What drives historical and future changes in photovoltaic power production from the perspective of global warming?

We investigate the drivers of global and regional changes in the potential for photovoltaic (PV) power production from the pre-industrial (1850) to present-day (1985–2014) and until the end of the century (2071–2100), based on output from the Coupled Model Intercomparison Project phase six (CMIP6). Our assessment separates regional contributions from changes in clouds, humidity, temperature, aerosols, and wind speed to the changes in PV power potentials for the first time. Present-day PV power potentials are adversely affected by anthropogenic aerosols compared to the pre-industrial, with a global decrease of the PV power potential by − 1.3%. Our results highlight a globally averaged decrease in future PV power potentials primarily driven by temperature and humidity increases by − 1.2% to more than − 3.5%, depending on the scenario. Regionally different contributions of changes in clouds and aerosols cause heterogeneous spatial patterns in changes of PV potentials, with typically stronger (weaker) influences from clouds (aerosols) in SSP5-8.5 compared to SSP1-2


Introduction
Part of the strategy to mitigate climate change is the transition from fossil-fuel-based electricity systems to systems that primarily use renewable power sources (e.g.Owusu and Asumadu-Sarkodie 2016).Electrical power has to be readily available upon demand.In contrast to controllable power production from fossil-fuel combustion to meet the demand, electricity generation from irradiance is inherently variable due to the spatial and temporal variability of the weather (Frank et al 2021).With increasing electricity provision from weather-dependent sources, the weather becomes economically more important in trading markets for electricity (e.g.Koch and Hirth 2019).In 2021 globally, roughly 1000 TWh of solar PV power was produced and a production of about 7400 TWh by 2030 is estimated for a net zero emission scenario by 2050 (Bojek 2022).Moreover, reliable information on future climate changes affecting renewable power production is needed, e.g. for planning sites for power plants and necessary changes in transmission and storage infrastructure (Schaeffer et al 2012).Weather and climate information are therefore crucial to ensure the security of electricity supply in the future.Specifically, the amount of electricity produced by a photovoltaic (PV) module depends on the incoming shortwave radiation, the ambient temperature, and the wind speed.The latter two affect the conversion efficiency of the PV module by altering its temperature.We quantify how the global PV power potential changes with anthropogenic climate perturbations.In so doing, we identify possible knowledge gaps in the physical science basis for estimating PV power potentials, which is needed for an energy transition.
Assessing the impacts of climate changes on PV power production is not new (e.g.Crook et al 2011, Wild et al 2015, Zou et al 2019, Feron et al 2020, Danso et al 2022) (Jerez et al 2015) was used to estimate mean changes in PV potentials in the future along pathways with 2.6 W m −2 to 8.5 W m −2 radiative forcing in 2100.The spatial focus of past analyses of the total changes in PV power potentials varies from regional studies over Europe (Gaetani et  We estimate the atmospheric influences for changes in the potential for PV power production (PV power potential hereafter) from 1850 to 2100 with a novel conceptual model using output from CMIP6.The conceptual model uses CMIP6 model output from 14 preindustrial control (piControl) experiments, 154 historical experiments, and 299 experiments for three future scenarios.The approach allows us to quantify the anthropogenic influences on the PV power potential for the present-day compared to pre-industrial times, and to assess the expected impacts of global warming on future PV power production for different socio-economic pathways at the end of the 21st century.Specifically, we calculate and evaluate the contribution of changes in temperature, wind speed, and irradiance to changes in PV power potential.Changes in irradiance are quantitatively separated into contributions from changes in clouds, humidity, and aerosols to assess their relative importance for changes in PV power potentials for the first time.In so doing, we also pinpoint simulated aspects that are key to a better understanding of changes in the regional PV power potential.To that end, we combine the degree of model agreement for the identified mean changes in the PV power potential from different factors and the state-of-the-science knowledge on uncertainties in CMIP6 models.The systematic assessment allows us, for the first time, to identify where further improvements in model capabilities are needed from the perspective of renewable power production from solar irradiance.

