Changes in photovoltaic potential over China in a warmer future

Simulated climatic data are derived from 25 km-resolution dynamically downscaled climate simulations in the East Asia domain, produced by the latest version of the Regional Climate Model version 4 (RegCM4; Giorgi et al 2012) within the Coordinated Regional Climate Downscaling Experiment-Coordinated Output for Regional Evaluations framework (CORDEX-CORE; Giorgi et al 2022). RegCM4 is driven by three GCMs, i.e. the Met Office Hadley Centre Earth System model (HadGEM2-ES; Jones et al 2011), the Max Planck Institute for Meteorology Earth System Model (MPI-ESM-MR; Giorgetta et al 2013), and the Norwegian Earth System Model (NorESM1-M; Bentsen et al 2013), with corresponding downscaled simulations named HdR, MdR, and NdR, respectively. Here, we used daily mean surface downward solar irradiance (R_s), near-surface temperature, and near-surface wind speed from historical simulations (1980–2005) and future projections (2006–2098) forced by the representative concentration pathway (RCP) 8.5 scenario (van Vuuren et al 2011). The main configuration of RegCM4 simulations is summarized in table S1. Here, we did not apply a bias correction method because it is unlikely to have an impact on the relative differences between the future and reference period of our interest (Ehret et al 2012, Maraun 2016).

In addition, the daily mean of hourly outputs from the fifth generation of the European Center for Medium Weather Forecasting atmospheric reanalysis (ERA5; Hersbach et al 2020) is employed as the reference data for evaluating the model results. The ERA products (ERA5 and its predecessor) are the official validation databases for the CORDEX downscaling initiatives (Gutowski et al 2016) and have been widely used as a reference in previous studies (e.g. Poddar et al 2021, Martinez and Iglesias 2022, Wu et al 2022). The ERA5 is shown to be outperforming other reanalysis products in estimating R_s throughout China (He et al 2021, Zhang et al 2021). We did not use observed R_s from meteorological stations because of the concerns about the uneven distribution of stations, missing records, and inhomogeneity issues due to station migration and changes in calculation methods (Wang et al 2015).

In the present study, we estimated the PV power potential (PV_POT) using the energy rating method (Mavromatakis et al 2010, Crook et al 2011), where PV_POT is defined as the fraction of the power output under standard conditions that a PV module may exhibit in the field:

$$\begin{equation}P{V_{{\text{POT}}}} = {P_{\text{r}}}\frac{{{R_{\text{s}}}}}{{{R_{{\text{STC}}}}}}\end{equation} \tag{ 1 }$$

where ${R_{{\text{STC}}}}$ is the solar irradiance at Standard Test Conditions (STC), which is equal to 1000 W m⁻². ${P_{\text{r}}}$ is referred to as the performance ratio, which accounts for changes in the efficiency of the PV cells due to changes in weather conditions. According to previous studies (Jerez et al 2015, Feron et al 2021, Sawadogo et al 2021), ${P_{\text{r}}}$ can be calculated as:

$$\begin{equation}{P_{\text{r}}} = 1 - \gamma \left( {{T_{{\text{cell}}}} - {T_{{\text{STC}}}}} \right)\end{equation} \tag{ 2 }$$

where ${T_{{\text{STC}}}}$ is the ambient air temperature at STC (25 °C), and the constant γ can be taken as equal to 0.5% °C⁻¹ in the case of monocrystalline silicon cells. ${T_{{\text{cell}}}}$ (°C) is the solar cell temperature, which is expressed as a function of solar irradiance (R_s), air temperature (T), and surface wind speed (v) as:

$$\begin{equation}{T_{{\text{cell}}}} = {c_1} + {c_2}T + {c_3}{R_s} + {c_4}v\end{equation} \tag{ 3 }$$

where ${c_1}$ = 4.3 °C, ${c_2}$ = 0.943, ${c_3}$ = 0.028 °C W⁻¹ m², and ${c_4}$ = −1.528 °C m⁻¹ s. Note that the above calculation presents the theoretical PV potential. The tilt of PV panels is not considered here. In addition, other technological and politico-economical aspects are not considered.

Although PV_POT is estimated based on daily average T, R_s, and v here due to the absence of high temporal resolution outputs from climate models, we recognize that those variables may change during the day. Given the context, based on ERA5, we compared PV_POT estimated based on daily average datasets and those estimated based on hourly datasets, and the difference in terms of both spatial pattern (figure S1) and temporal variation (figure S2) was confirmed to be small, which further consolidates our strategy in estimating PV_POT.

