for the 12 calendar months for the
locations of Kirkwall and Maribor. Both examples of distributional plots
adjust ERA-Interim slightly, but the original fits were quite good to start
with. As with wind speed, the DPD distributions are curtailed for
values < 0. Similarly to Fig. 9, Fig. 12 shows the differences between
the Weibull distribution parameters for DPD but after adjustment.
With the adjustments for dewpoint temperature using DPD, it is a simple task
to then calculate relative humidity (RH) using the adjusted air temperature.
Performing the bias adjustment this way, we are assured that all RH values
are between 0 and 100 %.
Daily precipitation totals
The same process was then used for daily precipitation totals but using a
gamma distribution, which has been found to perform well in many studies
(e.g. Wilks, 1995).
Fx;α,β=xβα-1exp-xββΓ(α)
Gamma distributions have two parameters, shape (α) and scale
(β), and were fit to the daily precipitation totals for each month for
ERA-Interim and for E-OBS. In Eq. (6), we show the probability density
function where Γ is the gamma function. Approaches to bias adjustment
of precipitation have been extensively discussed (see, for example, Piani et
al., 2010a, b). Several experiments were made ignoring all precipitation
values below a fixed low daily precipitation threshold over the whole domain.
Thresholds of 0.4, 0.6, 0.8, and 1.0 mm were experimented with and
best fits were achieved with 1.0 mm. This implies that the gamma
distributional fits are based only on days with precipitation values greater
than the threshold, with a different fit for each month. This threshold
ignores small precipitation totals, more so for ERA-Interim than for E-OBS
but, as both datasets are in essence areal averages, more than would be the
case for a station rain gauge series. In the adjusted ERA-Interim all
precipitation amounts below the threshold are set to zero, further improving
the agreement between E-OBS and ERA-Interim in the number of dry days per
month (i.e. days with rainfall less than the 1.0 mm threshold). Adjustment
is performed in a similar way to the temperatures, by back-transforming the
transformed ERA-Interim precipitation total with the scale and shape
parameters from the E-OBS dataset.
Figure 13 shows the differences in the scale and shape parameters of the
gamma distribution for October, by way of example. There is good agreement
between estimates for ERA-Interim and those calculated from E-OBS. As these
are both gridded datasets, the maps shown are fully coloured for each
0.5∘ grid box. Of the 4520 possibilities, Figs. 14 and 15 exhibit
the distributional fits of the E-OBS, original ERA-Interim, and bias-adjusted
ERA-Interim datasets for the 12 calendar months for the nearest land grid
boxes that approximate the locations of Kirkwall and Maribor. The fits for
the northern Scotland grid box are considerably better than for Maribor,
where the distributional fits are slightly worse for April–August. Here
only two locations are shown as examples: the full set of results for the
4520 grid-box comparisons can be viewed at the ftp site. The complete set
uses common scaling, which may be inappropriate in drier parts of Europe and
the fits (shown in the two examples in Figs. 14 and 15 are smoothed
representations of the distributions curtailed at zero rainfall). Although
the gamma distribution is widely used for rainfall data, it is not ideal in
all climates and across all seasons in Europe. Problems arise when there are
too few rainfall days within dry seasons (the southern Mediterranean and the
Middle East during summer). Similarly to Fig. 13, Fig. 16 shows the
differences between the scale and shape parameters after adjustment.
Differences in scale and shape parameters of the gamma distribution
between ERA-Interim and E-OBS for precipitation daily
totals > 1 mm. Based on daily precipitation totals for October
for 1981–2010.
Comparison of statistical distributions of daily precipitation
totals for northern Scotland (58.25∘ N, 3.75∘ W), for observations (black), ERA-Interim (orange), and
bias-adjusted ERA-Interim (green), based on the 1981–2010 period.
Comparison of statistical distributions of daily precipitation
totals for the Maribor grid box (46.25∘ N, 15.75∘ E), for observations (black), ERA-Interim (orange), and
bias-adjusted ERA-Interim (green), based on the 1981–2010 period.
Differences in scale and shape parameters of the gamma distribution
between bias-adjusted ERA-Interim and E-OBS for precipitation daily
totals > 1 mm. Based on daily precipitation totals for October
for 1981–2010.
