1 Introduction

Over the past decade, the cost of renewable energy has been declining steadily. The weighted average of the levelized cost of electricity at the global level fell by 15% for wind and 13% for solar energy in 2021 alone (Irena 2022). Over the past 5 years, the cost of solar energy has been declining at the rate of 13% per year while it was 7% for onshore wind energy (Lazard 2020). These trends have the potential to incentivize carbon neutrality and contribute to tackling the climate crisis by reducing reliance on fossil fuels (Pfeifer et al. 2021; Luderer et al. 2021).

To date, the burning of fossil fuels remains one of the largest producers of carbon greenhouse gasses contributing ~ 66% of global CO2 emissions (Foster and Elzing 2015). Evidence suggests that in less than a couple of decades, anthropogenic activities have doubled the atmosphere’s CO2 content since the beginning of industrialization (NASA 2022). The presence of CO2 in the atmosphere warms the earth and this leads to synoptic-scale changes in the climate system (Herwartz et al. 2021; Hartley and Turnock 2021). Most of these climatic changes which include extreme rainfall events such as floods and droughts impose cascading risks on lives, critical infrastructure, and property. Therefore, a shift to clean energy contributes to achieving Sustainable Development Goals Number 7 and 13 which stress the importance of ensuring access to affordable clean energy through climate action (UN 2015).

While critiques of renewable energy cite the spatiotemporal variability of wind and solar radiation as a weakness to a dependable electricity grid (Diesendorf and Elliston 2018), this shortcoming can be overcome by renewable energy mixes including wind, water, and solar power. Furthermore, technological advancements such as innovative energy storage, and optimized management techniques show promise of turning wind and solar energy resources into dependable electricity grids. Indeed, renewable energy sources like solar and wind energy are constantly being replenished at a higher rate than what humans can consume.

Many countries in the developed and developing world are turning to renewable energy as a pathway to climate change mitigation. The European Union for instance is accelerating the take-up of renewables to significantly contribute to the reduction of net greenhouse gas emissions by at least ~ 55% before the end of 2030 (EU 2022). In Africa, many countries including but not limited to South Africa, Nigeria, and Kenya have also set strong emissions-reduction targets. Zambia has also committed to reducing emissions by 25% by 2030 (USAID 2015). Overall, Africa has committed to cutting 32% of emissions by 2030 (Abudu et al. 2023).

With these emissions-reduction ambitions, global renewable energy is expected to expand rapidly. Projections show that by 2024, the global capacity of renewable energy will increase by 50%, and 60% of this will be sourced from solar photovoltaic resources (IEA 2019). During the same period, additional 22 gigawatts of renewable energy are expected across sub-Saharan Africa alone (IEA 2019). While these ambitions and projections will have a profound effect on CO2 emissions, it is notable that climate change, in turn, will also affect the future variability of wind and solar energy resources, affecting the ability of countries to produce electricity using renewables. Therefore, studies that assess future changes in climatic variables are indispensable for future renewable energy systems (Troccoli et al. 2018).

The overarching objective of this study is to explore future variations of climatic variables that are relevant to future photovoltaic solar power resources (PVRes) in Zambia. While much of Zambia experience roughly similar climate characteristics due to the plateau that characterizes the country’s topography (Fig. 1A), a few climatic differences exist, and these can be classified into four main categories of the Köppen–Geiger classification (Peel et al. 2007):

  1. 1.

    Tropical Savanna The Tropical Savanna climate which is classified as Aw in the Köppen–Geiger classification covers Kalabo district, parts of Shang’ombo, and Mongu in the Western Province of Zambia (Fig. 1B). In the Eastern Province, Katete, Petauke, and parts of Chipata are also classified as Tropical Savana. These areas generally experience a pronounced dry season characterized by monthly rainfall averaging 60 mm (Africa Groundwater Atlas 2019).

  2. 2.

