Mapping of the Greenhouse Gas Emission Potential for the Offshore Wind Power Sector in Guangdong, China

Abstract

This study aims to assess the potential greenhouse gas (GHG) emissions of delivering 1 kWh from planned offshore wind farm sites to the grid in the Guangdong Province, China. In contrast to most previous studies, we avoided underestimating GHG emissions per kWh by approximately 49% by adopting a spatialized life-cycle inventory (LCI)-improved stock-driven model under the medium scenario combination. We also developed a callable spatialized LCI to model the spatial differences in the GHG emissions per kWh by cells in planned offshore wind farm sites in Guangdong. The modeling results indicate that, under the medium scenario combination, the GHG emissions per kWh will range from 4.6 to 19 gCO_2eq/kWh and the cells with higher emissions are concentrated in the deep-water wind farms in the eastern ocean of the Guangdong Province. According to the mechanism by which the different scenarios affect the modeling results, increasing the unit capacity of turbines is the most effective approach for reducing the GHG emissions per kWh and decreasing the impact of natural conditions. Air density can be used as an empirical spatial variable to predict the GHG emission potential of planned wind farm sites in Guangdong. The modeling framework in this study provides a more reliable quantitative tool for decision-makers in the offshore wind sector that can be used directly for any offshore wind system with a monopile foundation and be extended to wind power systems with other foundation types, or even to the entire renewable energy and other infrastructure systems after certain modifications.

1. Introduction

The last decade has seen rapid growth in wind power capacity worldwide, but this is only a small part of a global 6000 $\mathrm{GW}$ wind development plan by 2050 [1] that one-sixth of will be built in China [2]. The main reason for the rapid growth of wind power is that the greenhouse gas (GHG) emissions during the life cycle of a wind power system (WPS) are as low as one-hundredth of those of fossil fuels (approximately 7–11, 990, and 530 gCO_2eq/kWh for wind, coal, and natural gas power, respectively [3]). However, the resource extraction, raw material production, manufacturing, transportation, maintenance, and waste management of material cycles of WPSs will consume energy and emit GHGs to the atmosphere [4]. With the rapid proliferation of wind turbines worldwide, the climate change impacts of GHG emissions from WPS [3,4,5,6,7] are currently a critical issue and will continue to be important in the pathway modeling of energy transformation.

The GHG emissions per $\mathrm{kWh}$ have been regarded as typical indicator identification hotspots of the climate change impacts of WPS under the life-cycle assessment (LCA) framework [3,5,6,8,9]. Many studies have been reported on such topics with higher data-intensity (higher technological [3] and spatiotemporal [10] resolutions). For example, Padey et al. [11] reported the GHG emissions per $\mathrm{kWh}$ of 17 installed turbines, which ranged from 3 to 77 gCO_2eq/kWh due to the wind speed and technological parameters of the turbines. Reimers et al. [12] reported that GHG emissions are a function of site conditions and technological parameters and their results show that the energy yield, distance to shore, and water depth had significant effects. Bonou et al. [3] provided an understanding of the GHG emissions per $\mathrm{kWh}$ of turbine technologies. Tsai et al. [13] compared 20 offshore wind farm sites and reported that the cumulative environmental burden from offshore wind farms is most significantly affected by spatial factors (water depth, distance from shore, and distance to the power grid). Dammeier et al. [10] suggested that the location-specific GHG payback times varied between 1.8 and 22.5 months for 4161 wind turbine locations in Northwestern Europe. Poujol et al. [14] developed a geo-located life-cycle inventory (LCI) and demonstrated that geo-located modeling is an effective technique to account for the spatial variability of the contribution of offshore wind energy to GHG emissions.

