Introduction
Regional climate change projections for all
terrestrial regions of the globe within the time line of the Fifth Assessment
Report (AR5) and beyond have been made available for climate researchers in
the framework of the CORDEX initiative. Within this initiative, a large
ensemble of high-resolution regional climate projections including Europe
(EURO-CORDEX, the European branch of the CORDEX initiative) have been made
available to provide climate simulations for use in climate change impact,
adaptation, and mitigation studies . Although most
of the simulations are run on a high grid resolution, systematic biases in
the regional climate models (RCMs) remain, due to errors related to (i)
imperfect model representation of the physical processes or phenomena and
(ii) to the parametrization and incorrect initialization of the models. Thus,
even when using the highest resolution available, RCMs still
require some adjustments .
Therefore, bias correction methods continue to be used in impact studies – for example in
hydrology e.g., agronomy
e.g., ecology, and more recently by climate
services e.g. – to reduce systematic bias
in (regional or global) climate models.
Traditionally, the bias correction method ensures equal mean values between
the corrected simulations and observations
e.g. – hence, explicitly addressing only
one aspect of the statistical properties of the simulated data. More advanced
methods consider the whole distribution of a weather variable to be adjusted,
including extremes, so that it matches the distribution of the observations
e.g..
Recent studies have compared different RCM bias correction methods.
evaluated seven bias correction
methods used to correct modelled precipitation by the RCM
MM5 using forcings from ERA-40 reanalysis. They concluded that quantile
mapping outperforms all methods considered, especially at high quantiles.
applied three bias correction methods to correct the
mean and variance of precipitation and temperature modelled by the RCM
COSMO-CLM driven by the ECHAM5-MPIOM global climate model (GCM) over all of Germany and nearby
surrounding areas, modelled at 7 km resolution and validated against
30 years of 1 km gridded observation data (1971–2000). They found
that some of the methods correct not only the means but also the higher
moments. confirmed that non-parametric
methods such as quantile mapping are more suitable in reducing systematic
errors in model data. They compared 11 bias correction methods used to
correct precipitation modelled by RCM HIRAM forced with the ERASE reanalysis
data and found that non-parametric methods performed the best in
reducing systematic errors, followed by parametric transformations with three
or more free parameters, with the lowest rank taken by the distribution-derived
transformations. applied six bias correction
methods to correct 11 different RCM-simulated temperature and
precipitation series, and found that all methods were able to preserve
the mean – however, other statistical properties were degraded.
applied four distribution-based bias correction
methods to correct precipitation modelled by the RCM
HadRM3-PPEUK driven by the GCM HadCM3 over seven catchments
in Great Britain. They found that gamma-based quantile mapping offers
the best combination of accuracy when evaluated on the first four order
moments (mean, SD, skewness, and kurtosis).
tested six distribution-based bias correction methods. They found that
all evaluated methods perform reasonably well in (i) reproducing statistical
properties of the observations including high-order moments and quantiles and
(ii) preserving the climate change signal.
Bias correction methods can also be categorized into parametric and
non-parametric methods. In the parametric methods the distribution of the
data is assumed to be known. For instance, it is well known that the
probability distribution of daily temperature values follows a normal
distribution , whereas the exponential
and gamma distributions
are often used to model the intensity of
daily precipitation. Likewise, the Bernoulli and geometric distributions are
often used to model the probability distribution of the occurrence of daily
precipitation (frequency) and the number of consecutive dry/wet days,
respectively . On the other hand, the
non-parametric methods are applied without prior assumptions about the
distribution of the data . Hence, they are
more attractive for many applications including those based on bias
correction. Another advantage is that non-parametric methods are more
suitable in reducing systematic errors in model data
.
Among existing methods, the non-parametric quantile mapping method, referred to
as quantile mapping (QM) for simplicity, has shown a good
performance in reproducing not only the mean and the SD but also other
statistical properties such as quantiles . As the
method belongs to the non-parametric family, it does not require prior
knowledge of the theoretical distribution of the weather variable, which makes
it very attractive, as it is easy to implement, in addition to its simple and
non-parametric configuration .