    Arid Steppe This climate zone covers the semi-arid region of Livingstone, Kaloma, Choma, and parts of the Luangwa valley (Fig. 1B). While the rainfall in these areas is not as low as that of desert climates, it is usually less than potential evapotranspiration and can, thus, be described as semi-arid.

  3. 3.

    Temperate with dry winters (generally June–August, see Marshall 2017) and warm summers The Cwb climate of Zambia mainly covers the northern tip of the country bordering the Democratic Republic of the Congo and Tanzania (Fig. 1B). These areas include Kaputa, Mpulungu, and Mbala. Given their high elevations, temperatures are usually lower than across the rest of the country.

  4. 4.

    Temperate with dry winters and hot summers Classified as Cwa in the Köppen–Geiger classification (Peel et al. 2007), this climate zone covers the rest of the country which is mainly characterized by dry winters and wet summers.

Fig. 1
figure 1

Overview of the study area showing: A the location of Zambia in Southern Africa (green square). The grey shading indicates topographical variations across the region based on the Global Land 1 km Base Elevation (GLOBE) digital elevation model (Hastings and Dunbar, 1999). The blue shading shows the location of major water bodies, B climatic zones of Zambia were developed using the Climatic Research Unit Time Series Version 3.21 (CRU TS 3.21) dataset produced and maintained by the Climatic Research Unit of the University of East Anglia (Jones and Harries 2013). The precipitation and temperature CRU data used to produce the Climatic Zones of Zambia are for the period 1951–2010 (Africa Groundwater Atlas 2019)

Full size image

2 Data and methods

2.1 Data

2.1.1 Reference datasets

To perform the present analyses, we explored three different atmospheric variables which include air temperature measured at 2 m above sea level (TAS), wind speed measured at 10 m above sea level (WS), and downwelling surface shortwave solar radiation (SR).

TAS was sourced from the latest version of high-resolution monthly data (CRU TS v4.05) from the Climatic Research Unit of the University of East Anglia (Harris et al. 2020; Table 1). The dataset covers all land areas across the globe apart from Antarctica and is gridded at a resolution of 0.5°. CRU TS v4.05 was developed using angular-distance weighting to interpolate in situ data from a dense network of meteorological stations across the globe (Harris et al. 2020).

Table 1 Summary of the reference datasets used in this study. The 1981–2005 period used in this study was chosen to align with the availability of climate model datasets
Full size table

For wind speed, we used the fifth-generation global climate reanalysis (ERA5). ERA5 is a product of the European Centre for Medium-Range Weather Forecasts (ECMWF) and was developed by assimilating in situ data into a global weather forecast model (Hersbach et al. 2020). ERA5 has a 0.1° × 0.1° horizontal resolution.

We retrieved SR from the archives of TerraClimate. This is also a high-resolution dataset gridded at 4 km (Abatzoglou et al. 2018). TerraClimate was developed by merging climatological normals from WorldClim version 1.4 and version 2 datasets (Fick and Hijmans 2017), CRU TS v4.0 (Harris et al. 2020), and JRA-55 (Ebita et al. 2011).

2.1.2 Climate models

In this study, we analyzed CORDEX-CORE models. Unlike ordinary CORDEX models, the CORDEX-CORE initiative is an improvement in terms of the horizontal resolution and a general homogenization of simulations across different CORDEX domains (Gutowski et al. 2016). CORDEX-CORE Regional Climate Models (RCMs) can, therefore, be thought of as a homogeneous set of high-resolution projections across all CORDEX domains driven by a common set of General Circulation Models (GCMs).

