Accelerating the energy transition towards photovoltaic and wind in China – Nature

Geospatial data in this study

An overview of our optimization model is shown in Extended Data Fig. 1. We optimized the placement and capacity of PV and wind power plants in our model driven by geospatial data (Supplementary Method 1), including land cover, solar radiation, wind speed, surface air temperature, ground slope, latitude and longitude of the installed PV panels, terrestrial and marine ecological reserves, water depths of offshore stations and marine shipping routes (Supplementary Table 2). All land pixels were categorized into forest, shrubland, savannah, grassland, wetland, cropland, urban and built-up land, mosaics of natural vegetation, snow and ice, deserts and water bodies⁴⁶. The suitability of the installation of PV panels or wind turbines was defined by land cover (Supplementary Table 3). Onshore wind turbines with the capacity of 2–2.5 MW and offshore wind turbines with the capacity of 5–10 MW are considered as the main models used in China at present⁴⁷, so we considered models for onshore (General Electric 2.5 MW) and offshore (Vestas 8.0 MW) wind power plants (Supplementary Table 4) at a hub height of 100 m above the ground to convert air kinetic energy to electricity based on the recommended power-generation curve⁴⁸ (Supplementary Fig. 3). Resources of solar and wind energy were associated with seasonal and diurnal variabilities and interannual differences. We estimated hourly solar radiation and wind speed at a hub height of 100 m above the ground as averages for 2012–2018 to provide a representative estimate of solar and wind energy in China (Supplementary Method 2). All geospatial data were projected to a resolution of 0.0083° in latitude and 0.033° in longitude for estimating the potential of power generation by installing PV panels or wind turbines in each pixel (Supplementary Methods 3–5).

The optimization model

We estimated the LCOE of the PV and wind power systems to indicate the grid parity of power generation, which is defined as the normalized net present value of all costs of investments, O&M, land acquisition, transmission and energy storage divided by the power generated during the lifetime (25 years (ref. ³⁰)) of the PV and wind power plants³⁵. Before solving the optimization problem, we sought the best strategy for installing PV panels or wind turbines with different shapes to achieve the maximal capacity of power generation in each county (Supplementary Fig. 4). We took the number of pixels installing PV panels or wind turbines and the time to build each PV or wind power plant by decade as decision variables. By accounting for the intertemporal dynamics of learning^26,30, we developed a unique method to optimize the capacity of each power plant, the order of building power plants, the time to build each power plant and the option of energy storage when building a new power plant by solving a cost-minimization problem based on the LCOE for generating power by the projected PV and wind power plants:

in which ϵ is a new power plant (ϵ = 1 to 3,844), x is a power plant built before ϵ, n_x is the number of pixels installing PV panels or wind turbines in plant x, t_x is the time to build plant x, s_x is the option of energy storage (1 for pumped hydro and 2 for chemical batteries) when building plant x, T is the average lifetime of a power plant, h is hour, q is a region (1–7 for Central China, East China, North China, Northeast China, Northwest China, South China and Southwest China, respectively), n_q is the number of power plants in region q, r_d is the discounting rate (5%)², τ_p is a year in the operation of plant x, τ_g is a year during operation of energy storage in plant x, L_g is the lifetime of storage (50 years for pumped hydro⁶ and 15 years for chemical batteries⁴⁹), E_ϵ is the total power generation, V_ϵ is the total investment in power plants, A_ϵ is the total cost of electricity transmission, G_ϵ is the total cost of storage, R_y is the ratio of O&M costs to investment costs (1% for PV⁵⁰ and 3% for onshore and offshore wind power plants⁵¹), V_x is the investment in plant x, ξ_x is the ratio of cost reduction by learning when building plant x, E_x,h is the hourly generation of power in a county, Θ_h is the hourly transmission of electricity, Λ_h is the hourly storage of electricity, η_tra is the fraction of electricity lost during transmission, η_store is the fraction of electricity lost during storage, M_q,h is the hourly consumption of electricity in region q and U_q,h is the hourly transmission of electricity from other regions to region q.