However, the QM has a few limitations. It is particularly sensitive to the
choice and the length of the calibration time period to make a reliable
estimation not affected by data sampling problems
. Thus, it requires a reference dataset to adjust
the modelled data to match the observations
. It is sometimes difficult to apply this method to
different climatic conditions, as unobserved values may lie outside the range
of those in the calibration time period
(i.e. values in the tail of the
distribution). Another issue is related to the misrepresentation of the
(physical) link between weather variables, which can be altered especially if
applied to each climate variable separately. For instance, in most
hydrological applications the dependence between daily precipitation and
temperature can affect the discharge .
A few studies related to projections of climate change have been dedicated to
Poland. For instance, climate projections originating from the ENSEMBLES
project were used as the basis to investigate the
impact of climate change on various sectors (agriculture, water resources, and
health) in Europe. assessed six regional climate
model simulations under the SRES A2 emission scenario and found
unfavourable changes in Polish climate, such as an increased frequency of
extreme events, reduced crop yields, and increased summer water budget
deficit. In the “KLIMADA” project, the ENSEMBLES projections were
additionally bias-adjusted (http://klimada.mos.gov.pl/en/) within the
framework of the Polish National Adaptation Strategy to Climate Change (NAS
2020) to estimate changes in climate variables and indices for two future
horizons – the near future (2021–2050) and the far future (2071–2100). The
outcomes of the latter project showed significant upward trend in temperature
and uncertain precipitation increases in the median of winter changes, and
slight decreases in summer .
assessed the spatial variability in
winter temperature over Poland for the near future (2021–2050) based on
three RCMs under the SRES A1B emission scenario. More
recently, applied a bias correction method on six
simulations from the ENSEMBLES project to assess various drought indices over
Poland and pointed out that the correction process altered the magnitude of
the trend in corrected modelled precipitation but not its direction.
There have also been a few studies carried out for Poland based on the newest
generation of climate model simulations (i.e. the fifth generation of the
Coupled Model Inter-comparison Project (CMIP5) and the European domain of the
Coordinated Downscaling Experiment Initiative (EURO-CORDEX)).
used bias-adjusted modelled temperature and
precipitation (seven GCM–RCM combinations from the EURO-CORDEX initiative
over 10 Polish catchments) and found that projections following the RCP4.5
emission scenario agreed on a precipitation increase of up to 15 %, and
a warming of up to 2 ∘C by the end of the 21st century.
applied a statistical downscaling model to
produce temperature and precipitation projections from two global climate
models following three different emission scenarios (RCP2.6, RCP8.5, and SRES
A1B) for the southwest of Poland and eastern Saxony. They found an acceleration of changes by the end of the 21st century leading to
negative consequences for the climatic water balance, particularly under SRES
A1B and RCP8.5 emission scenarios.
The main objective of this paper is to provide an update of climate
projections over Poland by adopting the new generation of concentration
pathways and recent developments in climate modelling. We hope the dataset
provided here will be beneficial for the research community – for instance for
impact studies in areas such as hydrology, ecology and agricultural sciences.
In this paper we also use a recently made available high-resolution gridded
observational dataset (CPLFD-GDPT5; see Sect. ) covering
more than 60 years as a reference for the bias correction procedure
(Sect. ).
Data analyses
Bias correction method
We used the quantile mapping to correct for systematic biases in RCM simulations. The quantile mapping tries to find a statistical
transformation or function F that maps a simulated variable y such that
its new distribution closely fits the distribution of the observed variable
x. In general, this transformation can be formulated as
x=F(y).