The CORDEX-CORE models used in the present study are for the historical period 1981–2005 and the future period 2025–2100. The models used include the Germany Climate Service Centre’s REMO model (Jacob and Podzun 1997), the Consortium for Small-Scale Modeling (COSMO) community’s CCLM (Rockel et al. 2008), and the Abdus Salam International Center for Theoretical Physics’ RegCM4 (Giorgi et al. 2012). The horizontal resolution of the CORDEX-CORE models is 0.22° × 0.22°. Despite the significant improvements the CORDEX-CORE initiative introduces, our literature review shows that no published peer-reviewed studies exist evaluating photovoltaic solar power resources (PVRes) across Zambia. As such, PVRes across Zambia remain largely unstudied, leading to major uncertainties and a lack of understanding of how climate change will affect renewable energy potential in the future.

It is important to note that CORDEX-CORE RCMs are only available for two Representative Concentration Pathways (RCPs), i.e., the low-end RCP2.6 and the high-end RCP8.5 scenario (Giorgi et al. 2022). RCP 2.6 is an ambitious target of keeping global temperature increments below 2 °C by the close of the twenty-first century (IPCC 2014). To achieve the RCP2.6 target, emissions should decline and reach zero by 2100. On the other hand, the RCP8.5 is a business-as-usual scenario; emissions continue to rise and as such, it is considered a worst-case scenario (Meinshausen et al. 2011). In this study, we examined both scenarios over the period 2025–2100 (Table 2). We retrieved all simulations from the Earth System Grid Federation (ESGF) node of the German Climate Computing Centre (DKRZ).

Table 2 Parameterizations of the CORDEX-CORE models used in this study
Full size table

2.2 Methods

2.2.1 Conceptual framework and reliability metrics

To establish a reliable understanding of future PVRes, the RCMs used must realistically reproduce the observed historical climate. Therefore, the analysis starts by evaluating the ability of CORDEX-CORE models to simulate TAS, SR, and WS as they are key to the generation of PVRes and, thus, to the renewable energy sector as a whole. Based on data availability, we used the 1981–2005 reference period that is available for the observational and model datasets. This reference period is widely used in renewable energy studies (Costoya et al. 2021; Ogunjobi et al. 2022). To ease the evaluations, we used Climate Data Operators (CDO) to regrid all datasets to a common resolution of the CRU TS v4.05 reference data.

Several methods of model evaluation exist; some are simple and straightforward while others are more complex. For example, some previous studies used percent overlap (PO), arguing that its simplicity in using probability density functions (PDFs) as a statistical measure of how well models reproduce observed PDFs of variables of interest makes it a useful tool (Perkins et al. 2007). In the present study, however, we focus on the ability of models to simulate spatiotemporal patterns as this is of importance when deciding where to install solar farms. In addition, it helps understand the seasonal cycle of solar energy potential with maximum/minimum PVRes to be expected. Using this performance criterion, we were able to exclude simulations whose outputs seemed unrealistic, e.g., those that produce output that is abnormal and in disagreement with the current understanding of the climate of Zambia. Spatiotemporal dynamics of climatic variables are widely used as a performance criterion for models (Cattiaux et al. 2013; Bartók et al. 2019; Vautard et al. 2019).

To compute the reliability metrics, we used the Satellite Application Facility on Climate Monitoring (CM SAF; Kothe 2022). CM SAF is a toolbox that automates the calculation of several model evaluation metrics in R Programming Language (R Core Team 2020). Some of these metrics include correlation (R), root mean square error (RMSE), and mean absolute error (MAE). A summary of the process we followed is given in Fig. 2.

Fig. 2
figure 2

Conceptual framework followed in the present study. CRU, Climate Research Unit; ERA5, European Centre for Medium-Range Weather Forecasts Reanalysis version 5. REMO, Regional Model; GERICS, The Germany Climate Service Center; COSMO, the Consortium for Small-Scale Modeling; RegCM, Regional Climate Model; TAS, ambient temperature; WS, wind speed; SR, downwelling surface shortwave radiation; Tcell, cell temperature; PVRes, solar photovoltaic energy resources, far future: 2025–2055; near future: 2075–2100