We optimized the increase in power capacity at an interval of 10 years during 2021–2060 because it generally takes 10 to 20 years for new technologies to be widely applied⁵². Given the variation of renewable energy within a decade, we performed a sensitivity experiment by optimizing the model at an interval of 5 years, in which the installed PV and wind power capacity and total costs both change moderately (Extended Data Fig. 8). Nevertheless, simulating the penetration of renewable energy within a decade will be useful to improve the optimization model.

We considered the connection of power plants in a county to one of the substations from the UHV transmission lines projected in the CFED plan⁷ (Supplementary Data Set 1) with the costs of building new UHV lines (Supplementary Method 6) and developing systems for storing energy (Supplementary Method 7). We estimated U_q,h using three assumptions. First, the electricity generated by all PV and wind power plants in a region is used to meet the power demand in this region as a priority. Second, the extra electricity is connected to a substation in the UHV line and transmitted to a region in which the next substation in the line is located. Third, the electricity transmitted to a region is distributed to each county based on the distribution of the consumption of electricity in this region. We have considered the impact of transmission access on where and when to build new PV and wind power plants. When optimizing the construction time of 3,844 PV and wind power plants, we have accounted for the costs of building new UHV transmission lines when a new power plant is built, which influences the LCOE of this plant and the construction time of all new plants. By optimizing the construction time of each new power plant, we considered that a new UHV line required for this new plant will be constructed at the same time.

As a caveat of this study, we do not have explicit information for all UHV transmission lines, so we assumed that the UHV lines projected by the central government are used as a proxy for UHV lines between the main regions in the country. This assumption is useful to determine the demand for electricity transmission between regions, but it could lead to bias in our cost estimation because of the lack of detailed information for all UHV lines. For example, the projection of 128 UHV lines with a capacity of 12 GW each from Huaidong to Wan’nan in our model indicates that at least a total transmission capacity of 1,536 GW is required for transmitting electricity from the region centred in Huaidong to the region centred in Wan’nan, but the ultimate UHV lines built between these two regions might be different from our prediction. This limitation can be addressed when the detailed information for all UHV transmission lines are available. To consider the outflow of electricity generated in a county, we sought the substation of UHV lines that is closest to this county and then we estimated the cost of electricity transmission from this county to the transmission substation and the cost of electricity transmission using one of the UHV lines. Although this study projected the construction of a large transmission capacity to optimize power systems, it is important to account for the physical, technical and economic constraints. These include the demand for advanced polymer matrix composites that can operate under a voltage of >1,000 kV, the construction of UHV lines over challenging terrains, the maintenance of these lines and ensuring the security of electricity transmission under extreme weather conditions.

We sought the optimal system for storing energy when building a new power plant using either mechanical storage (pumped hydro) with a lifetime of 50 years and a round-trip efficiency of 70% or chemical storage (batteries) with a lifetime of 15 years and a round-trip efficiency of 85% (see the parameterization of two systems in Supplementary Table 5) to minimize the LCOE (Extended Data Fig. 9).

Last, ξ_x is calculated as a function of the total capacity of installed PV or wind power (Supplementary Method 8) based on the measured rates of learning in China (Supplementary Table 1). We examined the sensitivity to adopting the international rates of learning in our model (Fig. 2c).

Capacity and costs of power generation

For a new PV or wind power plant x, the annual generation of power was calculated:

in which i is a pixel, j is the number of hours in a year and W_i,j is the hourly generation of power in a pixel installing PV panels (calculated in Supplementary Method 3), onshore wind turbines (calculated in Supplementary Method 4) or offshore wind turbines (calculated in Supplementary Method 5). The parameters used to estimate the projected PV and wind power generation are listed in Supplementary Table 6.