The non-parametric transformation is then defined as
x=F-1(G(y)),
where G is the cumulative distribution function of y and F-1 is the
inverse cumulative distribution function corresponding to x. The quantile
mapping of the simulated time series to the observed ones was performed for
each grid cell. Here, the number of quantiles was set to
Nq=1000 and was chosen to be regularly spaced. Two steps were
performed. First, RCM corresponding quantiles were taken from the empirical
cumulative distribution function based on observations. Second, these
estimates were used to perform a quantile mapping. It should be noted that
the set-up included a linear interpolation between the fitted transformed
values and simulated values lying outside the range of observed values in the
training period. Hence, they were extrapolated using the correction found for the
highest percentile as suggested by . Furthermore,
the method included an adjustment of wet-day frequencies for precipitation.
Here, a wet day was defined as a day with a precipitation amount higher than
0 mm day-1. The probability of wet days was first derived from
observations, and then used as a threshold, so that all modelled values below
this threshold were set to zero. This ensured an equal fraction of rainy days
between observed and modelled data. The transformations were, additionally,
fitted to the portion of the distributions corresponding to observed wet
days. The quantile mapping method was applied on each of the four seasons
separately to take into account seasonality in the biases, as different
seasons may be influenced by different physical processes. Then, the output
data were merged to reconstruct a full simulation. As discussed in
Sect. , the quantile mapping may modify the link between
individually post-processed climate variables. However, correcting the
present climate to be closer to the observations has been necessary for most
climate change impact studies . Quantiles of
the simulations for the control period (1971–2000) were mapped onto
corresponding quantiles in the observations considered as the most recent
63-year reference time period (1951–2013). The transfer functions were then
used to correct for the bias in the daily minimum and maximum temperatures
and precipitation simulations defined in Table .
Biases in RCM simulations
Each of the nine bias-corrected historical simulations was evaluated on its
ability to reproduce statistical properties of the pseudo-observed or
reference dataset. In our case, averaged error values (e) over time (t)
for each grid cell in terms of RMSEs were considered as
a measure of the model's performance. As model errors are often seasonally
dependent, the four seasons are treated separately. For a time t
(seasonal or annual), the model error is calculated as
et=st-ot,
where s and o refer to simulated and observed values. The RMSE averaged over space is then defined as
RMSE=1Ng∑i=1Nget2,
where Ng is the total number of grid cells.
The averaged RMSE informs about the magnitude of the overall deviation
between the simulations and pseudo-observations over all Poland, while the
error (e) indicates whether there was an over- (positive) or under-
(negative) estimation (bias) of the simulated values at each grid cell.
The RMSE was computed between values of bias-corrected and
raw monthly sums of precipitation, daily minimum and maximum temperatures, and
their corresponding observations. The model error was first computed on each
grid cell, then mapped across Poland and averaged from the spatial field
only – i.e. with the RMSE of the temporal means of all grid cells, not of the
single grid cells.
Tables to gives the mean and SD of
the RMSE derived from the ensemble of model simulations.
The RMSE in annual means averaged over all raw simulations
(i.e. multi-model ensemble mean) was 15.5±4.4 mm month-1
for precipitation and 1.1±0.7 ∘C and 1.6±0.45 ∘C for daily minimum and maximum temperatures,
respectively (Tables to ). For
seasonal means, the largest error was found in summer precipitation
(26.4±8.6 mm month-1), mainly due to the convection, not
well represented in the climate models. The lowest error was found in the
autumn, where precipitation is influenced by continental air masses. The same
tendency was additionally found for daily maximum temperature, i.e. large
error in the summer compared to the other seasons. However, for daily minimum
temperature, the largest error of 1.5 ∘C was obtained for
spring, followed by summer with a slightly lower bias of 1.3 ∘C. In general, biases in daily maximum temperature were 0.5 to
1 ∘C higher than those found in daily minimum temperature. The
lowest precipitation bias was simulated by the regional atmospheric model RCA4
driven by the global model M-MPI-ESM-LR (Simulation 8 in
Table ). Obviously, these biases or model errors might be
related to the complexity of the climate system in Poland, which has been very
difficult to predict, being influenced by air masses from all four directions
.