Bias for ERA-Interim vs. ground observations of daily mean of
surface solar irradiance for 57 stations. Downward triangles mean a negative
bias less than -5 Wm-2, upward triangles mean a positive bias
greater than 5 Wm-2, and circles mean an absolute value of the
bias less than 5 Wm-2. The size of the triangles increases with
increasing absolute value of the bias.
Surface solar irradiance
For the sake of simplicity, the adjustment was performed on the daily mean of
irradiance. Three methods have been investigated: ratio, affine, and quantile
mapping. Each method may be applied to the clearness indices Kt
as well. The possible improvement in bias delivered by each method was
assessed by comparing the original ERA-Interim estimates and the
bias-adjusted ERA-Interim with measurements from the 55 WRDC stations. The
method “ratio” consists of computing the means of HC3v5
IHC3v5‾ and ERA-Interim
IERA‾ for the calibration period of 2005–2014,
then computing the ratio of these means
(IHC3v5‾IERA‾), and eventually multiplying the ERA-Interim estimates by this ratio for the
entire period. The method “affine” consists in adjusting an affine function
between HC3v5 and ERA for the calibration period and then applying this
function to the ERA-Interim estimates. The method “quantile mapping” was
used here (applied to the clearness index) and consists of adjusting the
cumulative distribution function of ERA-Interim to that of HC3v5 for the
calibration period, thus yielding an abacus that is used to convert the
ERA-Interim estimates into adjusted irradiances.
Figure 17 exhibits the bias for ERA-Interim vs. ground observations of daily
mean of solar irradiance for the 55 stations. Downward triangles mean a
negative bias of more than -5 Wm-2, upward triangles mean a
positive bias greater than 5 Wm-2, and circles mean an absolute
value of the bias less than 5 Wm-2. The size of the triangles
increases with increasing absolute value of the bias. Bias is often positive:
i.e. ERA-Interim tends to overestimate the surface solar irradiance. Only 12
stations out of 55 exhibit an absolute bias of less than 5 Wm-2.
Improvement of bias after bias adjustment for daily mean of surface
solar irradiance. Absolute values of the bias after adjustment are coded in
three colours: blue for absolute value < 5 Wm-2, yellow for
5< value < 10 Wm-2, and red for value
> 10 Wm-2. A change in bias is coded by symbols: a circle for
changes in absolute value less than 5 Wm-2, a downward triangle
for improvement in bias, and an upward triangle for degradation. The size of the
triangles increases with increasing absolute value of the bias.
(a) Air temperature differences between adjusted
ERA-Interim and E-OBS for January of the years 1979–2014 and (b) as
(a) but differences between WFDEI and E-OBS for January of the years 1979–2014.
(a) Air temperature differences between adjusted
ERA-Interim and E-OBS for April of the years 1979–2014;
(b) as (a) but showing differences between WFDEI and E-OBS
for April of the years 1979–2014.
(a) Air temperature differences between adjusted
ERA-Interim and E-OBS for July of the years 1979–2014;
(b) as (a) but showing differences between WFDEI and E-OBS
for July of the years 1979–2014.
(a) Air temperature differences between adjusted
ERA-Interim and E-OBS for October of the years 1979–2014;
(b) as (a) but showing differences between WFDEI and E-OBS
for October of the years 1979–2014.
(a) Precipitation total differences between adjusted
ERA-Interim and E-OBS for January of the years 1979–2014;
(b) as (a) but showing differences between WFDEI and E-OBS
for January of the years 1979–2014.
(a) Precipitation total differences between adjusted
ERA-Interim and E-OBS for April of the years 1979–2014;
(b) as (a) but showing differences between WFDEI and E-OBS
for April of the years 1979–2014.
(a) Precipitation total differences between adjusted
ERA-Interim and E-OBS for July of the years 1979–2014;
(b) as (a) but showing differences between WFDEI and E-OBS
for July of the years 1979–2014.
(a) Precipitation total differences between adjusted
ERA-Interim and E-OBS for October of the years 1979–2014;
(b) as (a) but showing differences between WFDEI and E-OBS
for October of the years 1979–2014.
HC3v5 does not cover latitudes north of 60∘ N. Two stations, Lerwick
(Scotland) and Borlange (Sweden), are located along this latitude (Fig. 17).