However, there is still room for improvement based on previous studies. First of all, the product-level GHG emission assessment LCA in previous studies under the LCA framework (cycle of core WPS components throughout their lifespan) [3] is typically a steady-state, rather than a dynamic, aspect and it will lead to ignorance of further technological trends [15]. Secondly, product-level LCA will lead to underestimates for long-term and large-scale wind power development plans, as they did not consider the metabolism (replacement of failed turbines) revealed by dynamic material flow analysis (DMFA) [16] at the system level [17]. For example, if the average lifetime of turbines is 18 years (in accordance with the Weibull distribution, scale parameter = 19, shape parameter = 4.1) [16], 67$\%$ of installed turbines (rates vary with the lifetime distribution of turbines and wind farm operation duration) will need to be replaced due to failure during the 20-year operating cycle [18]. Such underestimation greatly reduces the objectivity, accuracy, and reference significance of the model results. Additionally, wind power is more influenced by spatial factors such as wind resource conditions and land use than coal power and gas power, and high-resolution spatial analysis [10] would significantly improve the modeling performance. Although previous studies have reported spatialized GHG emissions for installed turbines [10,11], they had no callable spatialized toolkit for a long-term and large-scale wind power development plan. The long-term planned installed capacity of the offshore wind power sector in Guangdong is 66.85 $\mathrm{GW}$ [19] and, for almost a third of China, the total planned area for wind power is over 10,000 ${\mathrm{km}}^2$, with water depths ranging from 0 to 76 $\mathrm{m}$, distances to shore ranging from 170 to 380 km [20], and wind speeds at 119 m (baseline in this study) ranging from 7.1 to 9.0 m/s [21]. To better predict the potential contribution of offshore WPSs in Guangdong to the 2060 “carbon neutral” target of China [22], system metabolism, spatial variability, and callability must be comprehensively considered to provide a more thorough and accurate assessment methodology for strategic decision-makers from a hypothetical perspective.

In this study, we adopted a spatialized LCI [23] with a spatial resolution of 1 km × 1 km for offshore WPSs and combined it with a stock-driven model [16] to identify physical pathways for mitigating GHG emissions [24] from offshore WPSs. Additionally, the material, energy, and GHG emission (${\mathrm{CO}}_2$ equivalent) flows were quantified based on a monopile supported turbine model assumption [25,26], the three reference turbines [27,28,29], existing materials [26], energy [30], and GHG emission data for all life-cycle stages [3], and five spatial parameter databases (water depth, air density, wind speed, transportation distance, and transmission distance [20,21]). Moreover, similar to previous studies, 1 $\mathrm{kWh}$ of wind farm power generation was used as the functional unit, which was normalized to a common reference for the life-cycle impact assessment results analysis [3,31], and the results were analyzed and mapped using the geographic information system software SuperMap. The three objectives of this work were as follows: (1) provide a spatially differentiated understanding of GHG emissions per $\mathrm{kWh}$ of the offshore wind power sector from the point of view of materials metabolism, (2) reveal the mechanism of how spatial variables and scenarios affect the modeling results, and (3) inform the industry participants and policymakers in the Guangdong Province with development policies on the basis of the results analysis.

2. Materials and Methods

2.1. System Boundary, Assumptions, and Framework

We adopted a multiple-interaction framework using material, energy, and GHG emission flows reported in the literature [24] and considered the system boundary of offshore WPSs as the duration from the raw material to the waste-management stages and with the planned offshore wind farm sites in Guangdong (Figure 1a) as the spatial scope of this study. We made several assumptions for the sake of simplicity. First, we did not consider the extraction, refining, and transportation of natural resources. Second, we assumed that all turbines were constructed in the comprehensive industrial base within Shantou Port [32] and Yangjiang Port [33] on the east and west coasts of the Guangdong Province, respectively, including manufacturing, port, and waste-management functions. Third, we assumed that the transportation distance (Figure 2d) was the linear distance between the port and installation site and the transmission distance (shortest distance calculated based on the digital elevation map (DEM) atlas, as shown in Figure 2e) was the sum of the distance from the installation site to the wind farm center and from the wind farm center to the landing point of the submarine cable.

Based on these assumptions, the modeling process consisted of five steps (Figure 1b): (1) spatialized LCIs [3,23] were created based on the three reference turbines (

Table 1. Engineering parameters of the selected reference turbines (each cell has an area of one square kilometer).

