Impacts of large-scale Saharan solar farms on the global terrestrial carbon cycle

A fully coupled EC-Earth ESM version 3.3.1 was used to investigate the impacts of large-scale solar farms in the Sahara on the climate and carbon cycle. The ESM is a complex model. In addition to the classical climate model, consisting of physical models of the atmosphere, sea-ice, ocean, and terrestrial ecosystem (Doscher et al 2021), in this set-up a second-generation dynamic vegetation ecosystem scheme was used—the LPJ-GUESS model (Smith et al 2001, 2014). This has state-of-the-art vegetation dynamics, land use functionality and terrestrial biogeochemistry, including carbon and nitrogen (Doscher et al 2021). The combination of vegetation dynamics and soil carbon and nitrogen processes and associated constraints on plant production and ecosystem carbon balance make it appropriate to study the impact on the terrestrial carbon cycle. The LPJ-GUESS model is based on individual and patched representation of land ecosystem structure and dynamics, allowing for more accurate modelling of dry land and mixed forest ecosystems (Smith et al 2014, Whitley et al 2017). The sensitivity of the EC-Earth3-veg model to a warming climate is within the multi-model CMIP6 ensemble spread (Wyser et al 2020).

Four simulations were run using the EC-Earth model 3.3.1. These are CTRL, S05, S20 and S50 and all used active atmosphere, ocean, sea-ice, and dynamic vegetation components, had a horizontal resolution of T159 (about 1^∘) for atmosphere/land/vegetation and had 62 vertical levels in the atmosphere. Initial conditions are present day as constrained by observation and each simulation was started from this state in 1990, the default set-up for modern climate in EC-Earth 3. Greenhouse gas levels, aerosol forcing, and other land use and land-cover properties are set as their 1990 levels. A quasi-equilibrium climate—determined by stable global mean surface air temperature and SST, was reached after 150 year spin up period and the following 60 years model output is used for analysis.

S05, S20 and S50 are scenarios where 5%, 20% and 50% respectively of all the grid points in the North African region between 15–30^∘N, 20^∘ W–45^∘E have prescribed reduced bare soil albedo. This represents the instalment of solar farms which decrease the relatively high surface albedo found on highly reflective desert soils. See supplementary figures 1((a)–(c)) for the placement of these grid points with altered albedo in the simulations. The prescribed albedo of 0.235 approximates the effective albedo of photovoltaic (PV) solar panels (Li et al 2018). The effective albedo is the percentage of incoming shortwave radiation not absorbed by the land surface covered by PV panels so they are not heated in effect. It includes both the shortwave radiation directly reflected back to the atmosphere, plus the energy absorbed by PV panels but exported away from the region in the form of electrical energy thereby stopping it from heating local land surface (Lu et al 2021). This technique of lowering surface albedo to investigate the impact of large scale solar farms has been used previously in modelling studies by Hu et al (2016) and Li et al (2018). In the simulations, for example in S20, gridpoints are picked for one in every five grids over the irregular land grid (supplementary figure 1) in an idealised spatial pattern. This method creates a simplified version of a solar farm, allowing for investigation into the effect of albedo change in the Sahara but of course does not encompass all the changes that will occur with a solar farm installation. S05 is a regional consumption scenario, supplying approximately 24% of world's power consumption (Lu et al 2021). However, as illustrated by Lu et al (2021), climate and vegetation cover responses in S05 are very limited with no systematic shift in the climate region or the terrestrial ecosystem. This indicates the 5% coverage is below a threshold to induce significant impacts and therefore we expect minimal impacts on the carbon cycle. We instead focus more on the plausible but high end S20 and the purely academic S50 scenario. Ideally S20 could yield 86.3—91.2 TW of electrical power averaged over a typical year. This is equal to 10¹²J s¹. This theoretical electricity yield is an estimation based on calculations as explained in Lu et al (2021). Although the ideal power generated by S20 is still larger than global demand, it is a more suitable simulation to study the feedbacks and impacts. We state S50 as a purely academic exercise to better understand robust forcing mechanisms and local and global feedbacks. Solar farm scenarios are compared with a 1990 simulation—CTRL. A full description of the simulations are found in Lu et al (2021). When analysing the 60 year model output, the grid cells containing prescribed albedo were masked out for carbon fluxes NPP, and NBP and vegetation cover variables tree percentage and grass percentage. This means the grid cells containing the hypothetical solar panels are not included in the carbon flux or vegetation results and do not affect the result. This aims to better represent the presence of a solar farm. All the simulated response emerges from 60 year climate/ecosystem conditions which pass the significance test can be regarded as robust and distinguished from the model internal variability.