No adjustment is performed to the grid boxes which are outside the coverage
of HC3v5, except for the grid cells along the border where the new irradiance
values are set to the mean of the original and adjusted irradiances to avoid
spatial discontinuities. Figure 18 exhibits the improvement of bias after
bias adjustment for surface solar irradiance for the 55 sites. Absolute
values of the bias after adjustment are coded in three colours: blue for
absolute value < 5 Wm-2, yellow for 5< value < 10 Wm-2, and red for value
> 10 Wm-2. Change in bias is coded by symbols: a circle for
changes in an absolute value less than 5 Wm-2, a downward
triangle for improvement in bias, and an upward triangle for degradation. The
size of the triangles increases with the improvement in bias. For example, a
green downward triangle means that the bias has been decreased (downward
triangle, i.e. improvement) and that after bias adjustment, the absolute
value of the bias is less than 5 Wm-2. One may see that there is
an improvement or status quo for all stations; i.e. there is no upward
triangle, only circles and downward triangles. In total, 22 stations out of
55 exhibit a bias less than 5 Wm-2 in their absolute values,
which is a strong improvement compared to the 12 for the original ERA-Interim
data.
Once daily means are adjusted, the ratio between the original ERA daily mean
and the adjusted daily mean is applied to each of the eight 3 h irradiances
within each day. Therefore, no alteration is made to the diurnal cycle of
irradiance. This yields the final set of bias-adjusted 3 h surface solar
irradiance.
Comparison of adjusted-ERA and WFDEI averages against E-OBS monthly
averages for the 1979–2014 period
In this section we plot differences for air temperature averages and
precipitation totals for the four mid-season months (January, April, July,
and October) between our adjusted ERA-Interim data compared with E-OBS values
for 1979–2014. Additionally, we plot differences between WFDEI and E-OBS for
the same 36-year average. Figures 19–22 show the air temperature difference
maps with Figs. 23–26 showing the differences for precipitation totals. Each
figure contains two panels, first the adjusted ERA-Interim minus E-OBS and
the second WFDEI minus E-OBS. All 12 monthly difference maps are available at
the ftp site, with details given in Sect. 6. We have used E-OBS in our
bias-adjustment procedure, so this ought to portray our results in a
favourable light with respect to WFDEI. Our reason for this is that E-OBS
uses many more station series than other possible choices (e.g. the CRU TS
dataset used by WFDEI for air temperature or the GPCC dataset used by WFDEI
for precipitation totals). If our approach is used where a dataset of the
quality and temporal resolution of E-OBS is not available, it would be
necessary to use the WFDEI monthly-timescale approach.
For air temperature, differences between our adjusted ERA-Interim and E-OBS
are mostly within ±1 ∘C except at a few locations (Scandinavian
mountains, southern Spain, parts of Italy, the Balkans, and Turkey), more so
in May to August than in other months. Differences cover all of Turkey and
are related to an almost total lack of daily observational data for Turkey
within E-OBS. Additionally, most of the ≥ 1 ∘C differences that
do occur are positive, so our adjusted ERA-Interim is slightly warmer than
E-OBS. For the WFDEI minus E-OBS difference maps, there is always more colour
implying greater differences, which are located more in mountainous regions
and also near some northern coasts. The adjusted ERA-Interim differences are
relatively smooth, but the WFDEI differences are often spotty, probably
related to differences in elevations between WFDEI and E-OBS. As briefly
discussed in Sect. 4.2, we stated that we did not consider elevation
differences between our adjusted ERA-Interim and E-OBS, as both grids have
relatively smooth elevation fields. For each 0.5∘×0.5∘ grid box of E-OBS, the temperature value is an average of the
25 interpolated values on each 0.1∘ grid. Additionally, the elevation
field is also an average of all the elevation points available within the
0.5∘ grid.
For precipitation, the differences appear to be larger but some of this may
be due to expressing the differences as precipitation totals for each month
(as opposed to the daily averages for air temperature). The adjusted
ERA-Interim minus E-OBS differences are greatest in the summer months and
relatively small in winter months. Differences are also more positive than
negative (i.e. adjusted ERA-Interim is wetter than E-OBS). Somewhat in
contrast, the WFDEI differences with E-OBS are larger in the winter months
compared to summer months and their tendency is for WFDEI to be drier than
E-OBS. As with air temperature, the differences are spottier than for
adjusted ERA-Interim. Taken overall, WFDEI appears to be slightly better than
our adjusted ERA-Interim for May–August, but WFDEI is markedly poorer in the
other months (particularly November–March). Elevation differences
(particularly over the Norwegian mountains, the Alps, and also the Caucasus)
are factors, but some could be linked to differences in the elevation
datasets used by E-OBS (used in this paper) and GPCC (used by WFDEI). More
consideration of the differences between climatological averages (e.g. the
1979–2014 averages for E-OBS, GPCC, and CRU TS) is beyond the scope of this
study. Absolute elevation values can be very dependent on the elevation
dataset used (see the discussion in Danielson and Gesch, 2010).