To assess the possible effects of individual variables on the change in PV_POT, we rewrote the expression of PV_POT by merging equations (1)–(3), as follows:

$$\begin{equation}P{V_{{\text{POT}}}} = {\alpha _1}{R_{\text{s}}} + {\alpha _2}R_{\text{s}}^2 + {\alpha _3}{R_{\text{s}}}T + {\alpha _4}{R_{\text{s}}}v\end{equation} \tag{ 4 }$$

with ${\alpha _1}$ = 1.1035 × 10^–3, ${\alpha _2}$ = −1.4 × 10^–7, ${\alpha _3}$ = −4.175 × 10^–6, and ${\alpha _4}$ = 7.64 × 10^–6, in the corresponding units (PV_POT should be dimensionless). Thereby, changes in PV_POT are given as follows (Jerez et al 2015):

$$\begin{align}\Delta P{V_{{\text{POT}}}} &= \Delta {R_s}\left( {{\alpha _1} + {\alpha _2}\Delta {R_{\text{s}}} + 2{\alpha _2}{R_{\text{s}}} + {\alpha _3}T + {\alpha _4}v} \right)\nonumber \\ & \quad + {\alpha _3}{R_{\text{s}}}\Delta T + {\alpha _4}{R_{\text{s}}}\Delta v + {\alpha _3}\Delta {R_{\text{s}}}\Delta T \nonumber \\ & \quad + {\alpha _4}\Delta {R_{\text{s}}}\Delta v.\end{align} \tag{ 5 }$$

According to equation (5), the change in PV_POT due to a change in T can be obtained by taking $\Delta {R_{\text{s}}}$ = 0 and $\Delta v$ = 0, and the change in PV_POT due to the influence of changes in R_s (or v) alone can be derived similarly.

We considered extreme low daily PV power outputs to be those falling below the 10th percentile of the probability density function (PDF) of the daily PV_POT anomalies (the departure of the daily PV_POT value from the daily base climatology), in which the PDF is estimated within a 15 day rolling window of the PV_POT data over the reference period (1986–2005). This definition is inspired by the methodology used for assessing changes in the probability of occurrence of extreme events (Zhang et al 2011, Perkins-Kirkpatrick and Lewis 2020). Here, we focused on three metrics, i.e. PV10, PV10N, and PV10D (table 1), to quantify the potential change in the events of extreme low PV power outputs. We demonstrated that the effect of using daily average datasets instead of hourly datasets on the calculation of these three indices is small (figure S3). In addition, the method for computing PV10 was also applied for sequentially computing the share of days with extreme low solar irradiance (Rs10, according to the 10th percentile), the share of days with extreme high temperature (T90, according to the 90th percentile), and the share of days with extreme low wind speed (W10, according to the 10th percentile). Following previous studies (Sutanto et al 2020, Feron et al 2021), we created binary maps for the days that exhibited extreme low PV_POT (defined according to the 10th percentile) and extreme low irradiances (defined according to the 10th percentile), which enabled us to assess the number of days with concurrent extremes low PV outputs and extreme solar irradiance. Similar analysis can be performed for other concurrent extremes.

Here, we examined PV_POT changes at specific global warming levels, i.e. 1.5 °C, 2 °C, and 3 °C above the pre-industrial level (1861–1880) based on a time sampling method. Among the above warming levels, 2 °C and 1.5 °C are the Paris Agreement goals, whereas 3 °C is similar to late-century warming projected based on current mitigation policies (Rogelj et al 2016, Liu and Raftery 2021). Given that RCMs do not provide global datasets, we use outputs from the GCMs, which initiate the RCM runs, to calculate the timing of global temperature reaching the global warming targets. Following previous studies (Zhang and Zhou 2020, Zhang and Wang 2022), the warming thresholds are determined using the 11 year running average global mean surface air temperature for each GCM. The specific warming periods are then aggregated over the 11 year windows that are centered on the years when respective warming levels occur (table 2). Note that the results regarding specific global warming thresholds in the present study represent a transient response, which is different from the stabilized response at the same global warming level (King et al 2020, Zhang et al 2022a).