Full size image

2.2.2 Calculation of solar photovoltaic energy resources

The energy received from the sun touches the earth’s surface as shortwave solar radiation and is part of the electromagnetic radiation spectrum (Blal et al. 2020; Spiridonov and Ćurić, 2021). Solar radiation consists of visible light, infrared, ultraviolet light, X-rays, gamma rays, and radio waves. Ultraviolet and visible light together form shortwave downward radiation. It is the only one that has enough energy density to produce electricity when it reaches a solar cell (Costoya et al. 2022). To calculate solar photovoltaic energy resources (PVRes) in this study, we considered the amount of shortwave downward radiation at the surface and corrected it based on the efficiency of PV cells which generally reduces with increasing temperature (Radziemska 2003). Our calculation of PVRes was based on the work of Jerez et al. (2015) who studied the impact of climate change on PVRes. Mathematically, PVRes can be expressed as follows:

$${\text{PV}}_{{{\text{Res}}}} = P_{{\text{R}}} *{\text{RSDS}},$$
(1)

where RSDS is the shortwave downward radiation at the surface given in W/m2 and PR is a performance ratio that considers the effect exerted by temperature on the efficiency of PV cells (Jerez et al. 2015). PR can be expressed mathematically as follows:

$$P_{{\text{R}}} = 1 + \gamma (T_{{{\text{cell}}}} - T_{{{\text{STC}}}} ).$$
(2)

Here, \(\gamma\) refers to the power thermal coefficient for mono-crystalline silicon cells which indicates how strongly the PV power output is dependent on the temperature of the cells. Here, we use a constant value of − 0.005 °C (Tonui and Tripanagnostopoulos 2008). It is negative because the power output is inversely proportional to increasing cell temperature. TSRC refers to the temperature of the cell under standard test conditions and has a constant value of 25 °C. Tcell refers to a multiple regression model which considers the effects of solar radiation, temperature, and wind speed (Chennai et al. 2007). It can be expressed as shown in Eq. 3:

$$T_{{{\text{cell}}}} = c_{1} + c_{2} *{\text{Tas}} + c_{3} *{\text{RSDS}} + c_{4} *{\text{WS}},$$
(3)

where Tas is the air temperature around the cells and is given in °C. RSDS is the downward solar radiation at the surface (W/m2), and WS is the near-surface wind speed (m s−1). C1, C2, C3, and C4 are coefficients that are dependent on the properties of the PV materials used. According to Jerez et al. (2015), we apply the following values:

C1 = 4.3 °C,

C2 = 0.943,

C3 = 0.028 °C m2 W−1

C4 = − 1.528 °C s m−1

2.2.3 PVRes trend analysis

While there are several trend detection methods existing (e.g.,Adamowski et al. 2009; Sonali and Nagesh Kumar 2013), the non-parametric Mann–Kendall trend test (MK test) and its modified version are inarguably the most used in climate research (Tesche and Kumar 2016; Agbo et al. 2022). The modified MK test is especially preferred because it ensures that the time-series data under study are not serially correlated and this prevents the detection of false trends (Hammed and Rao 1998).

In the present study, we employed the modified MK test to quantify the evolution of PVRes trends. To do this, we used the ‘modifiedmk’ Package (Patakamuri and O’Brien 2021) in R Programming Language (R Core Team 2020). In modified MK tests, the equivalent normal variants of the rank of the de-trended series are obtained using Eq. 5:

$$Z_{i} = \varphi^{ - 1} \left( {\frac{{R_{1} }}{n + 1}} \right){\text{for}}\quad i = 1:n,$$
(5)

where Ri refers to the rank of the de-trended time series, n refers to the length of the time series, and \(\varphi^{ - 1}\) is the inverse standard normal distribution function with a mean of 0 and a standard deviation of 1 (Hammed and Rao 1998).