ERA-Interim data were downloaded
from
http://apps.ecmwf.int/datasets/data/interim-full-daily/levtype=sfc/, and our regridded version at the 0.5∘×0.5∘ grid is
available as the original dataset (see Sect. 6).
E-OBS for both daily air temperature and precipitation grids is available at http://www.ecad.eu/download/ensembles/ensembles.php.
CRU for both monthly air temperature and precipitation grids (CRU TS 3.23) is available at https://crudata.uea.ac.uk/cru/data/hrg/.
GPCC for monthly precipitation grids is available at https://www.dwd.de/EN/ourservices/gpcc/gpcc.html.
HadISD for sub-daily station data for wind speeds, dewpoint, and air
temperatures is available at http://www.metoffice.gov.uk/hadobs/hadisd/.
For all the above datasets, the data are freely available for use, but this
is qualified on some sites as use is sometimes just for research and
educational purposes and it may be necessary to register to gain access.
Station data for surface solar irradiance were downloaded from the website
(www.wrdc.mgo.rssi.ru) of the World Radiation Data Center (WRDC) after
registration. Data are available only for research and educational
communities of the countries belonging to the WMO for non-commercial
activities.
HelioClim-3v5 datasets were downloaded from the SoDa Service website
(http://www.soda-pro.com) managed by the company Transvalor. Data are
available to anyone for free for the years 2004–2006 as a GEOSS Data-CORE
(GEOSS Data Collection of Open Resources for Everyone) and for a fee for the
most recent years, with the amount depending on requests and
requester.
Discussion
As stated earlier in the paper, the work reported
here specifically targets energy sector applications; however, the bias
adjustment carried out here could be applied to a wide range of potential
applications. ECEM and its users plan to use both the adjusted and unadjusted
ERA-Interim gridded products through ESCIIs (Energy Sector Climate Impact
Indicators), which will relate the climate variables to energy-relevant
indices. Whether the bias adjustments improve agreement between these ESCIIs
and the direct measures of energy production (e.g. renewable energy from
solar and wind farms) is a simple way of assessing their effectiveness.
The WFDEI bias-adjusted datasets (Weedon et al., 2011, 2014) are similar
datasets covering a much larger region than the ECEM European window. The
adjustments have been performed on a monthly basis. Comparison with E-OBS
shows some seasonal differences in performances between WFDEI and the
proposed dataset, with our dataset being better in the September to April months.
WFDEI datasets have been used extensively, based on citation counts. The
proposed dataset applies adjustments to the distributions of a similar set of
variables, providing daily and 6-hourly estimates. Outside the energy sector,
the bias-adjusted datasets could be used for driving hydrological and
land surface models in a similar way to Orth and Seneviratne (2015). Our bias
adjustments, therefore, could be assessed beyond the energy sector. For
Europe, they could be compared with WFDEI data (often referred to as forcing
data in hydrology, as opposed to bias-adjusted reanalyses) through comparison
of results from hydrologic and/or crop climate models (e.g. using discharge
or yield data). Bias adjustment ought to be an improvement, and this can be
assessed in a similar way to the ESCIIs within ECEM.
Is there a way of simultaneously bias adjusting all variables or at least
adjusting pairs to start with? Whereas Weedon et al. (2014) and this paper
have not attempted multivariate adjustment, this is being tested in the ECEM
project. However, as the number of variables increases, this becomes more
impractical. The usefulness of all bias-adjusted datasets can be assessed
through ESCIIs and discharge–yield data (i.e. using variables external to
reanalysis) which would be expected to be best simulated just as if perfect
observational data were available. Multivariate bias adjustment was
experimented with in ECEM (using wind speed and temperature), but the results
are dependent on the availability of adequate station data for variables
measured together (Dekens et al., 2017). Access to data is a crucial aspect
of all the datasets used in this study. ERA-Interim would be improved with
greater numbers of station input data, as would E-OBS and the other data
products considered in this paper. Improved access, however, is unlikely to
reduce the need for bias adjustment.