Conceptual model
We isolate the benefits and losses in the potential PV power production associated with climate change.The calculation of the potential of PV power production is based on earlier methods (Jerez et al 2015, Bichet et al 2019) and our own derivation of equations for separating contributions to the changes in PV power production, namely for clouds, air humidity, and aerosols.We calculate the PV power production as a function of time PV(t) following Jerez et al (2015) with: where I(t) is the temporally dependent irradiance at the surface and I STC = 1000 W m −2 is the irradiance used during standard testing conditions of PV panels.
The performance ratio, P R (t), is calculated as: Here, T cell (t) is the temporally dependent temperature of the PV cell in • C, and T STC = 25 • C is the temperature used during standard testing conditions of the PV panel.We use parameters for present-day monocrystalline silicon solar panels.Their response factor is, γ = −0.005• C −1 , which is taken from Tonui and Tripanagnostopoulos (2008).We compute T cell (t) following the approach from Jerez et al (2015) with: , and c 4 = −1.528 The mean change in PV power production potentials, ∆PV, is then computed here as the difference between two thirty-year periods t 1 and t 2 as ∆PV = PV(t 1 ) − PV(t 2 ), written with equation (4) as: (5) The different terms in equation ( 5) are interpreted according to their physical meaning allowing us to quantify the net contributions from different changes in meteorological variables to the mean change in the potential of PV power production.The five terms are composed of variables describing the mean values for a particular time period, e.g. the present-day temperature, T sfc , and of variables for mean changes, e.g.changes in irradiance between two time periods, ∆I.The name convention for the terms is chosen such that the variables driving the change in PV power production potentials determine the name, e.g.∆PV I is the change in PV power production potential associated with ∆I.The magnitude of products of two changes in different variables, namely ∆PV IT and ∆PV Iu , are small compared to the magnitude of the other terms and are, therefore, not shown.
Irradiance can change due to differences in clouds, air humidity, and aerosols between two time periods.These variables do not necessarily change similarly with warming and can be associated with different levels of model agreement.We, therefore, decompose irradiance-associated changes in PV power production, ∆PV I , into net contributions from changes in clouds, aerosols, and humidity, as: describing the sum of changes in the PV power production potential due to irradiance differences associated with clouds, ∆PV cloud , aerosols, ∆PV aer , and air humidity ∆PV q .The difference in irradiance between all-sky (I) and clear-sky (I cs ) conditions describes the net effect of clouds on radiation.We therefore calculate the influence of clouds on ∆PV I with: The contribution from changes in air humidity to ∆PV I is estimated using the water vapor content: Meteorological data as input for the conceptual model are available from CMIP6 output, except for ∆I q .We, therefore, compute ∆I q with the help of our simulations with the Radiative Transfer Model for Global climate models (RRTMG, Clough et al 2005).For simplicity, and because the tests with other profiles had no substantial impact on the results for the change in PV power potentials, we use the clear-sky U.S. Standard Atmosphere 1976 (COESA 1976).We specifically performed two sensitivity tests for ∆PV q with atmospheric profiles for the tropics and the subarctic winter taken from Anderson et al (1986).The mean difference for the global (table 1) and regional (figure 3) ∆I q in the tests was <0.03% compared to the RRTMG simulation with the U.S. Standard Atmosphere.We perform multiple RRTMG simulations to create a lookup table (Servera et al 2019) for I q .The simulation setups differ concerning (1) the water vapor profile scaled by integrated water vapor of 0-90 kg m −2 with steps of 0.01 kg m −2 , and (2) different solar zenith angles from 0 • -90 • with steps of 0.5 • .
Values from the lookup table are used for I q values in ∆I q for the calculation of changes in PV power production potentials.All calculations are with monthly mean output in each grid point from CMIP6 output paired with I q values from the lookup table according to the monthly integrated water vapor content and the mean zenith angle of the latitude and month.The percentage change between two time periods is then used to estimate ∆I q following: where the indices 1, 2 refer to the two time periods for which ∆PV is computed, e.g. the end of the 21st century against the present.The residual term is ascribed to changes in aerosols from both natural and anthropogenic sources: The three scenarios therefore span a range of future warming levels and are useful to document the associated uncertainty in ∆PV at the end of the 21st century.We use output from experiments with models that have interactive aerosol parameterizations.Dust aerosols and their effects on the irradiance at the surface are for instance represented in the data, such that we account for associated losses in the potential for PV power production.
In total, these are 14 experiments from 14 models for PI, 154 experiments from 18 models for PD, 98 experiments from 14 models for SSP1-2.6,109 experiments from 12 models for SSP2-4.5, and 92 experiments from 15 models for SSP5-8.5.The larger number of experiments is due to having several realizations of the same experiment type, e.g.there are ensembles of historical experiments from single models.Due to the different ensemble sizes and spatial grids, we pre-processed the CMIP6 output before reading it as input in the conceptual model.For each CMIP6 model, we first compute the mean over all ensemble members of the same model.These model means are then interpolated from their original spatial grid to a 1 • × 1 • latitude-longitude grid with second-order conservative remapping (Jones 1999).For ∆PV, we first calculate the mean annual cycle of ∆PV for each model between two time periods and then calculate the CMIP6 multi-model mean.
Retaining the information on the temporal changes per model allows us to assess the degree of model agreement in trends across the CMIP6 ensemble.All changes are computed for each SSP against PD and for PD against PI.
Our regional means are computed as spatial averages for seven regions, namely Europe, North America, South America, South Africa, Northwest Africa, India, and Southeast Asia, to allow a more detailed assessment of month-to-month changes.The regions are chosen for their regionally different results in ∆PV.Some of the regions are known for high (dust) aerosol optical depth, which inspired us to use for these cases regions as in Pu and Ginoux (2018).Additionally, we include North America and Europe due to, as we will see in the results, the strong influence of cloud changes.The margins of the regions are listed in table B2 and marked in figure 3.

