Future hydrogen economies imply environmental trade-offs and a supply-demand mismatch

Hydrogen will play a key role in decarbonizing economies. Here, we quantify the costs and environmental impacts of possible large-scale hydrogen economies, using four prospective hydrogen demand scenarios for 2050 ranging from 111–614 megatonne H2 year−1. Our findings confirm that renewable (solar photovoltaic and wind) electrolytic hydrogen production generates at least 50–90% fewer greenhouse gas emissions than fossil-fuel-based counterparts without carbon capture and storage. However, electrolytic hydrogen production could still result in considerable environmental burdens, which requires reassessing the concept of green hydrogen. Our global analysis highlights a few salient points: (i) a mismatch between economical hydrogen production and hydrogen demand across continents seems likely; (ii) region-specific limitations are inevitable since possibly more than 60% of large hydrogen production potentials are concentrated in water-scarce regions; and (iii) upscaling electrolytic hydrogen production could be limited by renewable power generation and natural resource potentials.


Supplementary Note 2. Optimization problem -hybrid energy systems
This study considers hybrid energy systems for renewable electricity generation, to be used by the electrolyzer for hydrogen production.4][5] The energy system optimization is performed over one year of system operation (T = 8760 hours) for hundred global locations (see Supplementary Figure 7 and Supplementary Note 3. to explain this selection).
The main objective is to design hybrid low-carbon hydrogen production systems with minimal costs, as shown in supplementary equation (1).The total annual cost [ C/year] includes the annualized investments (C inv ), annualized land costs (C land ), annual operation and maintenance costs (C om ), where replacements are assumed to be covered in O&M expenditures.It is worth noting that the fuel costs (i.e., electricity) are entirely provided by renewables.Further, land costs are included to account for different space requirements of onshore wind and solar photovoltaic (PV), hence technologies that require more land area are penalized, which is reasonable as these are subsequently used in the geospatial analysis.Land area costs are estimated at around 5,000 euro/ha land (c land , lower end of Refs. 6,7), on the premise that land can be used for other purposes than renewable energy generation only.
minimize C an = C inv + C om + C land . ( Supplementary equations ( 2)- (4) show the underlying cost components as shown in supplementary equation (1). 3,4inv = Next, we formulate constraints used in the optimization problem.

Energy balance -annual H 2 production
Supplementary equation ( 5) is introduced to ensure sufficient annual hydrogen production as output from the electrolyzer (here, p t is the hydrogen energy output from the electrolyzer).The set of time steps is described as T = {1, 2, . . ., T }.
where D is the daily hydrogen production [kWh/day], ∆t denotes the time step, and '365' refers to the annual number of days [day].The daily hydrogen production is assumed to be 10 tonnes, in line with Ref. 3

Electrolyzer
Hydrogen (p t ) is produced by polymer electrolyte membrane (PEM) electrolyzers, using renewable electricity (f t ).Here, we assume a constant efficiency, which is in line with earlier findings as shown in supplementary equations (6-7 ): 8

Renewable energy and curtailment
Solar PV, onshore wind, and offshore wind power generation profiles are pre-determined using different Python packages [9][10][11] (generally denoted by P t ).Their electricity generation (p t ) profile is expressed as in supplementary equation ( 8): 3,4 p t ≤ P t s, ∀t.
The inequality constraint ensures the possibility of curtailing renewable electricity generation.

Battery electricity storage
Battery electricity storage is modeled with a similar approach as Refs 3,4,12 .Please refer to this literature for more explanation.Supplementary equation ( 9) describes the battery dynamics in terms of energy stored (e t ), considering the battery charging (f t ), discharging (p t ), and a self-discharging factor (Λ).
Supplementary equations ( 10)-( 11) are introduced to prevent simultaneous (dis)charging of the battery (introducing binary variable x t ), and to ensure that the charging and discharging power are within the boundaries of the battery, where a energy charging duration (τ ) of 2 hours has been used to size the power capacity of the battery.The terms on the right side of supplementary equations ( 10) and ( 11) are bi-linear and are linearized with similar linear terms as described in Refs 4,13 .
Supplementary equation ( 12) is used to ensure the minimum (δ) and maximum state of charge (δ) of the battery.Supplementary equation ( 13) is a periodicity constraint. 13 δs ≤ e t ≤ δs, ∀t, Next, the optimization is performed for hundred global locations to determine trends with regard to renewable energy capacity and battery capacity installed.These non-linear trend lines are given in the next section.