2.2.4 Bias correction techniques

In an effort to increase confidence in future PVRes projections, we applied linear scaling (LS) to simulations. LS is a widely used bias-correction technique that considers the difference between mean reference data and model outputs and then applies it to simulations (Ines and Hansen 2006; Shrestha et al. 2015).

3 Results and discussion

3.1 Regional climate model evaluation

During the 1981–2005 reference period, the Regional Climate Models REMO, CCLM, and their ensemble mean reproduce wind speed, TAS, and SR with spatial correlations ranging between 0.5 and 0.7 (Fig. 3 and Table 3). At ~ 2.3 ms−1, the RMSE for wind speed is in the order of the mean wind speed suggesting a generally poor performance across the models. A temperature bias of ~ 3 °C and SR bias of ~ 22 W/m2 further adds credence to the observation made with wind speed, i.e., the state-of-the-art models exhibit substantial deficiencies in reproducing the observed climate of Zambia. These observations point to the need for a thorough bias correction before analyzing any future trends.

Fig. 3
figure 3

Spatial patterns of mean annual observed and modeled near-surface wind speed at 10 m (ms−1; first column), downwelling surface shortwave radiation (W/m2; middle column), and near-surface air temperature at 2 m (°C; right column) during the 1981–2005 period

Full size image
Table 3 Summary of the correlation coefficients between mean annual observed and modeled near-surface wind speed at 10 m (ms−1), downwelling surface shortwave radiation (SR, W/m2), and near-surface air temperature at 2 m (TAS, °C) during the 1981–2005 period
Full size table

Notwithstanding the observed deficiencies, REMO, CCLM, and their ensemble mean show above-average performance, especially considering that they can capture the spatial distribution of the reference dataset with generally higher wind speeds of up to 4 m s−1 across the central parts of Zambia while the rest of the country experiences ~ 3 m s−1. Further, the two models and their mean show a downward gradient of SR increasing from the north of the country to the south. While these spatial distributions are captured correctly by the models, it is noticeable that they tend to overestimate SR. Overall, the reference dataset shows SR values of up to 195 W/m2 around Luapula Province and over the Mwinilunga area of Northwestern Province, but the models and their mean simulate values of ~ 205 W/m2 (Fig. 3). Regarding TAS, models accurately simulate higher values of 26 °C over the Eastern and Southern Provinces of the country similar to the observational dataset (Fig. 3).

While it is noted in this study that all models struggle with reproducing historical climate, REMO and CCLM perform far better than RegCM. For instance, for TAS and SR, R values for REMO are up to 0.7 (TAS, RMSE and MAE are 1.5 °C and 2.5 °C, respectively; SR, RMSE, and MAE = 25 W/m2 and 24 W/m2, respectively), and up to 0.5 for WS. For CCLM, R values for TAS and SR go up to 0.6 (TAS, RMSE and MAE are 2.6 °C and 1.6 °C, respectively; SR, RMSE and MAE = 25 W/m2 and 24 W/m2 respectively). The above-average performance of the two models (i.e., REMO and CCLM) has been documented by other studies such as Ndiaye et al. (2022) who studied future changes in solar PV and wind energy potential over West Africa and found that REMO and CCLM behaved very well across the Guinean coast, Sudan, and Sahelian climatic zones. It is notable, however, that while RegCM performs well across other regions, challenges to simulate wind speed, SR, and TAS are very vivid across Zambia (Fig. 3). The RegCM model notices no climatic variations across the whole country for all three variables. This observation was also made earlier with wind speed simulations (Libanda and Paeth 2022) and may be related to model parameterization, local climate features which RCMs cannot reproduce due to discrepancies in resolution, or observational errors (Wilcke et al. 2013). Given these unrealistic simulations by RegCM, we excluded it from further analyses.