Global mean
The globally averaged multi-model mean change in PV power potential for present-day (PD) against the pre-industrial (PI) has decreased and is primarily explained by the reduced surface irradiance due to regionally different pronounced aerosol increases.We list the spatial and temporal mean changes in the PV power potential (∆PV) and the contributing terms in table 1.From PI to PD, ∆PV indicates a reduction in the global multi-model mean PV potential by −1.7%.Over land, where the majority of PV modules are located, the multi-model mean of ∆PV is more pronounced with −2.4%.The dominating contribution to the reduction in the PV potential for PD against PI stems from the increase of aerosols due to anthropogenic activities (∆PV aer ).For instance, sulfur dioxide, black carbon, and organic carbon emissions increased by several Tg yr −1 from 1850 to 2010 (Zhang et al 2016).Emissions of anthropogenic aerosols led to reduced irradiance at the surface (Wild 2009).We estimate a contribution of the anthropogenic aerosol increase to ∆PV of −1.3% in the global mean and −2.0% to the land-only mean for PD against PI.The strongest effect is seen over East Asia with up to −13.9% (figure A1) for PD against PI, consistent with the spatial patterns of PD emissions for anthropogenic aerosols.In comparison to the global contribution from aerosols, the contribution by the PI to PD changes in temperature, clouds, and humidity is one order of magnitude smaller (table 1).
Projected negative mean ∆PV at the end of the century in two of the three scenarios are primarily controlled by the degree of future warming that adversely affects the conversion efficiency of PV modules to produce power.Positive contributions from reduced clouds and negative contributions from The degree of warming to a given radiative forcing is controlled by the climate sensitivity, which remains uncertain in CMIP6 (Zelinka et al 2020).CMIP6 includes models with a higher climate sensitivity than was the case in CMIP5, although the difference in the climate sensitivity between CMIP5 and CMIP6 is statistically insignificant (Zelinka et al 2020).CMIP6 models project warmer climates for a given radiative forcing, e.g. higher summer temperatures over Europe (Palmer et al 2021).One could expect weaker signals for changes in the potential for PV power production from models with smaller climate sensitivity.Specifically, the changes in ∆PV associated with temperature changes, which are particularly pronounced for the change between SSP5-8.5 against PD, would be weaker for a smaller climate sensitivity.Reasons for the model diversity in climate sensitivity are thought to arise from model differences in cloud feedbacks (Myers et al 2021) and in ocean heat transport (Singh et al 2022).Cloud properties additionally influence the shortwave radiation transfer with an impact on ∆PV cloud .Model differences in ∆PV cloud might arise from differently simulated extra-tropical low-level cloud cover responses associated with the climate sensitivity (Zelinka et al 2020).

Regional differences
The spatial patterns of ∆PV for the future projections SSP1-2.6,SSP2-4.5, and SSP5-8.5 are mainly explained by the combined impact of ∆PV cloud and ∆PV aer (figure 1).Changes in temperature and humidity lead to a global reduction in PV potential that does not strongly change from one region to the next.The regional magnitude of ∆PV rather depends on irradiance changes due to clouds and aerosols that in some regions enhance and in other regions compensate for the adverse impacts on PV power production arising from the temperature and humidity increases.Regionally positive ∆PV in the northern hemisphere are consistent with other future estimates of changes in PV power potential on land (e.g.