RMSEs in annual and seasonal means of daily
maximum temperature (∘C). RMSEs
were computed between historical simulations and observations and averaged over all grid cells.
N
GCM/RCM simulation
Annual
Winter
Spring
Summer
Autumn
1
CNRM-CM5/CCLM4-8-17
1.5
2.4
2.8
1.6
1.6
2
CNRM-CM5/RCA4
1.3
1.6
2.8
1.1
0.6
3
EC-EARTH/CCLM4-8-17
1.6
1.5
1.8
1.6
2.0
4
EC-EARTH/RCA4
1.8
1.1
2.4
2.8
1.3
5
EC-EARTH/RACMO22E
1.9
0.9
2.7
2.7
1.8
6
EC-EARTH/HIRHAM5
2.4
1.5
2.1
3.4
2.8
7
IPSL-CM5A-MR/RCA4
1.7
0.7
3.3
2.8
1.0
8
MPI-ESM-LR/CCLM4-8-17
1.8
1.3
1.4
2.8
1.9
9
MPI-ESM-LR/RCA4
0.8
0.8
0.9
2.1
0.6
Ens. mean*
All
1.6
1.3
2.3
2.3
1.5
Ens. SD*
All
0.5
0.5
0.8
0.7
0.7
* Ens. stands for ensemble, SD for standard deviation.
RMSEs in annual and seasonal means of daily
minimum temperature (∘C). The RMSEs were computed between historical simulations and observations and averaged over all grid cells.
N
GCM/RCM simulation
Annual
Winter
Spring
Summer
Autumn
1
CNRM-CM5/CCLM4-8-17
0.7
1.3
1.0
2.3
0.8
2
CNRM-CM5/RCA4
1.13
1.7
1.9
0.7
0.9
3
EC-EARTH/CCLM4-8-17
0.6
0.7
0.6
0.9
0.8
4
EC-EARTH/RCA4
1.4
1.2
1.6
2.1
1.3
5
EC-EARTH/RACMO22E
2.7
2.2
4.0
2.1
2.6
6
EC-EARTH/HIRHAM5
0.6
1.0
0.7
0.6
0.8
7
IPSL-CM5A-MR/RCA4
1.0
1.2
1.8
1.6
0.8
8
MPI-ESM-LR/CCLM4-8-17
0.8
1.0
1.2
0.7
0.8
9
MPI-ESM-LR/RCA4
0.8
1.5
1.0
0.9
0.9
Ens. mean*
All
1.1
1.3
1.5
1.3
1.1
Ens. SD*
All
0.7
0.4
1.1
0.7
0.6
* Ens. stands for ensemble, SD for standard deviation.
Moreover, model errors or biases are often spatially dependant and varied
among the simulations and seasons. In our case, all historical simulations
showed wet and warm biases as well as dry and cold biases across the region,
which were more pronounced in the mountainous areas located in the southern
parts of Poland, due to topographical features not well represented in the
models (Figs. 1–27 in the Supplement). This can also be related to the low observational
network density in this region. As we did not intend to perform a thorough
comparison between all model simulations, only an example of the bias in the
RCM CCLM4-8-17 driven by the CNRM-CM5 GCM
for the historical climate (Simulation 1 in
Table ) is detailed here
(Figs. –).
Bias evaluation for Simulation 1
(Table ). The maps show RMSEs
estimated on the
difference between historical simulations (all available years included) and observations (CPLFD-GDPT5) for both raw (a) and
bias-adjusted (b) monthly sums of precipitation modelled by the CCLM4-8-17 RCM driven by the CNRM-CM5 GCM. The legend “RMSE” indicates the areal mean bias estimated from the gridded annual and seasonal aggregates and the
black polylines show the delimitation of the Polish provinces.
![]()
As Fig. but for daily maximum
temperature.
![]()
As Fig. but for daily minimum
temperature.