To further elucidate the forcing effects of temperature and precipitation on the terrestrial carbon fluxes, we conduct sensitivity experiments with a configuration of offline vegetation component of EC-Earth-LPJ-GUESS (Smith et al 2014). The climate forcing variables of temperature, precipitation and radiation for driving LPJ-GUESS were taken from EC-Earth simulations CTRL and S50. We select S50 rather than S20 for stronger forcing signals. Of the four sensitivity experiments, two benchmarks used full climate forcing of CTRL and S50, while the other two largely follow the setup of CTRL but replace temperature (S50-temp) and precipitation (S50-prec) by that of S50, respectively (supplementary table 1). In this way the terrestrial carbon flux response to temperature and precipitation anomalies induced by solar farms in S50 can be isolated.

Each offline LPJ-GUESS simulation was initialized with a 500 year spin-up period from bare-ground conditions as the model reaches an equilibrium state where the global vegetation carbon pool becomes stable. The simulations were continued in the equilibrium state under the prescribed climate forcing and the output was used for analysis. These simulations were run at the same spatial resolution (about 1^∘) as the fully coupled EC-Earth simulations. The atmospheric CO2 concentration, other land-use land change were set the same as in EC-Earth modern condition.

The responses of net biome production (NBP) is investigated for S05, S20 and S50 (figure 1). On an annual basis the most prominent NBP changes occur over equatorial Africa, with over 69% increase in NBP for S50. This carbon uptake is compensated by carbon release elsewhere, spread out globally. This results in global NBP anomalies ranging −0.0015 (S05) to 0.0004 (S50) kgC m⁻² y⁻¹. S20 anomalies for Northern Africa do not exceed 0.0058 kgC m⁻² y⁻¹ whilst in the Amazon S20 anomalies are less than −0.0029 kgC m⁻² y⁻¹ (figure 2(a)), indicating a quasi-equilibrium state.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Change is more evident at the seasonal scale for NBP (figures 1(d)–(i)), particularly in the S20 and S50 scenarios. In summer NBP response is strong and positive in equatorial Africa and Sahel region (S20 increasing 25%, S50 35%), but is negative in East Amazon (S20 decreasing 3.8%,S50 4%). In winter these changes are roughly opposite and cancel out summer NBP anomalies. Seasonal NBP response is less obvious in S05 (figure 1), implying a threshold for the extent of solar farm in inducing significant impacts on regional NBP.

To understand NBP change, we break it down into its component fluxes of net primary production (NPP), soil respiration rate or heterotrophic respiration (RH), and fire carbon flux (fFire). We first focus on the S20 solar farm scenario (see Method for justification). The solar panels in S20 enhance the carbon sink over North Africa, with NPP increasing by 0.0495 kgC m⁻² y⁻¹ (24% rise) in annual average value, particularly around Sahel. Simultaneously there is a weakening of the carbon sink in Amazon with a reduction in annual NPP of 0.0160 kgC m⁻² y⁻¹ (2.3% loss), a magnitude smaller than the North African response (3(a)). On the other hand, RH increases (decreases) in North Africa (Amazon), balancing the carbon uptake change there (figure 3(b)). Both NPP and RH changes are more robust in summer in North Africa, but are more evenly distributed around the year in Amazon (figures 3(d)–(g)). The response of fire carbon fluxes is generally more stochastic with most of changes confined within equatorial Africa, and appear not to be impacted much by solar farms on a global scale. Terrestrial carbon flux in S05 simulations have minimal changes comparatively. A Northern African increase in carbon uptake is barely evident for NPP and RH on supplementary figures 2(a) and (d), whilst losses in the Amazon are weak. Terrestrial carbon flux changes in the S50 simulation (supplementary figures 2(e)–(h)) exhibit similar patterns as in S20, but with slightly over two times magnitude of changes compared to S20. Figures 2(a) and (c) shows the contrast in size of flux anomaly between the simulations. In S50, the NPP response further leads to a spatial transfer of the terrestrial carbon storage, with NPP average annual values increasing by 0.1144 kgC m⁻² y⁻¹ (37.4% rise) in northern Africa and declining 0.0371 kgC m⁻² y⁻¹ (5.6% loss) in the Amazon. The NPP change is largely balanced by local RH change, with RH average annual value rising 0.0952 kgC m⁻² y⁻¹ (32.4% rise) in North Africa but decreasing in Amazon by 0.0364 kgC m⁻² y⁻¹ (7% loss) (supplementary figures 2(e)–(h)).