Gernaat et al 2021).
The pattern of ∆PV aer over land is consistent with irradiance changes that one would expect from future changes in anthropogenic aerosols (figures 1(c), (f) and (i)).In all SSPs, the anthropogenic aerosol optical depth (AOD) decreases over East Asia, Europe, North America, and Australia from PD to the end of the 21st century (Gidden et al 2019).This decrease is most pronounced in SSP1-2.6,weaker in SSP2-4.5, and smallest in SSP5-8.5.An exception is the anthropogenic AOD over Central Africa, where the values are similar in SSP1-2.6 and SSP2-4.5 compared to PD but the values increase in SSP5-8.5 (Fiedler et al 2019).Aerosols change the attenuation of shortwave radiation (Eltbaakh et al 2012), i.e. lower (higher) AOD leads to increasing (decreasing) shortwave radiation at the surface.This is broadly consistent with the positive ∆PV aer over China, Europe, North America in all SSPs against PD, and with the negative ∆PV aer over Central Africa for SSP5-8.5 against PD.Over China, the reduction of ∆PV aer seen for PD against PI is projected to be substantially weakened by the end of the 21st century with a positive ∆PV of up to +16% in SSP1-2.6 and up to +12% in SSP5-8.5.Individual CMIP6 models qualitatively agree on the sign of the aerosol contributions due to similarly prescribed anthropogenic aerosol emissions or concentrations, indicated by the spatial correlation coefficients for ∆PV aer across the models with values around 0.9 for SSP1-2.6,0.85 for SSP2-4.5 and 0.8 for SSP5-8.5 (figure 2).
Over most land regions, ∆PV cloud is positive in all SSPs against PD.Over Europe and South America ∆PV cloud increases from SSP1-2.6 to SSP5-8.5 from values around +1% to more than +4%.The spatial correlation and the root-mean-square-differences (RMSD) across the CMIP6 models for ∆PV cloud indicate relatively larger model differences than for the other components affecting ∆PV.These model differences in ∆PV cloud become larger with warming, as seen by comparing the different scenarios (figure 2).This behavior indicates model-dependent responses of clouds to future warming with adverse impacts on the capability to assess future PV power from CMIP6 simulations.
Interestingly, there is negative ∆PV cloud over polar regions in contrast to positive signs in most other world regions.Around Antarctica, ∆PV cloud has values of up to −5%, −7% and −12% for SSP1-2.6,SSP2-4.5, and SSP5-8.5, respectively (figure 1).Moreover, the regional ∆PV aer around Antarctica is negative in all SSPs compared again PD, which coincides with higher surface wind speeds in the Southern hemisphere mid-latitude storm track (Priestley and Catto 2022) indicative of more sea spray aerosols (not shown) and therefore regionally higher aerosol optical depth.This regional change in winds could be associated with changes in the large-scale circulation, e.g. the broadening of the Hadley Cells, which is positively correlated with global warming (Frierson et al 2007, Nguyen et al 2015, Zhou et al 2020, Hur et al 2021).Also in the Arctic, negative ∆PV aer occur in SSP5-8.5 paired with a particularly strongly negative ∆PV cloud , indicative for regional changes in ∆PV being at least in parts driven by irradiance changes due to aerosols and clouds and not primarily the large regional ∆PV T , due to Arctic amplification, through the temperature influence on the conversion efficiency of PV modules (e.g.Kawajiri et al 2011, Gernaat et al 2021).
Taken together, the future spatial patterns of ∆PV are qualitatively similar to PD, but the regional magnitudes depend on the CMIP6 scenario (compare figure 1 against figure A1).Future mean PV power potentials might most strongly decrease in polar regions, over the tropical oceans, western North America, and northern Africa, while an increasing potential is projected for Europe, eastern North America, and China.Our results for the future ∆PV are consistent with regional results from other studies.For instance, the positive ∆PV with warming over European regions is consistent with regional studies (Crook et al 2011, Gaetani et al 2014, Wild et al 2015, Müller et al 2019, Feron et al 2020, Hou et al 2021).For western North America, we find that the PV potential is decreasing for SSP5-8.5 because of the