![]()
For precipitation, seasonal evaluations additionally showed that the
relatively high RMSE found in the raw data on an annual scale
(13.5 mm month-1) was due to high wet biases in summer and
winter, which were 45.2 and 11.7 mm month-1, respectively,
compared to the transition seasons with relatively low biases
(7 mm month-1 for spring and 8.5 mm month-1 for
autumn). The highest discrepancy of the models was obtained
in mountainous areas located in the south (negative bias range of
-75 to -100 mm month-1) and the eastern part of the region,
with a negative bias in the range of -50 to -75 mm month-1.
Obviously, the RMSEs from the bias-adjusted results were very low compared to
those obtained from the raw simulations and were mostly close to zero for
annual and seasonal means – apart from summer, where a small bias of
3.5 mm month-1 persisted in the adjusted precipitation. For
maximum temperature, there was an overall cold bias for all seasons except
for summer, which showed a warm bias everywhere, except for the mountains
located in the south exhibiting a more enhanced cold bias. The annual RMSE of
the raw data was 1.5 ∘C. Seasonal evaluations showed higher
cold biases for winter and spring of 2.4 and 2.8 ∘C,
respectively. For the corrected results, the bias was reduced to almost zero
for all seasons and on an annual scale. For daily minimum temperature, the RMSE
was slightly lower than for maximum temperature for all seasons except
summer, where a relatively high warm bias of about of 2.3 ∘C
was found, which was additionally influenced by mountains located in the
south. However, the spatial distribution of the biases showed a similar pattern
to that discussed earlier for maximum temperature. Obviously, the RMSEs based
on corrected datasets were all close to zero for both daily minimum and
maximum temperatures. However, there was still a spatial structure to the
errors. Biases were also removed in corrected precipitation – apart from summer
precipitation, where small biases lower than 10 % remained for all
simulations (Sect. 2 in the Supplement).
Sensitivity to the climate change signal
Although the bias correction significantly improved the quality of the
simulations in the trained control time period, it may alter the physical
link between climate variables in the model and
possibly modify the climate change signal . We further
investigated the influence of QM on the climate
change signal. Accordingly, we mapped the climate change signal in both raw
and corrected simulations and focused on the time period 2096–2100 with
regard to historical data as we would expect a stronger alteration of the
climate signal by the end of the century rather than earlier. An example of
results based on the RCM RCA4 driven by the GCM MPI-ESM-LR is illustrated in what follows (Simulation 9 in
Table ).
Figures – suggest that the sign and
magnitude and the spatial distribution of the estimated changes were
maintained and hence were not affected by the correction procedure. This
demonstrated the reliability of the projected climate changes by the corrected
RCM simulations. One possible explanation could be related
to the use of the long reference record (1951–2013) on which the calibration was
performed – i.e. the training distribution has been built on a long record
encompassing different climate conditions rather than the short reference
periods that are commonly selected (e.g. 1971–2000 or the new
normal, 1981–2010). However, a few exceptions were found. For instance, the
magnitude of the climate change (in root mean square terms) between
historical and future (2096–2100) simulations for bias-adjusted modelled
precipitation was reduced by approximately 15 to 25 % in corrected
summer and spring changes compared to corresponding changes in the raw data,
respectively, although reported
that the impact of the bias correction on the climate change signal may be
larger than the signal itself. Overall, the spatial distribution of the
climate change signal was, however, consistent in all corrected simulations
–
i.e. no random effect was introduced by the correction. Similar results were
obtained for the other RCM simulations and even when
assuming the RCP8.5 scenario (Figs. S28–S54).
Precipitation change signal (mm month-1) for Simulation 9
(Table ). The maps show absolute changes
in the future (2096–2100) with regard to historical simulations (all years included) for both raw (a) and bias-adjusted (b)
monthly sums of precipitation modelled by the RCA4 RCM driven by the MPI-ESM-LR GCM. The legend
“RMSCC” indicates the areal mean change estimated from the gridded annual and seasonal aggregates and the black polylines show the
delimitation of the Polish provinces.
![]()
As Fig. but for absolute changes in
daily minimum temperature (∘C).