The ECEM dataset: its description and how to access it
All the ERA-Interim (original and bias-adjusted) are available as netcdf
files from the CDS of the Copernicus Climate Data
Service. As this CDS is currently being developed, this ftp site
(ftp://ecem.climate.copernicus.eu) can currently be used to access all
files discussed in this paper. This site currently has no password, but once
on the CDS, there will likely be a registration procedure. Datasets are named
according to the ECEM project. The original or unadjusted filenames have
“noc” in the file name. They are as follows for air temperature (T2M), dewpoint
temperature (DP), solar irradiance (GHI, Global Horizontal Irradiation), wind
speed (WS) and precipitation (TP):
The adjusted files are labelled similarly but include the distribution and
“bc” instead of “noc”. So, for air temperature and dewpoint, they include
“nbc”, for solar irradiance “qbc”, for wind speed “wbc”, and
for precipitation “gbc”. A final file contains the bias-adjusted relative
humidity file.
Two example locations for each variable of the distributional comparisons are
given in the paper (Figs. 2, 3, 6, 7, 10, 11, 14, and 15). The ftp site also
includes all the distributional comparisons as pdfs, with the stations
ordered by their WMO number when comparing with HadISD and by latitude and then
longitude when comparing with E-OBS. These files have the following names, for air
temperature (Tmean), dewpoint temperature, wind speed (WS), and precipitation
(dly_precip), respectively:
Additionally, the ftp site includes all 12 monthly difference plots, which
are shown in Figs. 19–26 for the mid-season months:
adjERA-EOBS_Tmean_monthly_deltas_means.pdf
adjERA-EOBS_Precip_monthly_deltas_means.pdf
WFDEI-EOBS_Tmean_monthly_deltas_means.pdf
WFDEI-EOBS_Precip_monthly_deltas_means.pdf.
The authors declare that they have no conflict of
interest.
Acknowledgements
The authors would like to acknowledge funding for the European Climatic
Energy Mixes (ECEM) project by the Copernicus Climate Change Service, a
programme being implemented by the European Centre for Medium-Range Weather
Forecasts (ECMWF) on behalf of the European Commission. The specific grant
number is 2015/C3S_441_Lot2_UEA. The authors particularly thank the
reviewers (Graham Weedon and Helge Goessling) for their extensive and
thoughtful reviews, which have significantly improved the paper. The
authors also thank Robert Dunn from the UK MetOffice Hadley Centre who kindly
extracted the wind speed, dewpoint, and air temperature for all European
stations from the HadISD dataset. The authors thank all ground station
operators of the WMO network for their valuable measurements. They
additionally thank the World Radiation Data Centre for hosting a website for
downloading data. The authors thank the French company Transvalor, which takes care of the SoDa Service for the common good, thus providing an
efficient access to the HelioClim databases.
Edited by: Gert König-Langlo
Reviewed by: Graham Weedon and Helge Goessling
ReferencesBecker, A., Finger, P., Meyer-Christoffer, A., Rudolf, B., Schamm, K.,
Schneider, U., and Ziese, M.: A description of the global land-surface
precipitation data products of the Global Precipitation Climatology Centre
with sample applications including centennial (trend) analysis from
1901–present, Earth Syst. Sci. Data, 5, 71–99,
10.5194/essd-5-71-2013, 2013.Blanc, P., Gschwind, B., Lefèvre, M., and Wald, L.: The HelioClim
project: Surface solar irradiance data for climate applications, Remote
Sensing, 3, 343–361, 10.3390/rs3020343, 2011.Bojinski, S., Verstraete, M., Peterson, T. C., Richter, C., Simmons, A., and
Zemp, M.: The concept of essential climate variables in support of climate
research, applications, and policy, B. Am. Meteor. Soc., 95, 1431–1443,
10.1175/BAMS-D-13-00047.1, 2014.Boilley, A. and Wald, L.: Comparison between meteorological re-analyses from
ERA-Interim and MERRA and measurements of daily solar irradiation at surface,
Renew. Energ., 75, 135–143, 10.1016/j.renene.2014.09.042, 2015.Compo, G. P., Whitaker, J. S., Sardeshmukh, P. D., Matsui, N., Allan, R. J.,
Yin, X., Gleason Jr., B. E., Vose, R. S., Rutledge, G., Bessemoulin, P.,
Brönnimann, S., Brunet, M., Crouthamel, R. I., Grant, A. N., Groisman, P.