Turbine Model	Unit	$5\mathbf{M}{\mathbf{W}}_{\mathbf{N}\mathbf{R}\mathbf{E}\mathbf{L}}$	$10\mathbf{M}{\mathbf{W}}_{\mathbf{D}\mathbf{T}\mathbf{U}}$	$15\mathbf{M}{\mathbf{W}}_{\mathbf{I}\mathbf{E}\mathbf{A}}$
Rated power	[MW]	5	10	15
Swept area	[m2]	8003	16,031	19,328
Hub height	[m]	90	119	150
Cut-in wind speed	[m/s]	3	4	3
Rated wind speed	[m/s]	11.4	11.4	10.6
Cut-out wind speed	[m/s]	25	25	25
Blade mass	[tons]	19.89	41.70	65.00
Hub mass	[tons]	50.33	105.50	164.45
Rotor mass	[tons]	110.00	230.60	359.45
Nacelle mass	[tons]	240.00	446.00	657.55
Tower top mass, RNA	[tons]	350.00	676.70	1017.00
Tower mass	[tons]	347.46	628.40	860.00
Total capacity	[MW]	66,850	66,850	66,850
Number of cells	[−]	12,735	12,735	12,735
Capacity per cell	[MW/km2]	5.25	5.25	5.25
Spatial density of turbines per cell	[Turbines/km2]	1.05	0.52	0.35

) and the variables for modeling material, energy, and GHG emission flows [24] related to spatial heterogeneity in the inventory were selected and operated; (2) atlases of spatial parameters (water depth, air density, wind speed, transmission distance, and transportation distance) were prepared; (3) three key variables (operating cycle (OC), design lifetime (DLT), and unit capacity (UC)) were selected and the corresponding values for high, medium, and low scenarios were set; (4) material, energy, and GHG emission flows were calculated using the stock-driven model [16] based on the spatialized LCI, technological and spatial parameters, and equations of the above steps; (5) GHG emissions per 1 $\mathrm{kWh}$ were taken as the indicator for climate change impact assessment and the implications for policy- and decision-makers were provided based on the analysis and discussion of the results. Throughout the process, the layers were presented and results were analyzed at a spatial resolution of 1 km × 1 km using SuperMap. All of the maps in this article were created according to the standard maps supervised by National Administration of Surveying and Geographic Information and the base map was not modified. The official inspection NO. is GS (2017) 1267.

2.2. Data Collection and Scenario Setting

We selected three reference turbines as engineering models for calculating the material, energy, and GHG emission flows in this study, including $5{\mathrm{MW}}_{\mathrm{NREL}}$ [27], $10{\mathrm{MW}}_{\mathrm{DTU}}$ [28], and $15{\mathrm{MW}}_{\mathrm{IEA}}$ [29]; the engineering parameters are listed in Table 1. The results modeled using reference turbines would be larger than those of commercial turbines as the latter are optimized. However, the revelation of the technological differences and spatial heterogeneity trends is notable and the selected reference turbines will adapt to the goal of this study very well.

Five spatial variables were used in the simulation and the layers used to prepare these variables can be described as follows: (1) the water depth layer (Figure 2a) was extracted directly from the DEM atlas published by national oceanic and atmospheric administration [20]; (2) the air density and wind speed layers (baseline in this work shown in Figure 2b,c) were directly extracted from the Global Wind Atlas [21]; and (3) the transportation (Figure 2d) and transmission (Figure 2e) distance layers were based on the data of the DEM atlas [20]; SuperMap was then used to calculate the optimal distance.