![]()
As Fig. but for absolute changes in
daily maximum temperature (∘C).
![]()
Summary of changes in projected multi-model ensemble seasonal and
annual regional means of mean temperature (a, in
∘C) and precipitation (b, in %) for the near (2021–2050) and far (2071–2100) futures assuming both the
RCP4.5 and RCP8.5. Values in brackets indicate the 5th and 95th percentiles of the projected ensembles and hence represent the
90 % confidence interval of the mean estimates from the multi-model ensemble.
Scenario/Horizon
DJF
MAM
JJA
SON
Annual
(a) Temperature changes
RCP4.5 by 2021–2050
+1.2
+1.0
+1.0
+1.1
+1.1
[+0.3,+1.9]
[+0.6,+1.7]
[+0.7,+1.4]
[+0.6,+1.6]
[+0.7,+1.4]
RCP8.5 by 2021–2050
+1.6
+1.3
+1.1
+1.3
+1.3
[+0.5,+2.5]
[+0.9,+2]
[+0.7,+1.3]
[+0.6,+1.8]
[+0.8,+1.8]
RCP4.5 by 2071–2100
+2.5
+2.0
+1.7
+1.8
+2
[+1.1,+3.3]
[+1.1,+2.8]
[+1.3,+2.3]
[+1.4,+2.4]
[+1.4,+2.5]
RCP8.5 by 2071–2100
+4.5
+3.2
+3.1
+3.5
+3.6
[+3.8,+5.3]
[+2.5,+4.0]
[+2.5,+3.9]
[+2.7,+4.2]
[+3.0,+4.1]
(b) Precipitation changes
RCP4.5 by 2021–2050
+8.4
+7.6
+3.8
+5.6
+5.9
[+2,+17]
[+2,+14]
[-2,+9]
[-2,+14]
[+4,+9]
RCP8.5 by 2021–2050
+13.2
+10.5
+4.7
+6.8
+8.0
[+6,+22]
[+0.5,+22.9]
[+0.2,+11]
[+1,+15]
[+5,+11]
RCP4.5 by 2071–2100
+18.4
+14.8
+4.0
+6.5
+9.7
[+12,+27]
[+7,+23]
[-7,+12]
[0,+12]
[+6,+13]
RCP8.5 by 2071–2100
+26.8
+26.4
+5.2
+13.1
+15.7
[+18,+35]
[+16,+39]
[-5,+15]
[-1,+25]
[+9,+23]
Data availability
The CHASE-PL Climate Projection (CPLCP) dataset produced here was made available for use in two different ways: (1) in a long-lasting
research data repository and (2) through a dedicated CHASE-PL web geoportal. The first option
(Sect. ) is meant to serve mainly researchers, particularly users of environmental models to apply the
bias-corrected high-resolution climate data as a consistent forcing dataset for projecting climate change impacts on different
sectors in Poland. In this case, to achieve full consistency, it is recommended to use the observational (CPLFD-GDPT5) dataset
, used as a reference for model calibration and validation. The second option
(Sect. ) is expected to serve both researchers and a wider audience, including students, stakeholders, and public
authorities, as climate change science has not been disseminated widely in Poland to date .
Data repository at 4TU.Centre for Research Data
The bias-adjusted files were stored in NetCDF4 format and compiled using the
Climate and Forecast (CF) conventions. The data were made available at the
4TU.Centre for Research Data . The files
consist of nine bias-adjusted RCM simulations of daily
(minimum and maximum) temperature and precipitation for a spatial domain
covering the union of Poland and the Vistula and Odra basins for one
historical and two future time periods assuming the RCP4.5 and RCP8.5
scenarios. There are 135 files and the total size is 127 GB. The full
dataset covering the continuous time period (i.e. 1950–2100) can be obtained
upon request from the Norwegian Meteorological Institute. The CPLCP-GDPT5
dataset presented here is publicly available at
10.4121/uuid:e940ec1a-71a0-449e-bbe3-29217f2ba31d.