Future changes relative to the reference period (1986–2005) are derived from the difference between projections and historical simulations for individual RCMs. The statistical significance of changes is evaluated using Wilcoxon's rank-sum test imposing a significance level of 0.05. The ensemble-mean signal is the arithmetic mean of the three individual signals from RCMs. If at least one significant individual signal differs in the sign of the projected change, it is considered as 'uncertain'; else if the uncertainty condition is not fulfilled and less than half of the individual signals are significant, it is considered as 'negligible'; otherwise, the ensemble mean signals are referred to as 'robust'. To examine the regional characteristics of solar resources, we divided China into four subregions (figure 1(a)), i.e. the Tibetan Plateau (TP), northwestern China (NWC), northeastern China (NEC), and southern China (SC).

Zoom In Zoom Out Reset image size

Before examining future changes, we evaluated the performance of the dynamically downscaled simulations in reproducing the present-day spatial pattern of annual mean R_s and PV_POT over China. During the reference period (1986–2005), the highest R_s over China is on the TP, followed by NWC, while the lowest is in the west-central part of SC. The spatial pattern of PV_POT generally matches that of R_s (figures 1(a) and (h)), suggesting that western China is favored for its abundant PV power, whereas central and eastern China are the least favorable for PV power generation. The spatial patterns of both R_s and PV_POT are considerably captured by all three dynamically downscaled simulations, with a spatial correlation coefficient between RCMs and ERA5 ranging from 0.69 to 0.79 for R_s and 0.80–0.85 for PV_POT (figures 1(a)–(d), (h), (i) and table S2). All three RCMs slightly overestimate both R_s and PV_POT over China by a few percent (table S2), regarding the spatial distribution of biases, R_s and PV_POT are overestimated in SC and parts of NEC but are generally underestimated in NWC and over the TP (figures 1(e)–(g) and (l)–(n)). A rigorous evaluation of models' capability in reproducing surface solar irradiance over China is somehow out of the scope of the present study and has been reported in the previous literature (Wu et al 2022); however, the different RCMs provide similar and reasonable results, indicating good reliability and stability of the current model ability, which builds more confidence for future projection.

There are a few percent changes in R_s over China in a warmer future, projected by dynamically downscaled simulations. An important finding is that regions with high R_s (e.g. TP and SWC) under today's climate tend to experience a decrease in R_s in a warmer future, and vice versa (figures 2(a)–(c)). Although a broadly similar phenomenon has also been projected by an ensemble of GCMs participating in phase 5 of the Coupled Model Intercomparison Project (Wild et al 2015, Yang et al 2018), the RCMs employed here demonstrate a more refined spatial pattern and are expected to be of higher accuracy (Wu et al 2022). Moreover, projected changes in surface air temperature and wind speed, the other two variables that affect PV, are displayed in figure S4. Under a warmer climate in the future, the increase in temperature is most pronounced in high latitudes and high altitudes, while the signal of changes in wind speed is generally negligible. Under the combined effect of the above factors, changes in annual mean PV_POT show a similar pattern to those in R_s, but with differences in magnitude (figures 2(d)–(f)).

Zoom In Zoom Out Reset image size

In general, the mean PV potential over China decreases with escalating global warming in the future (figure 3(a)). Specifically, the national average changes in PV_POT relative to the reference period in 1.5 °C, 2 °C, and 3 °C warming futures are −1.1%, −1.5%, and −2.6%, respectively, based on the ensemble mean, implying that if the global warming level is limited to the Paris Agreement goals, the adverse impact of climate changes on PV_POT is expected to be avoided to a considerable extent. In terms of spatial distribution, the decrease in PV_POT is robust in all subregions, except for SC. The TP is projected to experience the most pronounced decrease in PV_POT, with magnitudes of −1.8%, −2.8%, and −4.4% in 1.5 °C, 2 °C, and 3 °C warming futures, respectively. This is followed by NWC and NEC. In terms of spatial distribution (figures 2(d)–(f)), under the 3 °C warming scenario (similar to late-century warming projected based on current mitigation policies), the northwest half of China tends to experience a robust decrease in PV_POT, while this signal is negligible in the southeastern half of China. In addition, under the 2 °C and 3 °C warming scenarios, the decrease in R_s is negligible in the northern and eastern parts of NEC, but this is not the case for PV_POT (figures 2(c) and (f)), which can be mainly explained by the negative effect of elevated temperature (figure S4) on PV_POT (equations (2) and (3)). It is noted that the projected decrease in PV_POT over the TP is robust even under the lowest 1.5 °C warming scenario. The above results indicate the adverse impact of escalating global warming on solar energy in China, suggesting the need for proactive action to mitigate climate change. Another finding is that the currently solar-rich regions in China, such as the TP and parts of NWC (Feng et al 2021, Qiu et al 2022), are expected to suffer the most remarkable decrease in PV_POT in a warmer future, although the magnitudes are confined to a few percentages, which is insufficient to change the overall spatial pattern of solar PV resources over China (figures 1(i)–(k) and S5).