We used the following three key variables in the scenario analysis: (1) the OC is the duration of the wind farm operation, which was determined by the agreements between the investment contract and the economic efficiency of operation. Owing to the increased failure in the later stages of DLT [34], we used the mainstream lifetime of turbines as reference [35] and set the OCs for the high, medium, and low scenarios to 30, 25, and 20 years, respectively. (2) The DLT is one of the main engineering parameters of the turbines. Most early DLTs of turbines are 20 years [36] and the average current value is 25 years [3], which will be extended to 30 years by 2030 [35]. Therefore, we set the DLTs for the high, medium, and low scenarios to 30, 25, and 20 years, respectively. We adopted a Weibull distribution with a fixed shape parameter of 4.07 and a variable scale parameter determined by the lifetime [16]. (3) The UC was based on the engineering parameters of the selected turbines, as shown in Table 1. The average UC of offshore turbines in Guangdong increased from 3.3 MW [37] in 2016 to approximately 5.5–7 MW [38] in 2020. Moreover, Siemens Gamesa plans to assemble and produce 14 MW [39] turbines by 2024 and Vestas planned a 15 MW [40] offshore turbine next year. According to the UC development trend, we applied the $15{ \mathrm{MW}}_{\mathrm{IEA}}$ [29], $10{ \mathrm{MW}}_{\mathrm{DTU}}$ [28], and $5{ \mathrm{MW}}_{\mathrm{NREL}}$ [27] reference turbines to the high, medium, and low scenarios, respectively.

2.3. Stock-Flow Modeling and Assessment Indicators

Under the framework we adopted, the stocks and flows of material, energy, and GHG emission were modeled based on the collected data and established equations. The energy and GHG emission flows from the materials account for 70–80% of the life cycle of WPS [3]. Therefore, we set the material flows as the basis for the multiple-interaction (

Table 2. Multiple interactions and inventory items for the material, energy, and GHG emission flow calculations.

LC Stage	Inventory Items	Calculation by Spatial Heterogeneity
		Material Flows	Energy Flows	GHG Emission Flows
Raw material	Turbine	Steel, aluminum, copper, polymer, etc. As shown in Table S2, the weight of components calculated by equations in Table S1.	Embodied energy and GHG emissions of materials based on the weight and data in Tables S2 and S3 of the Supplementary Materials.
	Monopile
	Substation
	Submarine cable
Manufacturing			11.2% to the value of embodied in all raw materials [3].
Transport from the manufacturing plant to the site			As shown in Table S7 of the Supplementary Materials.
Installation	Submarine cable laying		23.1% to the value of embodied in raw materials of submarine cable [3].
	Monopile setting		12.1% to the value of embodied in raw materials of monopile [3].
Operation	Service		1.4% to the value of embodied in all raw materials [3].
	Electricity generation and transmission		Calculated by equations in Table S6.	Lack of data; not considered.
Dismantling	Monopile removal	Based on the previous stage and data in Tables S2 and S4 of the Supplementary Materials.	8.6% to the value of embodied in raw materials of monopile [3].
	Submarine cable removal		7.7% to the value of embodied in raw materials of submarine cable [3].
Transport from site to waste plant			As shown in Table S7 Supplementary Materials.
Waste management		Based on the previous stage and data in Tables S2, S4 and S5 of the Supplementary Materials.	Based on the materials and data in Tables S2 and S3 of the Supplementary Materials.

) analysis of the material, energy, and GHG emission flows. We then used a component-by-component and stock-driven model [16], similar to our previous work [26,41], and added substations and submarine cables as new spatial variables in the material flow simulation (Table S1 of the Supplementary Materials).

The energy and GHG emission flows were modeled based on the material flows and generation and transmission of electricity (Table S6). In particular, the capacity factor, scale, and shape parameters of the Weibull distribution of wind speed at a certain height were determined using the $\mathrm{TREND}$ function of Excel based on the Global Wind Atlas [21]. The spatial heterogeneity of electricity generation was then quantified at a spatial resolution of 1 km × 1 km. The transmission distance of electricity mentioned above was used to calculate the spatial heterogeneity of energy loss through submarine cables. The transportation distance of shipping was also determined to simulate the energy and GHG emission flows (Table 2).