Supplementary Note 3. Curve-fitting as optimization result for hybrid systems
A curve-fitting approach is used to estimate the optimal design of large-scale hydrogen production facilities.
More specifically, non-linear curve fittings are used to quantify the following location-specific aspects: (i) the land share available to install onshore wind and solar PV, (ii) the battery energy storage capacity, and (iii) the electrolyzer capacity.Indeed, different methods exist to address the complexity of optimally designing hydrogen production systems using global scope.
Non-linear curve fittings are a reasonable approach due to their ability to capture non-linear correlations and offer interpretable coefficients, which allow researchers to reuse them.In this way, non-linear curve fittings provide an appropriate balance between computational complexity and the need to determine the (near-)optimal design at each geospatial grid cell (0.25°×0.25°).On the one hand, non-linear curve fittings are typically limited to capturing data outliers, which poses challenges in representing the actual complexity of the system.On the other hand, our analysis does not focus on designing case studies but looks at the global perspective, which makes outliers of individual system designs less important.Clustering might represent an alternative; for example, locations could be clustered based on weather conditions, renewable energy potential, intermittency, and land use.However, determining the optimal number of clusters is complex and likely requires many clusters to capture the set of location-specific aspects globally, potentially leading to an oversimplification of variability within clusters by assuming homogeneity.
The locations for the curve-fitting are specified using a global grid search with 0.5 degree longitudinal and latitudinal steps.For each wind vs. solar PV electricity ratio (i.e., the values on the x-axis of figures 4-6), twenty case studies were selected across the following segments: 0-1, 1-2, 2-3, 3-4, and 4-5.In this search, only one national case study is (max.)included in each segment (except for ratio 4-5 since there are limited case studies for this segment) to ensure a diverse selection, resulting in twenty different countries per segment.Thus, this results in hundred optimizations per scenario; a total of 400 optimizations for the four scenarios considered.This ratio, as well as the solar PV and onshore wind generation profiles, are calculated using the typical meteorological years function of PVGIS 14 .Python package pvlib has been applied to generate PV generation profiles, assuming open-rack ground-mounted multi-Si PV installations 9,10 .
In addition, Python package windpowerlib is used to calculate wind profiles assuming the power output of a Vestas turbine (V90/2000) 11 .
Supplementary Figure 4 shows the correlation between the ratio of annual wind energy and solar energy available (on the x-axis) and the amount of area available for installing onshore wind turbines (on the y-axis) as result of the optimization.The former ratio is determined by dividing the local amount of wind energy available by ground-mounted solar PV energy available.It is worth noting that optimizations are repeated for each future scenario due to different techno-economic assumptions.Supplementary Figure 5 shows the correlation between land share used for electricity generation from onshore wind (on the x-axis) and the energy storage capacity of the battery versus solar PV installed (on the y-axis).Finally, Supplementary Figure 6 visualizes the electrolyzer capacity ratio (on the y-axis) as a function of the land share used for wind electricity production (on the x-axis).
Supplementary Figure 4 and 5 highlight that a non-linear curve fitting has a slightly to significantly better performance than a linear one, thus, we decide to use the non-linear curve fitting to determine land shares available for onshore wind and solar PV.Indeed, the non-linear curve fitting of the battery capacity determination is significantly weaker than the wind share determination.However, we believe that there is a sufficient strong trend that shows higher battery capacity installed with more solar PV capacity installed and with a higher land share for solar PV electricity production, which aligns with findings described in Ref. 4 .
For locations with limited data for only one renewable energy technology, the electrolyzer is sized in a way that ensures that all renewable energy peak capacity can be used.This could for example happen at offshore locations or without solar PV or onshore wind data. Supplementary