The remaining two RCMs were found to reproduce the observed annual cycle of wind speed, SR, and TAS differently (Fig. 4). The annual cycle of TAS is well captured by the RCMs with the highest values of about 25 °C from October to November and plummeting to ~ 18 °C in June/July (Fig. 3C). Both RCMs tend to overestimate SR and wind speed (Fig. 4A, B). With an average of 3 ms−1, REMO consistently exhibited a positive bias of 20% compared to the reference dataset. CCLM, on the other hand, shows very good performance with a positive bias of only 4%. At an average of 2.7 ms−1, the mean of the models was 12% higher than the reference dataset. Positive biases are again seen with SR; for instance, REMO was found to exhibit a positive bias of ~ 11% while CCLM again exhibits better performance with a bias of only ~ 8%. The mean of the two models (i.e., REMO and CCLM) was found to be 10% higher than the reference dataset. Overall, both the models and the reference dataset exhibit the lowest SR values around July which constitutes the coldest month in Zambia. The low SR values in July can, therefore, be attributed to the fact that the sun's rays are generally more slanted during July than in other months (Marshall 2017). Regarding TAS, exceptional performance was found with REMO showing a near-zero bias of − 1% while CCLM was found to have − 4% bias. Taken together, the multi-model ensemble exhibited a bias of only − 2% (Fig. 4).

Fig. 4
figure 4

Mean annual cycle of observed and modeled a downwelling surface shortwave radiation (W/m2), b near-surface wind speed at 10 m (ms−1), and c near-surface air temperature at 2 m (°C) during the 1981–2005 period, averaged across longitude 21°–34° E and latitude 17.4°–7.6° S

Full size image

Taken together, the biases found in this work are very minor compared to those found in West Africa where, e.g., wind speed overestimations were as high as 50% (Ndiaye et al. 2022). While interrogating the causes of biases observed herein was beyond the scope of the present study, it is well established that model biases can be caused by a range of factors which may include limited spatial resolution, hence failure to read microscale meteorological processes, simplified thermodynamic processes, and utter incomplete understanding of the physics of the global climate system (ECMWF 2022).

Given the considerable overestimation of solar radiation during austral winter, a systematic overestimation of wind speed especially by REMO, and some minor errors in the seasonal temperature cycle, we calculated cell temperature (Tcell) and PVRes during the reference period (1981–2005) and compared the performance of models to the reference data. At first sight, the ability of models to accurately reproduce present-day climate would give credence to their ability to project future PVRes. We found strong and statistically significant correlations of up to 0.8 for Tcell (Fig. 5). Similar to the reference dataset, both models and the mean show the highest Tcell values of ~ 30 °C along the Luangwa Valley. Regarding PVRes, we noticed that all models simulate the spatial pattern correctly with the highest values over Livingstone and much of the Southern Province. However, some large biases are also noticeable, especially over the Copperbelt where the reference dataset shows PVRes of 190 W/m2 while models overestimate it at ~ 210 W/m2. This, therefore, suggests that models can accurately depict areas that are expected to have higher future PVRes although a noticeable systematic overestimation of up to ~ 10 ± 2% should be expected. After applying bias correction, substantial improvements of R values reaching 0.8 are observed (Fig. 5 Column 3). This finding re-echoes the earlier observation that a thorough bias correction prior to analyzing any future trends would be required.

Fig. 5
figure 5

Same as Fig. 2, but for cell temperature (°C; Tcell i.e., column 1), photovoltaic solar power resource (PVRes; W/m2, column 2), and bias-corrected photovoltaic solar power resource (PVRes; W/m2, column 3)

Full size image

The seasonal evolution of Tcell shows that all the models accurately depict month–month changes with all variations correctly mimicked although differences in the mean are evident (Fig. 6). Notably, models underestimate Tcell in January and February which represents the core of the rainy season in Zambia (Musonda et al. 2020); this underestimation could be related to models struggling to reproduce cloud cover and precipitation during the rainy season. It is estimated that during the January/February period, close to 70% of the country is characterized by cloudy to overcast conditions punctuated by rapid intermittent periods of sunshine (Nkolola et al. 2021). These cloudy conditions dictate energy budgets with ripple effects on the variability of other climatic variables, especially the amount of solar radiation reaching the earth’s surface.