Zoom In Zoom Out Reset image size

Changes to the global carbon cycle in S20 correspond with changes to the climate. The increase in carbon uptake surrounding the solar panels, coincides with a localised warming of 1.5 ^∘C mean surface air temperature (4(a)), stronger convective motion and thus more intensified monsoonal precipitation of approximately 15% north of 15^∘N (figure 4(b)). The rise in precipitation across Northern Africa causes an expansion and increase in vegetation (figures 4(c) and d)). Tree cover and grass cover increase 21% and 10% respectively between 0 to 30^∘ N. Tree cover is mostly constrained to the wetter, cooler area south of 15^∘N, whilst grass cover expands over Northern Africa. S05 climate variables indicate a mild warming over the Sahara and weak cooling over sub Saharan and vague drying over the Amazon (supplementary figures 3(a) and (b)) but the signal is far weaker than that of S20 and therefore incurs very little vegetation change (supplementary figures 3(c) and (d)). S50 results are consistent with S20, but with stronger changes (supplementary figures 3(e)–(h)). Also evident in the S50 simulation is the northward expansion of the treeline associated with a shift in vegetation productivity. This is notable in Scandinavia and northern Canada (supplementary figure 3(g)), indicating that along with the strengthening of the carbon sink in the Sahel, under the solar farm simulations there may be the emergence of an Arctic carbon sink. We use the S50 scenario to understand these climatic changes. The decrease in albedo, driven by the solar panel presence, has likely reduced the land-ocean thermal contrast (warmer land in Western African monsoon season) and subsequently lead to changes in wind convergence. This is evident on supplementary figure 4 which shows the S50 monsoonal winds convening over the African continent and convergence of moisture there (indicated by blue colour).

Zoom In Zoom Out Reset image size

Sensitivity experiments indicate precipitation change in North Africa plays a prominent role in driving this vegetation expansion (supplementary figures 6–8) and NPP increase. The comparison with DVGM sensitivity experiments on figure 2 show the prominence of precipitation in driving change to northern African NPP and RH, especially compared with temperature forcings. The reduction of the terrestrial carbon absorption from the atmosphere in the Amazon corresponds to a marginally drier climate, and partly to warmer temperatures. Average temperature increases by 0.3 ^∘C and average precipitation declines by 2% (figure 4). The increase in absorption of solar energy in the Sahara (due to the decrease in albedo) has likely caused an energy imbalance between the two hemispheres (Swann et al 2014) and to restore the energy balance, there is a northward shift of the Hadley circulation (Chiang and Friedman 2012), and a consequent northward shift of the ITCZ to transport more energy over the equator (Lu et al 2021). This firstly supplements moisture over Africa but also likely reduces moisture supply to the Amazon basin from the Pacific sector and the Atlantic Sector. This can be see on supplementary figure 4 with monsoonal winds redirecting away from the Amazon basin and either supplementing monsoon precipitation in the Pacific or returning to the Atlantic. This corresponds to an area of stronger moisture divergence focused over the Amazon basin. Over tropical South America, precipitation is further suppressed due to the Gill-type Rossby wave pattern (Gill 1980) induced descending motion (Sun et al 2019). Tree cover declines over the drier and warmer northeast Amazon (10^∘ N–10^∘ S, 65^∘ W–50^∘ W in figure 4), with average annual loss of 2%. Grass cover rises, increasing annual average cover by 8.5%, and expanding over the Amazon area undergoing forest degradation. RH response in North Africa is much stronger in the precipitation sensitivity experiment (supplementary figure 8). Changes in RH are likely a response to environmental changes (Bond-Lamberty et al 2018) and carbon input to soils. In Amazon, the reduced RH is contributed by both cooling and less soil moisture (supplementary figure 8) that slows the decomposition of carbon in the soil.