Access through the Climate Impact web portal
The Climate Impact web portal (http://climateimpact.sggw.pl) developed
within the CHASE-PL project presents spatial interactive data on three
aspects of climate change in Poland: (1) observations, (2) projections, and
(3) impacts. The “Observations” sub-page presents, among other things, the
5 km resolution gridded precipitation and temperature dataset
CPLFD-GDPT5 that was used in this study
as the reference dataset, aggregated to monthly/seasonal/annual time series
and long-term average values. The “Impacts” sub-page presents maps of climate
change impacts on water resources obtained
from hydrological modelling using SWAT driven by the dataset described in
this paper. In this section we focus on the “Projections” sub-page presenting
the contents of the CHASE-PL Climate Projections dataset
(Fig. ).
The web-map application was developed using ArcGIS Server, which makes the data
available using REST architecture, as well as using the temporal data
visualization portal using the JavaScript API for communication between the
client and the server. ESRI Geoportal Server was applied for meta-data
management. Two language versions, English and Polish, are available. The
geoportal stores in total 180 maps of projected variables (precipitation,
minimum and maximum temperature) for two time horizons (near and far future),
under two RCPs (4.5 and 8.5), for five temporal aggregation levels (annual
and four seasonal), and three ensemble statistics types (5th percentile,
median, and 95th percentile). All data are shown as original 5 km × 5 km raster files. Projected changes are shown, as in this
paper, as absolute differences between future and baseline periods for
temperature, and as percentage differences for precipitation. Ensemble
statistics are calculated for projected changes across all ensemble members.
By including three ensemble statistics, the geoportal informs end users both
about the magnitude and the spread of change (climate model uncertainty). The
web-map application has the following functionalities: (1) meta-data
searching, (2) searching by location, (3) identification of selected values
on the map (simultaneously for all seasons and year), and (4) data download in
NetCDF and GeoTIFF formats. The online help and glossary were also created in
order to enhance the use of the geoportal among users less advanced in web GIS
and/or climate model outputs.
Conclusions
A recent high-resolution gridded dataset (CHASE-PL Forcing Data
“CPLFD-GDPT5”) was used as long-term reference dataset produced over the
Odra and Vistula basins in Poland and surrounding regions to correct for any
systematic bias in daily precipitation and temperatures simulated by nine
EURO-CORDEX RCMs. The main purpose was to provide
up-to-date climate projections for Poland assuming the new generation of
representative concentration pathways.
Results showed that the bias correction method performed very well in
reducing the large biases found in the raw data of an ensemble of nine
EURO-CORDEX simulations.
Regarding the climate projections, we demonstrated that the climate change
signal was not affected by the bias correction method. Yet, any
misrepresentation of the former in the RCM, due, for
instance, to inherited misrepresentation of (i) the sea surface temperatures
and sea ice extent in the northern parts influenced by the Baltic Sea and
(ii)
topographical features (e.g. the mountains located in the southern parts)
could have an influence on the projected temperature and precipitation
changes . Nevertheless, we assumed that changes
in the climate parameters were less affected than their absolute values and
hence showed more robust estimates.
Based on the best estimates, projected changes in temperature and
precipitation suggest a warmer and wetter climate over Poland for the coming
decades, except for summer, during which a decrease in precipitation by less
than 7 % is also likely to occur. The warming over Poland is expected
to likely vary by 0.3 ∘C by 2021–2050 assuming the
intermediate-emission scenario. This accelerates to approximately
5 ∘C towards the end of the 21st century assuming the high-emission scenario. Similarly to temperature, precipitation over Poland is
expected to vary by between -7 and +40 %.
The highest increases in both temperature and precipitation are expected to
occur in winter.
We believe that the CHASE-PL Climate Projection product (CPLCP-CPLFD-GDPT5)
available for the period of 150 years (from 1951 to 2100) will serve as the
basis for further applications – for instance, to study the impact of climate
change over Poland on many sectors (e.g. agriculture, hydrology, ecology, and
tourism).