Y., Jones, P. D., Kruk, M. C., Kruger, A. C., Marshall, G. J., Maugeri, M.,
Mok, H. Y., Nordli, Ø., Ross, T. F., Trigo, R. M., Wang, X. L., Woodruff,
S. D., and Worley, S. J.: The twentieth century reanalysis project, Q. J.
Roy. Meteor. Soc., 137, 1–28, 10.1002/qj.776, 2011.Cosgrove, B. A., Lohmann, D., Mitchell, K. E., Houser, P. R., Wood, E. F.,
Schaake, J. C., Robock, A., Marshall, C., Sheffield, J., Duan, Q., Luo, L.,
Wayne Higgins, R., Pinker, R. T., Dan Tarpley, J., and Meng, J.: Real-time
and retrospective forcing in the North American Land Data Assimilation System
(NLDAS) project, J. Geophys. Res., 108, 8842, 10.1029/2002JD003118,
2003.
Danielson, J. J. and Gesch, D. B.: Global multi-resolution terrain elevation
data 2010 (GMTED2010): U.S. Geological Survey Open-File Report 2011–1073,
26 pp, 2011.Dee, D. P., Uppala, S. M., Simmons, A. J., Berrisford, P., Poli, P.,
Kobayashi, S., Andrae, U., Balmaseda, M. A., Balsamo, G., Bauer, P.,
Bechtold, P., Beljaars, A. C. M., van de Berg, L., Bidlot, J., Bormann, N.,
Delsol, C., Dragani, R., Fuentes, M., Geer, A. J., Haimberger, L., Healy, S.
B., Hersbach, H., Hólm, E. V., Isaksen, L., Kållberg, P., Köhler,
M., Matricardi, M., McNally, A. P., Monge-Sanz, B. M., Morcrette, J.-J.,
Park, B.-K., Peubey, C., de Rosnay, P., Tavolato, C., Thépaut, J.-N., and
Vitart, F.: The ERA-Interim reanalysis: configuration and performance of the
data assimilation system, Q. J. Roy. Meteor. Soc., 137, 553–597,
10.1002/qj.828, 2011.Dekens, L., Parey, S., Grandjacques, M., and Dacunha-Castelle, D.:
Multivariate distribution correction of climate model outputs:
a generalisation of quantile mapping approaches, Environmetrics, 10.1002/env.2454, online first,
2017.Dunn, R. J. H., Willett, K. M., Thorne, P. W., Woolley, E. V., Durre, I.,
Dai, A., Parker, D. E., and Vose, R. S.: HadISD: a quality-controlled global
synoptic report database for selected variables at long-term stations from
1973–2011, Clim. Past, 8, 1649–1679, 10.5194/cp-8-1649-2012, 2012.Dunn, R. J. H., Willett, K. M., Morice, C. P., and Parker, D. E.: Pairwise
homogeneity assessment of HadISD, Clim. Past, 10, 1501–1522,
10.5194/cp-10-1501-2014, 2014.Eissa, Y., Korany, M., Aoun, Y., Boraiy, M., Abdel Wahab, M., Alfaro, S.,
Blanc, P., El-Metwally, M., Ghedira, H., and Wald, L.: Validation of the
surface downwelling solar irradiance estimates of the HelioClim-3 database in
Egypt, Remote Sensing, 7, 9269–9291, 10.3390/rs70709269, 2015.Gudmundsson, L., Bremnes, J. B., Haugen, J. E., and Engen-Skaugen, T.:
Technical Note: Downscaling RCM precipitation to the station scale using
statistical transformations – a comparison of methods, Hydrol. Earth Syst.
Sci., 16, 3383–3390, 10.5194/hess-16-3383-2012, 2012.
Harpham, C., Troccoli, A., Jones, P., Ranchin, T., and Wald, L.: Comparing
monthly statistical distributions of wind speed measured at wind towers and
estimated from ERA-Interim, 16th EMS Annual Meeting, 12–16 September 2016,
Trieste, Italy, EMS Annual Meeting Abstracts, 13, EMS2016-336, 2016.Harris, I., Jones, P. D., Osborn, T. J., and Lister, D. H.: Updated
high-resolution monthly grids of monthly climatic observations: the CRU TS
3.10 dataset, Int. J. Climatol., 34, 623–642, 10.1002/joc.3711, 2014.Haylock, M. R., Hofstra, N., Klein Tank, A. M. G., Klok, E. J., Jones, P. D.,
and New, M.: A European daily high-resolution gridded dataset of surface
temperature and precipitation, J. Geophys. Res., 113, D20119,
10.1029/2008JD010201, 2008.Hersbach, H., Peubey, C., Simmons, A., Berrisford, P., Poli, P., and Dee, D.:
ERA-20CM: a twentieth-century atmospheric model ensemble, Q. J. Roy. Meteor.