Fig. 6
figure 6

Mean annual cycle of observed and modeled Tcell (°C) during the 1981–2005 period, averaged across longitude 21°–34° E and latitude 17.4°–7.6° S

Full size image

The ability of models to simulate PVRes is lower than that of Tcell (Fig. 7). We found a good representation in terms of the basic structure of the annual cycle but a substantial overestimation from March to November. Notwithstanding these biases, the lower bounds during the rainy season in January and February are simulated correctly: PVRes based on the reference dataset ranges from 165 to 229 W/m2 while that of CCLM, REMO, and the mean are 163–259 W/m2, 164–250 W/m2, and 165–255 W/m2, respectively (Fig. 7). These results suggest that future projection based on these models will give reliable ranges for minimum solar radiation whereas the maximum is fairly overestimated.

Fig. 7
figure 7

Mean annual cycle of cloud cover (%), observed and modeled photovoltaic solar power resource (PVRes; W/m2) during the 1981–2005 period, averaged across longitude 21°–34° E and latitude 17.4°–7.6° S

Full size image

Substantial biases of the maximum values of solar radiation reported herein were also found in West Africa where CORDEX-CORE models were used to study photovoltaic potential (PVP; Ndiaye et al. 2022). However, while RCMs overestimate PVRes in Zambia from March to November, they were found to underestimate it in West Africa, pointing to a regionally inhomogeneous bias in cloudiness across Africa. To get a reliable estimate of PVRes projections, we applied a delta change bias correction to the mean and found a 9.2% improvement (Fig. 7).

3.2 Future evolution of solar photovoltaic energy resources (PVRes)

3.2.1 Future temporal evolution of PVRes

For the future, results indicate that at an average of 237 W/m2 and 212 W/m2, respectively, RCP2.6 leads to 12 ± 3% more PVRes than RCP8.5 (Fig. 8). However, an exploration of the future seasonal cycle of mean cloudiness does not show major differences between RCP2.6 and RCP8.5 (Fig. 8); therefore, the observed less PVRes in RCP8.5 can be attributed to higher near-surface temperature, thus making PV less efficient. However, no major changes are expected with regard to the annual cycle. Under RCP2.6, January and February are still expected to be characterized by relatively low PVRes values ranging between 149 and 235 W/m2 while the highest PVRes values of up to ~ 300 W/m2 are expected in September. Similarly, under RCP8.5, the lowest PVRes values are expected in January and February ranging between 130 and 180 W/m2.

Fig. 8
figure 8

Projections of the annual cycle of photovoltaic solar power resource (PVRes; W/m2) for the period 2025–2100 and mean annual cycle of cloud cover for low, medium, and high clouds (%), averaged across longitude 21°–34° E and latitude 17.4°–7.6° S

Full size image

These findings suggest that regardless of the emission scenario, seasonality will be the major factor affecting PVRes in the future (Wiltberger et al. 2009). One key observation is that the January and February period when the country is expected to experience the lowest PVRes coincides with the time the country is characterized by lower wind power densities which also peak around August coincidentally with PVRes (Libanda and Paeth 2022). Thus, our findings embodied point toward a wind–solar energy mix for enhancing production during peak times, but a substitutionary energy source such as hydropower may be necessary during months of low production. The peak of wind speed around August is also advantageous to enhance PVRes because wind cools solar panels, making them more efficient (Gökmen et al. 2016). It is important to note that solar panels generate electricity with sunlight and not heat. Atoms get energized to a higher level in the presence of sunlight, thus leading to electricity generation (Tisdale et al. 2010). However, atoms tend to vibrate faster in a hot solar cell than in a cooler one. Electrons in a solar panel whose temperature is cooled by 1 °C are 0.05% more efficient (Glenn 2019), thus the cooling effect of wind on PVRes is beneficial.