Several emitted gases are considered GHGs, including ${\mathrm{CO}}_2$, ${\mathrm{N}}_2\mathrm{O}$, ${\mathrm{CH}}_4$, ${\mathrm{SF}}_6$, $\mathrm{Hydrofluorocarbons}$, and $\mathrm{Perfluorocarbonss}$ [42], and their carbon dioxide equivalent (${\mathrm{CO}}_{2\mathrm{eq}}$) was adopted as the standard unit for GHG emission measurements. Then, combined with the normalized reference 1 $\mathrm{kWh}$ electricity from WPSs delivered to the grid [31], we used ${\mathrm{gCO}}_{2\mathrm{eq}}/\mathrm{kWh}$ as the functional indicator [3] for the climate change impact assessment of the offshore wind power sector in Guangdong using the following equation:

$$\mathrm{GHG} \mathrm{emissions} \mathrm{per} \mathrm{kWh} \mathrm{electricity}\left(\frac{{\mathrm{gCO}}_{2\mathrm{eq}}}{\mathrm{kWh}}\right)=\frac{\mathrm{CGE}}{\mathrm{CNE}}$$

(1)

where $\mathrm{CGE}$ is the cumulative GHG emissions and $\mathrm{CNE}$ is the cumulative net energy delivered to the grid over the OC of the WPS.

2.4. Uncertainties and Sensitivity Analysis

Some assumptions were adopted in this study and we considered the key parameters based on the assumptions and uncertainties of empirical regression. The empirical equations for the monopile weight, topside and substructure weights of the substation, and energy loss through the submarine are detailed in the Supplementary Materials. Based on the statistical uncertainties, a sensitivity analysis was conducted for each factor individually to assess its effect on the modeling results. We adopted the method proposed for DMFA [43], which is the basis of energy and GHG emission flow analysis, and all numbers in this study were rounded to one or two significant digits. Based on the sensitivity of modeling to three key parameters, we set three scenario combinations for the results analysis and discussion, i.e., high (high OC, DLT, and UC values), medium (medium OC, DLT, and UC values), and low (low OC, DLT, and UC values) scenario combinations.

3. Results and Discussion

3.1. Effect of Parameters and Considering Metabolism

The mechanism by which the scenarios affected the modeling results was determined based on the sensitivity analysis of GHG emissions per $\mathrm{kWh}$ related to UC, DLT, and OC. As shown in Figure 3a, increasing the variables from the high to low scenarios resulted in changes −23% to 22% for UC, −11% to 17% for DLT, and −5% to 9% for OC, respectively. The results suggest that improving the UC of turbines is the most effective technical method for reducing GHG emissions, followed by increasing the length of the DLTs and OCs. The measures for extending the OC could dilute the GHG emissions per $\mathrm{kWh}$ from the raw materials [3] and prolonging the DLT would increase the reliability of turbines to reduce the newly installed capacity for maintaining a specific stock capacity. UC is an integrated technological indicator that considers both technological (hub height, swept area, and generator capacity) and wind resource parameters (air density and wind speed) [27,28,29]. Overall, the increasing UC driven by technological progress suggests a wider swept area of the rotor to support a larger generator capacity, higher hub height to capture faster wind, and a longer DLT driven by improved reliability.

Box plots were used to indicate the dispersion degree of cells in the net energy generation layer, GHG emissions, and GHG emissions per $\mathrm{kWh}$. As shown in Figure 3b, the corresponding interquartile ranges of the GHG emissions per $\mathrm{kWh}$ (high: from 0.59 to 1.4, medium: from 0.54 to 1.4, and low: from 0.54 to 1.4) were between those of the net energy generation (high: from 0.82 to 1.2, medium: from 0.81 to 1.2, and low: from 0.80 to 1.3) and GHG emissions (high: from 0.65 to 1.4, medium: from 0.59 to 1.4, and low: from 0.55 to 1.5) and were closer to the GHG emissions (dominated by the material flow [3]). This suggests that the impact on climate change per $\mathrm{kWh}$ delivered to the grid was attributed to the multiple interactions of energy and material flows. Meanwhile, the dispersion degree decreased from the low to the high scenario combination, indicating that the higher technological level with a larger UC and longer DLT would smooth the differences due to resource conditions, such as wind speed and water depth.