Soc., 141, 2350–2375, 10.1002/qj.2528, 2015.
Horton, E. B., Folland, C. K., and Parker, D. E.: The changing incidence of
extremes in worldwide and Central England temperatures to the end of the
twentieth century, Climatic Change, 50, 267–295, 2001.Jones, P. D.: The Reliability of Global and Hemispheric Surface Temperature
Records, Adv. Atmos. Sci., 33, 269–282, 10.1007/s00376-015-5194-4,
2016.Maraun, D.: Nonstationarities of regional climate model biases in European
seasonal mean temperature and precipitation sums, Geophys. Res. Lett., 39,
L06706, 10.1029/2012GL051210, 2012.Maraun, D.: Bias correction, quantile mapping, and downscaling: revisiting
the inflation issue, J. Climate, 26, 2137–2143,
10.1175/JCLI-D-12-00821.1, 2013.Maraun, D., Wetterhall, F., Ireson, A. M., Chandler, R. E., Kendon, E. J.,
Widmann, M., Brienen, S., Rust, H. W., Sauter, M., Themeßl, T., Venema,
V. K. C., Chun, K. P., Goodess, C. M., Jones, R. G., Onof, C., Vrac, M., and
Thiele-Eich, I.: Precipitation downscaling under climate change. Recent
developments to bridge the gap between dynamical models and the end user,
Rev. Geophys., 48, RG3003, 10.1029/2009RG000314, 2010.Marchand, M., Al-Azri, N., Ombe-Ndeffotsing, A., Wey, E., and Wald, L.:
Evaluating meso-scale change in performance of several databases of hourly
surface irradiation in South–eastern Arabic Pensinsula, Adv. Sci. Res., 14,
7–15, 10.5194/asr-14-7-2017, 2017.Orth, R. and Seneviratne, S.I.: Introduction of a simple-model-based land
surface dataset for Europe, Environ. Res. Lett., 10, 044012,
10.1088/1748-9326/10/4/044012, 2015.Oyerinde, G. T., Hountondji, F. C. C., Lawin, A. E., Odofin, A. J., Afouda,
A., and Diekkrüger, B.: Improving hydro-climate projections with
bias-correction in Sahelian Niger basin, West Africa, Climate, 5, 8,
10.3390/cli5010008, 2017.
Piani, C., Haerter, J. O., and Coppola, E.: Statistical bias correction for
daily precipitation in regional climate models over Europe, Theor. Appl.
Climatol., 99, 187–192, 2010a.Piani, C., Weedon, G., Best, M., Gomes, S., Viterbo, P., Hagemann, S., and
Haerter, J.: Statistical bias correction of global simulated daily
precipitation and temperature for the application of hydrological models,
J. Hydrol., 395, 199–215, 10.1016/j.jhydrol.2010.10.024, 2010b.Rienecker, M. M., Suarez, M. J., Gelaro, R., Todling, R., Bacmeister, J.,
Liu, E., Bosilovich, M. G., Schubert, S. D., Takacs, L., Kim, G., Bloom, S.,
Chen, J., Collins, D., Conaty, A., da Silva, A., Gu, W., Joiner, J., Koster,
R. D., Lucchesi, R., Molod, A., Owens, T., Pawson, S., Pegion, P., Redder, C.
R., Reichle, R., Robertson, F. R., Ruddick, A. G., Sienkiewicz, M., and
Woollen, J.: MERRA: NASA's Modern-Era Retrospective Analysis for Research and
Applications, J. Climate, 24, 3624–3648, 10.1175/JCLI-D-11-00015.1,
2011.Rigollier, C., Lefèvre, M., and Wald, L.: The method Heliosat-2 for
deriving shortwave solar radiation from satellite images, Sol. Energy, 77,
159–169, 10.1016/j.solener.2004.04.017, 2004.