The year-to-year variations indicate a very negligible but statistically significant decrease in PVRes of 0.02 W/m2 under RCP2.6 and 0.005 W/m2 under RCP8.5 (Fig. 9). The highest PVRes values are expected toward the middle of the century reaching their peak of ~ 316 W/m2 under RCP2.6 between 1930 and 1960 and 281 W/m2 between 1950 and 1970 under RCP8.5. Similar observations were made in the Sahel-Savannah region where progressive PVRes decreases of up to 2% were found, becoming more pronounced and reaching a high of − 3.5% in the far future (Ndiaye et al. 2022). The decrease of PVRes especially toward the end of the century can be explained by the expected temperature increments toward 2100. Although the projected PVRes decrease will affect the size of future solar projects in Zambia, it is unlikely to have a significant effect on their viability (Bichet et al. 2019).

Fig. 9
figure 9

Projections of the spatial mean of photovoltaic solar power resources (PVRes; W/m2) for the period 2025–2100 averaged across longitude 21°–34° E and latitude 17.4°–7.6° S

Full size image

3.2.2 Future spatial evolution of PVRes

Results of the future spatial trend of PVRes indicate that although on average Zambia is expected to experience slight reductions (Fig. 10), there are spatial differences across the country with some places such as Eastern Province being expected to experience a negligible positive annual trend of up to 0.003 W/m2 (Fig. 10). Both RCP2.6 and RCP8.5 agree on the trend of PVRes across Eastern Province, they, however, disagree across Western Province where RCP2.6 projects a slight upward trend while RCP8.5 projects a downward trend (Fig. 10A, B, respectively). Given the marginal differences, what stands out from these results is that future changes of PVRes in Zambia are likely to be near zero.

Fig. 10
figure 10

Projections of the future spatial trend of PVRes (W/m2) across Zambia for A RCP2.6 and B RCP8.5 for the period 2025–2100

Full size image

4 Concluding thoughts

We set out to explore future variations of climatic variables with a focus on changes in future photovoltaic power resources (PVRes) over Zambia. We used CORDEX-CORE simulations over the Africa domain to study future PVRes for the period 2025–2100. The main findings of this study can be summarized as follows:

  • Regional Climate Models REMO, CCLM, and the mean reproduce wind speed, SR, and TAS better than RegCM.

  • Temporally, RCMs were found to accurately reproduce the annual cycle of wind speed, SR, and TAS although overestimations are evident, thus necessitating bias corrections before analyzing future simulations.

  • RCP2.6, a greener and climate-friendly pathway, points toward 12 ± 3% better renewable energy across Zambia compared to RCP8.5.

  • Spatially, some positive trends are evident especially across Eastern Province although these are very marginal, thus suggesting that future changes of PVRes in Zambia are likely to be near zero.

Given seasonal fluctuations of PVRes, the present analysis showed that a wind-solar energy mix has a lot of promise for enhancing production during peak times, but a substitutionary energy source may be necessary during months of low production. While this study has highlighted key future spatiotemporal PVRes trends in Zambia, it suffers from the lack of robust reliable observational datasets and a limited number of models. Furthermore, it is important to note that the multi-model ensemble used herein does not factor in the influence of aerosols on PVRes projections. Aerosols are well known to be strong modulators of SR (Mallet et al. 2021) and clouds (Deroubaix et al. 2022), consequently PVRes especially in arid environments (Neher et al. 2019). Nevertheless, although RegCM inherently has an aerosol module, its unrealistic simulation of the past climate of Zambia suggests that including it in the ensemble would lead to larger biases. It is also notable that types of aerosols included in RegCM (e.g., Sea salt and dust; Solmon et al. 2006; Zakey et al. 2006, 2008) are not common in Zambia. Future PVRes investments in Zambia should, therefore, be undertaken after addressing these limitations thoroughly.