Zoom In Zoom Out Reset image size

We further examine changes in events of extreme low PV power outputs. By definition, PV10 should be close to 0.1 over the base period; however, PV10 estimates calculated over a certain period in the future may be notably different due to changes in the climate. Specifically, PV10 is projected to increase in the vast majority of China in a warmer future, and the spatial range with robust changes in PV10 variation is wider than that with robust changes in annual mean PV_POT (figures 4(a)–(c)). The changes in PV10 over China relative to the reference period are 16.0%, 22.3%, and 36.1% in 1.5 °C, 2 °C, and 3 °C warming futures, respectively, based on the ensemble mean. All subregions will experience a robust increase in PV10, except for SC. The projected increase in PV10 over the TP reaches more than twice the national average, while that over NWC and NEC is slightly below the national average (figure 3(b)). PV10N and PV10D are also projected to substantially increase under future scenarios (figures 3(c) and (d)), and correspondingly, the changes in PV10N (PV10D) over China relative to the reference period are 40.2% (20.7%), 52.2% (24.4%), and 87.9% (35.0%) in 1.5 °C, 2 °C, and 3 °C warming futures, respectively. Regarding the spatial distribution, changes in PV10, PV10N, and PV10D show similar patterns in general, although with different magnitudes (figures 4(c)–(i)). The most pronounced increase in events of extreme low PV power outputs locates in the TP, followed by northern China (north of 40° N) and southern China coast, while the rest of the region shows a relatively moderate change.

Zoom In Zoom Out Reset image size

In addition to changes in mean climate state, changes in time variability of PV_POT value are also essential for power outputs, especially since persistent low PV power output affects the network stability and even causes power outages. Although previous studies have investigated projected variabilities of PV_POT at different time scales (expressed as a standard deviation; Jerez et al 2015, Bichet et al 2019, Müller et al 2019), the present study focuses on extreme low daily PV_POT events, i.e. those falling below the 10th percentile of the probability distribution of the daily PV_POT anomalies, shows a more intuitive result. The increasing frequency of extreme low PV output days would make it harder to maintain the production/consumption balance within an electrical grid, and further compromise the economic sustainability of PV projects (Patt et al 2013, Feron et al 2021).

PV cells directly convert sunlight into electricity using a semiconductor junction. According to equations (1)–(3), PV energy yields depend on the downwelling shortwave irradiance that is modulated by aerosols and clouds (Zou et al 2019, Li et al 2020, Danso et al 2022), but are also affected by air temperature and surface wind speed. Specifically, cooler conditions generally improve the performance of PV cells and vice versa, whereas air flow typically cools the PV module, thereby improving the performance of PV cells. Although the explanation for mechanisms of the PV potential changes is limited in this work because the relevant physical quantities are not available in CORDEX-CORE outputs, we nevertheless calculate the contributions of individual variables to the projected PV_POT change based on the equation (5). Here, we take the results under the scenario of 3 °C warming as an example, whereas similar but weaker signals are noted under the scenarios of 2/1.5 °C warming (figure S6). The spatial pattern of R_s-induced changes (figure 5(a)) is similar to that of overall changes (figure 2(f)), implying that future changes in PV_POT are dominated by spatially uneven changes in R_s. Moreover, the elevated air temperatures (T) also contribute to the decrease in PV_POT with relatively uniform spatial distribution (figure 5(b)). Compared to the first two, the effect of wind speed (W) is generally negligible (figures 5(c) and (d)). For the national average, the effect of R_s and T on the decrease in PV_POT is comparable; however, the relative contributions of R_s and T vary markedly across subregions. The decrease in R_s superimposed on the negative impact of T on PV outputs results in a remarkable decrease in PV_POT over the TP, while in SC, the positive effect of increasing R_s is offset by the negative impact of T on PV outputs (figure 5(d)). Moreover, regarding the change in days with extreme low PV power outputs in the future, it can be seen that the days with extreme low PV power outputs mainly coincide with those with extreme low solar irradiance, followed by those with extreme high temperatures (figures S7–S8). The substantial increase in extreme low irradiance events in a warmer future dominates the increase in events of extreme low PV power outputs, while the escalating extreme high-temperature events also contribute slightly.

Zoom In Zoom Out Reset image size