One of the improvements of this model over previous studies [3,5,6,8,9] is that it considered the materials metabolism in WPSs for a long-term and large-scale development target. Figure 3c shows that the cumulative failure rate of turbines in WPSs is 97% under the low DLT and high OC scenario combination, 49% under the medium DLT and medium OC scenario combination, and 12% under the high DLT and low OC scenario combination. This indicates that, if we follow the method in the existing studies, which did not consider metabolism, the results would be severely underestimated for a particular wind farm or area. As wind power plays an increasingly important role in the electricity system, adopting a stock-driven model [16] that includes the metabolism of WPS at the system level would perform better than single LCA [44], which only covers the stages from the cradle to the grave of turbines or other core components at the product level.

3.2. Spatially Differentiated Mapping and Analysis

The spatial heterogeneity was clearly reflected in the results, with a wide range of cell values in the layers. Figure 4a-1–a-3 show the net energy generation by cells under the high (360–640 $\mathrm{GWh}$), medium (230–480 $\mathrm{GWh}$), and low (160–350 $\mathrm{GWh}$) scenario combinations for an example with a partial view of the eastern Guangdong (the full map is shown in Supplementary Materials, Figure S1), respectively. The net energy generation was dominated by the spatial distribution characteristics of wind speed (Figure 2c); the east coast of Guangdong is significantly more productive than the west coast. Figure 4b-1–b-3 show the GHG emissions by cell under the high (1400–6200 $\mathrm{tons}$), medium (1400–7300 $\mathrm{tons}$), and low (1500–8400 $\mathrm{tons}$) scenario combinations, respectively. Unlike net energy generation, the GHG emissions were dominated by the spatial characteristics of water depth (Figure 2a). The deeper water area of the eastern wind farm away from the coast emitted significantly more GHGs than the nearshore shallow water wind farm. In particular, the areas with higher net energy generation and GHG emissions were concentrated in the eastern wind farms (east of GD 2–5, as shown in Figure 1a). The farm site in the deeper water area also received a faster wind speed in the planned wind farm sites in Guangdong.

Figure 4c-1–c-3 show the GHG emissions per $\mathrm{kWh}$ by cell under the high (3.1–12 $\mathrm{g}/\mathrm{kWh}$), medium (4.6–19 $\mathrm{g}/\mathrm{kWh}$), and low (6.8–29 $\mathrm{g}/\mathrm{kWh}$) scenario combinations, respectively. However, unlike the net energy generation and GHG emissions, the distribution of the GHG emissions per $\mathrm{kWh}$ related to any single spatial parameter was not notable. Therefore, we adopted the ordinary least squares ($\mathrm{OLS}$) function in SuperMap (the $\mathrm{OLS}$ results are summarized in Table S16 of the Supplementary Materials) to elucidate the contribution of five spatial parameters (explanatory variables) to the GHG emissions per $\mathrm{kWh}$ (the dependent variable). The results indicate that the air density is the independent variable with the greatest contribution, followed by the wind speed and water depth. However, the $\mathrm{OLS}$ results are only applicable to the database in this study, as the air density and GHG emissions per $\mathrm{kWh}$ results had similar distributions (as shown in Table S17 of the Supplementary Materials). For the general scenario, the wind speed should be the variable with the greatest contribution, followed by the water depth. Meanwhile, the air density can still be used as an empirical spatial indicator to predict the GHG emissions per $\mathrm{kWh}$ for the planned wind farms in Guangdong.

Additionally, different from the data provided in the existing literature, this study is oriented to specific geographical locations. The results of this study are a range, for example the GHG emissions per $\mathrm{kWh}$ by cell ranged from 4.6 to 19 $\mathrm{g}/\mathrm{kWh}$ under the medium scenario and this presents the effects of different geographic locations on the results. However, the results in the literature are a single datum, for example, 11 $\mathrm{g}/\mathrm{kWh}$ for the offshore wind sector in Europe [3], and do not show the impact of spatial heterogeneity on the results. A detailed understanding of geographically specific carbon emissions per kWh of offshore wind power will provide a more scientific reference for China’s future large-scale installations and help China achieve carbon neutrality more successfully.