Supplementary Note 4. Review, discussion & limitations
In this work, our main novelty is to give indications of the global economic and environmental implications of large-scale hydrogen production in future socio-economic pathways, a short review of current literature on this topic is given below.After that, we discuss the main limitations and suggest potential improvements for future work with a main focus on results and methodological assumptions.
The main body of the article discussed the current limitations of literature on the potential upscaling towards large-scale hydrogen economies.Here, a couple of recent studies on large-scale hydrogen production are discussed.For example, Terlouw et al. 3 provided a comprehensive environmental and techno-economic assessment of optimized hydrogen production now and in the future for five case studies on geographical islands in Europe.These geographical islands exhibit beneficial hydrogen production potential and are used in an analysis upscaling towards 9.5 EJ (per year) hydrogen production via water electrolysis using the impacts of these case studies.Hence, they excluded a comprehensive geospatial quantification of location-specific aspects, except for the five case studies, thus, not considering other global regions for hydrogen production potential.A recent study by Weidmer et al. 15 determines the environmental sustainability of future hydrogen economies in 2050 of 500 Mt/hydrogen economies by considering different hydrogen production pathways including grey, blue, and green hydrogen.The impacts are compared to planetary boundaries using prospective life cycle assessment (LCA).While the authors mention that location-specific impacts are crucial, case studies for hydrogen production pathways were used to derive to the impacts of prospective large-scale hydrogen production quotas.Finally, Tonelli et al. 16 , incorporated water and land considerations to explore the limitations of a future global hydrogen economy solely focusing on electrolytic hydrogen production using geospatial analysis but considering one socio-economic narrative and excluding the optimal sizing, cost, and other environmental factors.
Thus, prior analyses have been limited by exploring a single socio-economic narrative, while we argue that future hydrogen production scenarios are highly uncertain and depend on socio-economic development pathways.Given the diverse range of possible development scenarios, there is an urgent need for comprehensive assessments that address such uncertainties to determine the potential global environmental implications of future hydrogen production economies.Additionally, there are still environmental burdens and trade-offs that have not been evaluated, such as material utilization and overall trade-offs of hydrogen production pathways.These omissions impede a comprehensive understanding of the potential challenges and opportunities of transitioning to a hydrogen-based economy.Lastly, prior global geospatial analyses have often overlooked the optimal sizing of hybrid hydrogen production systems-considering both regional solar PV and wind potentials-or have done so in a simplified way.Besides, these analyses frequently-in our view unreasonably-assumed the existence of a local power grid in remote regions.Such simplifications typically lead to either under-or over-dimensioning hydrogen production systems and neglecting the con-sideration of more complex (off-grid) hybrid energy systems with flexibility options, such as batteries and renewable electricity curtailment.We argue that these systems should be designed in a (near-)optimal way using optimization techniques to minimize cost, and to better account for curtailment and battery electricity storage 3,4,12,13 .
We quantify a comprehensive set of environmental impact categories, in particular, material utilization and water consumption.Life cycle environmental flows are typically well-defined for greenhouse gas (GHG) emissions.However, some of the material inventories are not well represented in the ecoinvent database.
Therefore, we used critical material requirements from literature for onshore wind, offshore wind, and solar PV.For the other technologies, no comprehensive material demand for the critical materials was found, therefore, we aggregated material demands from the ecoinvent database using the specific material flows with 'in ground'.This represents some inconsistency, hence more efforts should be taken to represent better critical material inventories in the ecoinvent database.
As discussed in the main article, some environmental impact categories assessed are very locationspecific.In fact, water consumption is calculated per location, however, the environmental water flows throughout supply chains are not tracked.For example, the location-specific production and construction of technologies (such as solar PV) typically exhibit water consumption in other global locations than where the solar PV installation is installed.Indeed, it is challenging to track these water flows, however, future efforts should provide more insights into this issue since we show that water scarcity and consumption are critical aspects of a future low-carbon (hydrogen) economy.One solution is to apply regionalized LCA, which has the potential to keep better track of such global water flows 17,18 .
Commodity prices are used in line with the year 2020 (pre-covid).However, the prices of electricity, natural gas, and coal have been subject to significant fluctuations in recent years.Thus, this represents large uncertainties, although the last couple of months show a trend towards pre-covid commodity prices, which supports the prices used for our analysis.Another limitation is the use of a single commodity price for all world regions while commodity prices and inflation are typically very different between world regions, which was not accounted for in our main analysis.However, it is important to highlight that the main analysis is focused on low-carbon hydrogen production, making the prices of fossil fuels less influential.
In this work, no value is given to oxygen as a by-product of water electrolysis as it is typically vented into the atmosphere.In fact, oxygen can be a valuable by-product (e.g., for hospitals) and might have economic value 19 , which could make the business case of hydrogen via water electrolysis more economically attractive.
The geospatial analysis used pre-defined exogenous assumptions for techno-economic and LCA parameters.The inclusion of endogenous learning into the geospatial analysis could be interesting to consider as hydrogen production could be responsible for a substantial share of global final energy production and therefore the additional installation of key components-such as electrolyzers, wind turbines, and solar PV-have significant implications for the costs of individual components.However, the actual application of endogenous learning could be challenging due to additional complexities, such as data requirements and model complexity.
We applied a simplified method, in our geospatial analysis, to assess the potential for land utilization and to determine the optimal sizing of energy technologies, such as solar PV and wind capacity.This involved acquiring various land use types and applying corresponding land use factors from various literature sources.
Additionally, the ratio of solar PV to wind availability decides on the sizing of the electrolyzer and any required battery systems, which is based on the energy system optimization of a set of hundred case studies for four future scenarios.Optimally designing a hydrogen production site for each individual grid pixel individually would be the most accurate.However, such a method would be computationally intensive, involving calculations for approximately one million grid pixels.Therefore, we used a curve-fitting technique to estimate the land allocation between wind and solar PV, the capacity of the electrolyzer, and the capacity of battery electricity storage.Satisfactory R 2 values are obtained from the curve fits, ranging between 0.6-1.
To increase accuracy, we recommend applying energy system optimization when there are fewer case studies.
Furthermore, refining these curve fits might be achieved by incorporating a broader set of case studies and data as well as considering different correlations.Importantly, we did not directly account for social aspects in our assessment of potential land use, although some societal considerations are embedded into our land use factors.In fact, social acceptance might represent another significant barrier to the large-scale expansion of renewable electricity.
Single impact factors for other hydrogen production technologies are used to complement hydrogen production mixes based on electrolytic hydrogen production; for biomass-based hydrogen production, coal gasification, and steam methane reforming.However, hydrogen production has location-specific impacts, especially for biomass-based hydrogen production, mainly depending on the biomass source.In our study, the water consumption of Coal gasification premise (v2.0.0), water adjusted using Ref. 31 hydrogen production, gaseous, 30 bar, from hard coal gasification and reforming, at coal gasification plant RER