The investment cost of a new PV or onshore wind power plant x was calculated⁶:

in which i is a pixel, P_i is the installed capacity of PV panels (calculated in Supplementary Method 3) or onshore wind turbines (calculated in Supplementary Method 4), S_i is the area of pixels installing PV panels or wind turbines, P_fix is the capacity of a voltage transformer (300 MW), μ_fix is unit capital costs, μ_land is unit cost of land acquisition, μ_line is unit cost of line connection and μ_tran is unit cost of voltage transformation.

We assumed that the installed voltage transformer has a capacity of 300 MW (ref. ⁵³). When estimating the unit cost of land acquisition, we considered that onshore wind turbines take up only 2% of area in a pixel and PV panels take up 100% of area in a pixel⁵⁴. We derived μ_fix as the sum of costs for modules (μ_module), inverters (μ_inverter), mounting materials (μ_mounting), secondary equipment (μ_sec), installation (μ_ins), administration (μ_adm) and grid connection (μ_grid) using the data for PV panels ($0.64 per watt) published by the China Photovoltaic Industry Alliance⁵⁵ and using the data for onshore wind turbines ($0.68 per watt) from a previous study⁵⁶. We demonstrated the impact of using different capital costs by examining the sensitivity to adopting high capital costs ($0.73 and $0.88 per watt for PV panels and onshore wind turbines, respectively)³⁵ or low capital costs ($0.23 and $0.76 per watt for PV panels and onshore wind turbines, respectively)³⁶ in the sensitivity tests (Fig. 2c).

The investment cost of an offshore wind power plant x was calculated on the basis of the distance of offshore wind turbines in this power plant to the onshore power station⁵⁷:

in which i is a pixel, μ_baseline is the unit cost of offshore wind turbines, P_i is the installed capacity of offshore wind turbines (calculated in Supplementary Method 5), D_L,i is the distance of the offshore wind turbines to the onshore power station and D_P,i is the water depth of the installed offshore wind turbines. The coefficients z₀ (0.0057), z₁ (0.7714), z₂ (0.0084) and z₃ (0.8368) were calibrated using engineering data⁵⁷. The parameters used to determine the costs of PV and wind power generation are listed in Supplementary Table 7. The parameters used to determine the costs of UHV transmission and energy storage are listed in Supplementary Table 8.

We adopted a fixed ratio of O&M costs to investment costs for the projected PV and wind power plants^50,51. We adopted 25 years (ref. ³⁰) as the average lifetime of PV or wind power plants. We considered the costs of electricity transmission by UHV when increasing the installed capacity of a power plant. We sought the geographic centre among all pixels suitable for power generation and then increased the number of surrounding pixels (n_x) installing PV panels or wind turbines. The capacity of power generation by each power plant increases as the number of pixels installing PV panels or wind turbines increases in the order of the distance to the geographic centre. The inclusion of more pixels in a power plant, however, increases not only the capacity of this PV or wind power plant but also the total costs in the power systems. The LCOE for a new power plant first decreased when we increased the power capacity by increasing the number of pixels for installation of PV panels or wind turbines, because the capital costs were divided by the generated power, but then increased owing to the increasing costs of purchasing land and the decreasing power-use efficiency (Extended Data Fig. 2). We sought the optimal capacity for each power plant for achieving the minimum of the LCOE.

MAC

We assumed that the electricity generated by new PV and wind power plants was used to replace oil, gas and coal in the order of fuel price to produce the highest profits⁶. Solving the cost-minimization problem in equation (1) was constrained by the target of the annual abatement of CO₂ emissions by substituting fossil fuels when a new PV or wind power plant ϵ was built (F_ϵ):

in which E_ϵ is the total power generation, S_x is the area of pixels installing PV panels or wind turbines, θ_fossil is the CO₂ emission factor of coal (0.84 kg CO₂ kWh⁻¹), oil (0.72 kg CO₂ kWh⁻¹) or gas (0.46 kg CO₂ kWh⁻¹)⁵⁸ that is substituted by PV and wind power, γ_x is the flux of the terrestrial carbon sink disaggregated from a bottom-up estimate⁵⁹ and v_x is the concentration of soil carbon in lands covered by vegetation⁶⁰ transferred to PV panels or wind turbines (Supplementary Table 9).