Seasonal differences
Understanding seasonal differences in PV power production is important to satisfy the seasonally varying electricity demand, e.g. for heating in winter when irradiance is naturally lower than during summer.We find that regional changes in the annual cycle of ∆PV from PI to PD is dominated by changes in aerosol-induced changes in irradiance (figure 3).A change in the annual cycle of ∆PV aer is evident over Europe, North America, and Northwest Africa, with a maximum absolute change in summer when irradiance and AOD are high (compare figure 3 against figure A2).In contrast to the mid-latitudes, ∆PV aer shows a constant offset of −3% to −5% independent of the month over sub-tropical regions of Southeast Asia and India where seasonal changes in irradiance are smaller than at higher latitudes.
The reduction of PV by aerosols since PI is reversed in almost all regions for SSP1-2.6 and to a lesser extent for SSP2-4.5 and SSP5-8.5 (figure 3).An exception is Northwest Africa, where the aerosol load increases in SSP5-8.5 and reduces the PV potential, especially in the summer months when the temperatures are the largest and the demand for electricity for cooling is potentially high.Here, dust aerosols play a major role in the regional aerosol burden, yet the response of dust emissions in a warming world is insufficiently understood from CMIP-class models (Evan et al 2014, Kok et al 2023).We infer that our estimate of the regional contribution to the PI to PD change in the ∆PV over Northwest Africa is conservative, given that observed changes in dust aerosols are larger than simulated by the models (Kok et al 2023).
From PD to the future scenarios for 2070-2100, the change in the annual cycle of ∆PV due to clouds (∆PV cloud ) is largest in summer in northern hemisphere mid-latitudes with again impacts scaling with the degree of warming (figure 3).In the mid-latitudes, we also see a summertime negative contribution from humidity (∆PV q ) and temperature (∆PV T ) partially balancing the positive ∆PV cloud in Europe, North America, and China.The large contribution of ∆PV cloud in the mid-latitudes during summer makes here the response of clouds to warming crucial to project ∆PV in summer when temperatures are the highest and hence the electricity demand might be increased, e.g. for operating cooling systems during heat waves.

Conclusions
Global climate change impairs PV power production, with global mean changes in the potential PV power production (∆PV) of −0.3% for the present-day compared to the pre-industrial, and up to −2.4% for the end of the 21st century in the future scenario SSP5-8.5 against present-day.Although the percental differences in the global mean might seem moderate, the absolute differences in the magnitude of PV power production have important implications.Most importantly, ∆PV are globally not uniformly distributed and are largest over some land regions with large populations that demand electricity.It might be possible to balance the adverse impacts of climate change on the potential for PV power production through future technological advances and combining PV power with other technologies for renewable power production.The offset of climate change impacts via emergent PV technologies are for instance assessed by Saxena et al (2023).
Over most land regions and scenarios, the contributions from changes in clouds and aerosols to ∆PV are positive, yet uncertainties in the understanding of such changes remain from CMIP data (Bock and Lauer 2023, Zelinka et al 2023).Changes in temperature and humidity are comparatively well understood, reflected in our results by a high model agreement on associated changes in regional ∆PV.However, we see a large model spread in changes of ∆PV arising from changes in clouds and aerosols.Both changes in aerosols and clouds regionally and temporally contribute by different magnitudes in ∆PV and also influence each other, e.g. through aerosol-effects on cloud microphysical processes (e.g.Twomey 1977), and rain formation by clouds leading to wet deposition of aerosols.Yet aerosol effects are still uncertain (Bellouin et al 2020) and, for instance, accurately simulating rainfall in CMIP-class models remains an unresolved challenge (Fiedler et al 2020).
The magnitude of future ∆PV scales with the degree to which the Earth system is warming and regionally depends on changes of clouds with warming, primarily in summertime mid-latitudes of the Northern hemisphere.Although the representation of clouds has improved from CMIP5 to CMIP6 (Vignesh et al 2020), simulating clouds is a challenge in models with parameterized convection (Bony et al 2015).It is reflected in our results by often poor model agreement on the sign of the regional ∆PV associated with cloud changes with future warming.
Moreover, past changes in dust aerosols are not fully understood nor reproduced by CMIP6 models paired with large model differences in future projections of dust emissions (Zhao et al 2022, Kok et al 2023), i.e. the future development of dust aerosols and their impact on ∆PV is an open question.One promising approach to overcome the challenge is the use of kilometer-scale models for global (e.g.Slingo et al 2022) or regional (e.g.Kendon et al 2019) experiments, which can better simulate dust-emitting winds associated with dynamical processes on the mesoscale (e.g.Tegen et al 2013).Such simulations running with a few kilometer grid spacing are via the link to clouds and aerosols interesting for future assessments of ∆PV and other weather-dependent renewable power production.
al 2014, Jerez et al 2015, Müller et al 2019, Hou et al 2021), Africa (Gaetani et al 2014, Soares et al 2019, Danso et al 2022) and Australia (Poddar et al 2021) to global assessments (Crook et al 2011, Wild et al 2015, Zou et al 2019, Feron et al 2020, Dutta et al 2022).Global assessments quantified climate change impacts on PV power production for different future scenarios alongside estimates for other renewable sources (Gernaat et al 2021).Recent work also considered losses associated with the installation and operation of PV modules (Saxena et al 2023).Atmospheric influences on PV power production are substantial, but not fully understood.The temperature impact on the PV power potential has been known for some time (Kawajiri et al 2011).Contributions from changes in temperature, wind speed, and irradiance to PV power potentials have been estimated (e.g.Jerez et al 2015, Bichet et al 2019, Poddar et al 2021, Danso et al 2022).However, separating influences on irradiance, e.g. from clouds and aerosols on global scales, for understanding their relative contributions to PV power changes has not been done to date, although for Europe examples separating contributions of irradiance changes associated with clouds exist (Müller et al 2019, Hou et al 2021).Especially, the contribution of the individual changes from aerosols and humidity was not explicitly quantified before.Both aerosol and clouds are large uncertainties in our understanding of climate change (e.g.Bony et al 2015, Bellouin et al 2020), which affects PV power estimates with a currently unknown magnitude.We aim to fill this knowledge gap by estimating the separate atmospheric influences on changes in PV power potentials from the global perspective, based on data from the contemporary model output from the Coupled Model Intercomparison Project phase six (CMIP6, Eyring et al 2016).