Sheffield, J., Goteti, G., and Wood, E. F.: Development of a 50-yr
high-resolution global dataset of meteorological forcings for land surface
modeling, J. Climate, 19, 3088–3111, 2006.Simmons, A. J., Berrisford, P., Dee, D. P., Hersbach, H., Hirahara, S., and
Thépaut, J.-N.: A reassessment of temperature variations and trends from
global reanalyses and monthly surface climatological datasets, Q. J. Roy.
Meteor. Soc., 143: 101–119, 10.1002/qj.2949, 2017.Smith, A., Lott, N., and Vose, R.: The Integrated Surface Database: Recent
Developments and Partnerships, B. Am. Meteorol. Soc., 92, 704–708,
10.1175/2011BAMS3015.1, 2011.Themeßl, M. J., Gobiet, A., and Leuprecht, A.: Empirical-statistical
downscaling and error correction of daily precipitation from regional climate
models, Int. J. Climatol., 31, 1530–1544, 10.1002/joc.2168, 2011.Themeßl, M. J., Gobiet, A., and Heinrich, G.: Empirical-statistical
downscaling and error correction of regional climate models and its impact on
the climate change signal, Climatic Change, 112, 449–468,
10.1007/s10584-011-0224-4, 2012.Thomas, C., Wey, E., Blanc, P., and Wald, L.: Validation of three
satellite-derived databases of surface solar radiation using measurements
performed at 42 stations in Brazil, Adv. Sci. Res., 13, 129–136,
10.5194/asr-13-129-2016, 2016a.
Thomas, C., Wey, E., Blanc, P., Wald, L., and Lefèvre, M.: Validation of
HelioClim-3 version 4, HelioClim-3 version 5 and MACC-RAD using 14 BSRN
stations, 2015 Solar Heating and Cooling, Energy Procedia, 91, 1059–1069,
2016b.Tye, M. R., Stephenson, D. B., Holland, G. J., and Katz, R. W.: A Weibull
approach for improving climate model projections of tropical cyclone
wind-speed distributions, J. Climate, 27, 6119–6133,
10.1175/JCLI-D-14-00121.1, 2014.
Vrac, M. and Friederichs, P.: Multivariate–intervariable, spatial, and
temporal–bias correction, J. Climate, 28, 218–237,
10.1175/JCLI-D-14-00059.1, 2015.Vrac, M., Drobinski, P., Merlo, A., Herrmann, M., Lavaysse, C., Li, L., and
Somot, S.: Dynamical and statistical downscaling of the French Mediterranean
climate: uncertainty assessment, Nat. Hazards Earth Syst. Sci., 12,
2769–2784, 10.5194/nhess-12-2769-2012, 2012.Weedon, G. P., Gomes, S., Viterbo, P., Österle, H., Adam, J.C., Bellouin,
N., Boucher, O. and M. Best, M.: The WATCH forcing data 1958–2001: A
meteorological forcing dataset for land surface and hydrological models,
Technical Report No. 22, 41 pp., available at:
http://www.eu-watch.org/media/default.aspx/emma/org/10376311/, last
access: 3 July 2017, 2010Weedon, G. P., Gomes, S., Viterbo, P., Shuttleworth, W. J., Blyth, E.,
Osterle, H., Adam, J. C., Bellouin, N., Boucher, O., and Best, M.: Creation
of the WATCH forcing data and its use to assess global and regional reference
crop evaporation over land during the twentieth century, J. Hydrometeorol.,
12, 823–848, 10.1175/2011JHM1369.1, 2011.Weedon, G. P., Balsamo, G., Bellouin, N., Gomes, S., Best, M. J., and
Viterbo, P.: The WFDEI meteorological forcing data set: WATCH Forcing Data
methodology applied to ERA-Interim reanalysis data, Water Resour. Res., 50,
7505–7514, 10.1002/2014WR015638, 2014.Wilcke, R., Mendlik, T., and Gobiet, A.: Multi-variable error correction of
regional climate models, Climatic Change, 120, 871–887,
10.1007/s10584-013-0845-x, 2013
Wilks, D. S.: tatistical Methods in Atmospheric Sciences, Academic, New York,
467 pp., 1995.Yang, W., Andreasson, J., Graham, L. P., Olsson, J., Rosberg, J., and
Wetterhall, D.: Distribution-based scaling to improve usability of regional
climate model projections for hydrological climate change impacts studies,
Hydrol. Res., 41, 3–4, 10.2166/nh.2010.004, 2010.