We derived the MAC for a new PV or wind power plant ϵ, MAC_ϵ, based on the abated CO₂ emissions:

in which ϱ is the price of coal, oil or gas. We obtained the prices of coal ($0.043 ± 0.015 per kWh as the 95% confidence interval)⁶¹, oil ($0.141 ± 0.057 per kWh)^62,63 and gas ($0.058 ± 0.016 per kWh)⁶⁴ in China as the averages during 2010–2020, when they are considered to generate electricity with an efficiency of 35%, 38% and 45%, respectively⁶⁵. The generation of power by fossil fuel in the 2010s in China was dominated by coal, with a contribution⁶⁶ of 96% in 2020, but the composition of fossil fuel in the future is as yet unknown for China. We assumed that the future fossil-fuel composition in China will be constant in our central case, but we performed two sensitivity experiments to consider the impact of changes in fuel composition. First, the share of oil for generating power increases from the current level in 2020 (0.3%) to 50% in 2060 by substituting coal when the share of gas is identical to that in the central case. Second, the share of gas for generating power increases from the current level in 2020 (4.1%) to 50% in 2060 by substituting coal when the share of oil is identical to that in the central case (Extended Data Fig. 6).

Hourly power demand

We predicted the hourly power demand by 2060 based on the flexibility of power loads by sector (Supplementary Table 10). First, we scaled up the historical hourly power loads from electrical grids⁶⁷ in 2018 by the increase in total power demand under the projected rate of electrification³³ in 2060 (58%) for six non-power sectors, including agriculture, industry, transport, building, service and household electric appliances, in 31 provinces. We assumed that the power loads are flexible for agriculture, industry, building, service and household electric appliances, except for heating and cooling in houses and electric cars, so we could simulate the profiles of the hourly power demand to match the hourly power generation by PV and wind endogenously in our optimization model. Second, we predicted the hourly power demand by electric cars. We obtained the profile of traffic flow in each street every five minutes in 2018 in Shenzhen⁶⁸, which was assumed to represent the variation of traffic flow in the future owing to a lack of data for other cities in China. When electricity is used by electric cars, one-third of vehicles are charged immediately, one-third are charged in one hour and one-third are charged in two hours²⁵.

Third, we predicted the hourly power loads for heating and cooling in houses. We obtained the hourly energy used for space heating and cooling in houses by region in China⁶⁹. Finally, we considered the impact of temperature on the power demand for heating and cooling in houses and electric cars based on the projected temperature under climate warming. The electricity used for heating increases by 0.98% as the annual average temperature decreases by 1 °C when the hourly temperature is below 16 °C (ref. ⁷⁰), whereas the electricity used for cooling increases by 0.63% as the annual average temperature increases by 1 °C when the hourly temperature is above 28 °C (ref. ²⁵). We predicted the hourly power demand for heating and cooling for 31 provinces for 2021–2060 based on the gridded hourly temperature⁷¹ averaged for 2016–2020 and the projected change in annual average temperature during 2021–2060 under the SSP1-2.6 scenario from an Earth System Model¹⁷. The power demand shifts in the daytime to match the peak of hourly PV and wind power generation (Supplementary Fig. 5).

Actual PV and wind power plants

We adopted a pixel resolution of 1 × 3 km² for installing PV panels or wind turbines, which allows us to predict the location and capacity of individual PV and wind power plants in our optimization model. We cannot validate the locations and capacities of the projected PV or wind power plants that had not yet been built, so we used the locations and capacities of the commissioned PV and wind power plants in OpenStreetMap³¹ that were closest to the projected PV or wind power plants to evaluate our prediction (Extended Data Fig. 3). We estimated the geographical distance of locations between the projected and actual PV and wind power plants. A full validation of our optimization model required detailed information on the PV and wind power plants when they are to be built in the coming decades, so we only compare the projected capacities of power plants normalized by current area with the actual capacities of the existing power plants in OpenStreetMap³¹.