Figure 1 .
Figure 1.Future changes in potential PV power production.Shown are (left, ∆PV) the change in total PV power production, (middle, ∆PV cloud ) the contributions from clouds to ∆PV, and (right, ∆PVaer) the contribution from aerosols to ∆PV.All values are calculated with equations (5)-(10) and shown as percental changes for 2071-2100 in the scenarios (top to bottom) SSP1-2.6,SSP2-4.5, and SSP5-8.5 against 1985-2014 in the historical CMIP6 experiments.The hatched areas mark low (<80%) model agreement for the sign of changes.

Figure 2 .
Figure 2. Taylor diagrams for PVaer and PV cloud .Shown are (circles) PVaer and (crosses) PV cloud of individual CMIP6 models against the CMIP6 multi-model mean as reference for (a) 1985-2014 against 1850, and 2071-2100 from (b) SSP1-2.6,(c) SSP2-4.5, and (d) SSP5-8.5 against 1985-2014.We follow Taylor (2001) and show both variables in the same diagram for easier comparison.Note the different standard deviations of the two variables and hence the two RMSD grids.

Figure 3 .
Figure 3. Annual cycle of ∆PV and its contributing terms as means (left to right) for 1985-2014 (PD) against 1850 (PI), and for 2071-2100 (SSP1-2.6,SSP2-4.5, and SSP5-8.5)against PD (rows) for seven regions.The regions are indicated in the left corner of the subplots and their margins are listed in table B2.The bold line shows the CMIP6 multi-model mean and the shaded areas the CMIP6 model-to-model differences measured by the 25th and 75th quantiles.

Figure A2 .
Figure A2.Like figure 3, but for the change in surface shortwave downwards radiation (∆I) in Wm −2 , change in aerosol optical depth (∆AOD) times 100 (to use the same y-axis) and change in cloud fraction in %.
. Previous studies of projected PV power potential have used climate model output from CMIP3 (Crook et al 2011), CMIP5 (Wild et al 2015, Müller et al 2019, Zou et al 2019) and CMIP6 (Hou et al 2021, Dutta et al 2022).Moreover, model output from the Coordinated Regional Climate Downscaling Experiment (CORDEX, Giorgi et al 2009) for Africa (Bichet et al 2019, Soares et al 2019) and Europe

Table B2 .
Margins of the areas shown in figure3.Additionally, the table lists the mean change of potential PV in % (∆PV) in the selected regions for PDagainst PI (1850), and SSP1-2.6,SSP2-4.5,andSSP5-8.5 (2071-2100)against PD.Shown are the CMIP6 multi-model means ± the model-to-model standard deviation.