3.3. Policy Implications

First, from the comparison with methods in existing studies [3,11,12] that did not include metabolism, using a stock-driven model [16] can avoid the underestimation of the GHG emissions per $\mathrm{kWh}$ for offshore WPS. Additionally, the introduction of a spatialized LCI [23] highlights the spatial differences. These improvements have significant benefits for strategy-making, and, most importantly, allow for a more accurate estimate of the contribution of the offshore wind power sector to carbon neutrality in specific locations, which also provides a reference for location-oriented engineering optimization. Policymakers can then use this as a basis to develop market-access policies that encourage the use of more low-carbon turbines and effectively increase the contribution of the offshore wind power sector to carbon neutrality. The improved methodology can also be extended to the assessment of other environmental indicators, the whole renewable energy sector, and other infrastructure sectors.

From the results of the sensitivity analysis, increasing the UC of turbines is the best approach for promoting the national carbon neutral strategy at the technological level, and a larger UC can reduce the impact of natural conditions on the GHG emissions per $\mathrm{kWh}$. In 2020, the first 10-$\mathrm{MW}$ offshore wind turbine [45] in China was successfully installed and connected to the grid in Xinghua Bay, Fujian Province, China. As a local manufacturer in Guangdong, Ming Yang released an 11$\mathrm{MW}$ offshore turbine in 2020 [46] and a 16 MW offshore turbine in 2021 [47]. Prolonging the DLT is another effective approach for reducing the GHG emissions per $\mathrm{kWh}$. Additionally, thus, a larger UC also indicates a longer DLT, which is widely mentioned in various policies [48,49]. A series of technology development support policies have proven to be effective and such policies should be maintained and improved. However, the period of use for sea areas, which is one of the factors that determine the OC, is rarely mentioned by existing regulations. A longer OC can reduce the GHG emissions per $\mathrm{kWh}$ by diluting the emissions flow embedded in the raw materials of the WPS and it should be mentioned in future industrial policies to allow owners to improve the environmental benefits of offshore WPSs.

Finally, the electricity loss rate also affects the GHG emissions per $\mathrm{kWh}$ and ranged from 2.7% to 20% among the cells in the planned wind farm sites of Guangdong, exceeding 9.8% in almost 40.8% of the cells. However, the distance to the grid is difficult to reduce and increasing the transmission voltage from 220 $\mathrm{kV}$ or switching the alternating current into a direct current from the substation to the grid is likely the most effective solution [50]. Such measures should be accompanied by active policy support or even direct fiscal subsidies. These methods have a good basis for implementation in Guangdong, where the voltage class of the main transmission network is 500 $\mathrm{kV}$ and the flexible direct current transmission lines are under consideration by the China Southern Power Grid [51].

4. Conclusions

In this study, we provide a spatially differentiated understanding of GHG emissions per $\mathrm{kWh}$ of the offshore wind power sector from the point of view of materials’ metabolisms. Firstly, we resolved the underestimates (49% under the medium scenario combination) of GHG emissions for WPSs by adopting a stock-driven model. The results are in sharp contrast to those of previous studies that did not consider the metabolism of the system. Furthermore, we used a callable spatialized LCI to improve the modeling performance (difference in GHG emissions per $\mathrm{kWh}$ even exceeded four times between cells, the results shown 4.6–19 $\mathrm{g}/\mathrm{kWh}$ under the medium scenario) over studies that did not consider spatial variables. The modeling framework proposed in this study provides a reliable and accurate methodology for decision-makers throughout the renewable energy sector and can even be extended to other infrastructure sectors.

However, the use of a reference turbine model may lead to an overestimation of material stocks and flows, so further improvements should include data of specific offshore wind farm cases. Meanwhile, this study selected a monopile as the foundation model, which may lead to deviation from the actual results in a deep-water area. Further research will consider the introduction of a semi-submersible foundation and a jacket foundation to optimize the model. Another shortcoming is the lack of a localized offshore wind industry supply chain database; this can lead to an underestimation of regional differences, so collaboration within the industry will be necessary in the future.