Distribution of income

We estimated the impact of finances embodied in the flow of electricity generated by new PV and wind power plants on the redistribution of income in 2060. First, we estimated the distribution of income among the population at a county level based on the frequency distribution of income among the residents of urban and rural populations derived from a national survey⁷² in 2015. We considered that the annual growth rate of population is 2% for Gansu, Inner Mongolia, Ningxia, Qinghai, Xinjiang and Xizang, which is higher than other provinces (1%) resulting from more new jobs and higher income created in these less-developed provinces⁶⁶. Second, we compiled the per-capita disposable income for urban and rural populations at the county level⁶⁶ for 2015–2019. We made a linear projection of per-capita disposable income to 2060 based on the rate of growth of income by province during 2021–2060. We calibrated the growth rate of per-capita income for each income group at the county level during 2015–2060 to guarantee that the projected per-capita income as an average for each county in 2060 was equal to the projection for 2060. Given a carbon price (ς), we assumed that only power plants with MACs below this carbon price were constructed. We estimated the revenue (R_ϵ) from power generation when building a new PV or wind power plant (ϵ):

in which ϱ is the price of coal, oil or gas in China that is substituted by PV or wind power, F_ϵ is total abatement of CO₂ emissions, E_ϵ is total PV and wind power generation and LCOE_ϵ is the LCOE for the projected PV and wind power plants after building plant ϵ.

Revenue from PV and wind power generation could be derived from the electricity price minus the LCOE, but the electricity price can be influenced by many socio-political factors^25,50. We focused on analysing the impact of carbon price as a proxy for climate policies on the revenue of PV and wind power, so we considered that the electricity price depends on the fossil fuel prices and carbon price. The prices of fossil fuel may increase in the future owing to the scarcity of fossil fuel⁷³, which can increase the revenue of replacing fossil fuel with renewables, including PV and wind power. To estimate the prices of fossil fuel in 2060, we randomly draw the prices from the normal distributions, the average and standard deviations of which are estimated from the prices that are available from 2010 to 2020 (Extended Data Fig. 6). Predicting the impact of energy scarcity on the prices of fossil fuel and thus the revenue of PV and wind power is not considered in this study.

We represented the income inequality by dividing the population into 2,002 groups based on the order of income in each county. We excluded pixels in urban areas for the construction of utility-scale PV or wind power plants, so we allocated the revenue among the rural population in each county. When a carbon tax was levied on fossil fuels, we predicted the change in the per-capita income for each group in the population by considering the increased costs of power generation, the revenue from PV and wind power generation and the costs of carbon tax saved by reducing the use of fossil fuel (Supplementary Method 9). Finally, we estimated the income Gini coefficient in China using a formula⁷⁴ based on the changes in the fractions of income and population in each population group for 2,373 counties in 2060 when the carbon price increases from $0 to $100 per tCO₂.

Uncertainty analyses

We estimated the uncertainties in the MAC and the Gini coefficient by running an ensemble of Monte Carlo simulations 40,000 times⁷⁵. We randomly varied the parameters in these simulations, including: (1) the variability of PV power generation (±5%) over a suitable pixel (W_ijy) owing to the impact of aerosol deposition on PV panels⁷⁶ and the variability of wind power generation (±2%) over a suitable pixel (W_ijy) owing to the impact of climate change on wind resources⁷⁷, (2) the growth rate of power demand by province during 2020–2060 (±1%)⁷, (3) the parameters used in the calculation of initial investment costs based on the variability of capital costs (±10%) from previous estimates^55,56, (4) the historical rates of learning for different cost components measured in China (Supplementary Table 1) and (5) the parameters used for calculating the costs of UHV transmission and energy storage from different studies (Supplementary Table 8). Last, we adopted the medians of the MAC and the Gini coefficient to represent our best estimates, whereas we used the 90% uncertainties and interquartile ranges to represent their uncertainties.