Figure 4 :
Curve-fitting from optimization to determine onshore wind and solar PV land share occupied in a geospatial grid cell for hybrid energy systems, which adds up to '1' (100%) in total, for four scenarios: a reference, b business-as-usual, c 2 • C, and d 1.5 • C. BAU = business-as-usual, Ref. = reference scenario.

16 Supplementary Figure 9 :
Complete set of life cycle environmental burdens from green hydrogen production (first row, green colors) routes compared fossil-fuel-based routes with (blue hydrogen, blue colors) and without CCS (second row, grey colors).Hydrogen production via: a. PEM water electrolysis using groundmounted solar PV electricity.b.PEM water electrolysis using residential solar PV electricity.c.PEM water electrolysis using offshore wind electricity.d.PEM water electrolysis using onshore wind electricity.e. Autothermal reforming with CCS.f.Autothermal reforming.g.Steam methane reforming with CCS.h.Steam methane reforming.LU = land use (quality).AC = acidification.CC = climate change.ETF = ecotoxicity: freshwater.ETFI = ecotoxicity: freshwater, inorganics.ET FO = ecotoxicity: freshwater, organics.ER = energy resources: non-renewable.EFF = eutrophication: freshwater.EFM = eutrophication: marine.EFT = eutrophication: terrestrial.HT C = human toxicity: carcinogenic.HT CI = human toxicity: carcinogenic, inorganics.HT CO = human toxicity: carcinogenic, organics.HT NC = human toxicity: non-carcinogenic.HT NCO = human toxicity: non-carcinogenic, organics.HT NCI = human toxicity: non-carcinogenic, inorganics.IR = ionising radiation: human health.MM = material resources: metals/minerals.OD = ozone depletion.PM = particulate matter formation.PF = photochemical oxidant formation: human health.WU = water use.Cf = capacity factor.CCS = carbon capture and storage.SMR = steam methane reforming.ATR = autothermal reforming.PV = photovoltaic.Supplementary Figure 28: Contribution analysis on ozone depletion.It is worth noting that ozone layer depletion impacts for the production of reverse osmosis membranes are tremendously high due to CFC-113 emissions, however, these impacts are likely much lower today due to the adoption of regulations such as the Montreal Protocol 46 .Cf = capacity factor.CCS = carbon capture and storage.PEM = polymer electrolyte membrane.PV = photovoltaic.

Table 1 :
20kilogram of hydrogen production via biomass gasification with carbon capture and storage is around 40-45 kilogram H 2 O per kilogram H 2 .However, some recent studies report (much) higher water and land footprint of biomass gasification, up to more than 3400 kilogram H 2 O per kilogram H 2 in Ref20.To illustrate this, applying a higher specific water consumption results in a substantial increase of global water consumption in the 1.5 • C scenario up to 413 bcm for biomass gasification with carbon capture and storage only.Life cycle inventories used for activities of hydrogen production.GLO = global.PV = photovoltaic.RER = Europe.RoW = rest of the world.CH = Switzerland.PEM = polymer electrolyte membrane.NMC = nickel manganese cobalt.CCS = carbon capture and storage.MDEA.= methyldiethanolamine.

Table 2 :
hydrogen production, gaseous, 30 bar, from hard coal gasification and reforming, with CCS, at coal gasification plant RER Techno-economic parameters used in calculating hydrogen production costs and impacts.These techno-economic parameters are selected from a wide body of literature sources.The cost figures refer to the reference and future scenarios for 2050.CCS = carbon capture and storage.PV = photovoltaic.SoC = state of charge.O&M = operation & maintenance.CAPEX = capital expenditures.Eff.= efficiency.Disch.= discharge.