Attaining stable electrolyzer operation under multi-annual climatic variations in almost fully renewable electricity systems: the case of Sweden

Hydrogen produced from renewable electricity can play an important role in deep decarboniza-tion of industry, such as primary steel-making. However, adding large electrolyzer capacities to a low-carbon electricity system also increases the need for additional renewable electricity generation which will mostly come from variable renewable energies (VRE). This will require hydrogen production to be variable, unless suﬃcient ﬂexibility is provided by other sources. Existing sources of ﬂexibility in hydro-thermal systems are (a) hydropower and (b) thermal generation. However, increasing the ﬂexibility of hydropower generation may have negative consequences for river ecosystems and the use of fossil and non-fossil fuels in generation may increase if thermal power is increasingly used to balance short-falls in wind power during electrolyzer operation. We assess here for our Swedish case study the utilization of electrolyzers with a dispatch model, assuming that additional VRE generation matches the additional electricity demand of hydrogen production on average. The ﬂexibility of hydropower and thermal generation is restricted in four scenarios, and we run our model for 29 diﬀerent weather years to test the impact of variable weather regimes. We show that (a) in all scenarios, electrolyzer utilization is above 60% on average, (b) the inter-annual variability of hydrogen production is very high if thermal power is not dispatched for electrolysis, (c) this problem is aggravated if hydropower ﬂexibility is also restricted, and therefore (d) either

increasing the flexibility of hydropower generation may have negative consequences for river ecosystems and the use of fossil and non-fossil fuels in generation may increase if thermal power is increasingly used to balance short-falls in wind power during electrolyzer operation. We assess here for our Swedish case study the utilization of electrolyzers with a dispatch model, assuming that additional VRE generation matches the additional electricity demand of hydrogen production on average. The flexibility of hydropower and thermal generation is restricted in four scenarios, and we run our model for 29 different weather years to test the impact of variable weather regimes. We show that (a) in all scenarios, electrolyzer utilization is above 60% on average, (b) the inter-annual variability of hydrogen production is very high if thermal power is not dispatched for electrolysis, (c) this problem is aggravated if hydropower flexibility is also restricted, and therefore (d) either

Introduction
Hydrogen is considered as one important option for deep decarbonization of energy and industrial sectors, in particular in the fields of transportation [1], as feedstock for fuel or chemical production, as reductant in primary steel-making [2], or as energy storage in stationary heat and electricity applications [3][4][5]. Today, natural gas constitutes the primary source for hydrogen production [6].
However, to allow for an actual contribution to decarbonization, hydrogen has to be generated in a carbon-free way. The electrolysis of water, using carbon-free renewable electricity, is one potential technological pathway.
Significant additional amounts of renewable electricity generation have to come, however, from intermittent generation such as solar PV or wind power, i.e., variable renewable generation (VRE), at least in markets where hydropower potential is already used up to its full potential. In effect, variability of generation is set to increase in the absence of additional measures. Operating electrolyzers in a variable way to align with renewables' variability is possible in principle. Yet, due to their high investment cost, the economics of electrolyzers demand high utilisation rates, and 'peak shaving', i.e., using electrolyzers to produce hydrogen from limited peaks in intermittent generation, is therefore economically not competitive [7]. Also, the quantities of hydrogen produced would be relatively small if only otherwise curtailed electricity is used.
Attaining a high utilization rate of electrolyzers under high VRE penetration therefore may increase the need for other flexibility options in the system. Hydro-thermal systems, such as our Swedish case study system, offer already significant flexibility today as both hydropower with storage and thermal generation can adapt their output according to system demands.
Hydropower dominated systems can offer flexibility on all time scales, [8][9][10][11] from hours to seasons to years. The highly flexible operation of hydropower plants for system integration of VRE, however, causes increased rapid sub-daily fluctuations in water flow and water levels (hydropeaking).
This conflicts with other environmental quality objectives as short-term river regime alteration poses a key threat to river ecosystems [12][13][14][15]. The negative impacts of a higher penetration of renewables on hydropeaking have indeed been assessed before [16].
Using thermal power generation more flexibly has minor consequences in terms of a lower efficiency of thermal power plants in part load operation [17]. However, depending on the system setup, thermal generation may even increase if used to guarantee a high utilization rate for electrolyzers. This comes at the cost of increased usage of fuels and associated air pollution impacts [18] and CO 2 -emissions. The increased use of biomass as an alternative to fossil fuels also has its own environmental risks [19] and is limited by the sustainable sourcing potential. Thermal power generation should therefore be limited.
The role of electrolysis-based hydrogen in energy and electricity systems has been widely investigated in previous work, and several recent reviews show the important role of hydrogen in a future decarbonized energy world [3,4,20]. In particular, electrolyzers have been shown to increase the share of renewables in power systems such as in Europe [21], a sub-region in Norway [22], California [23], and Japan [24]. Hydrogen storage has also been shown to mitigate the problem of hydro-peaking [25] and to lower the spilling of water in hydropower cascades [26]. In most studies, hydrogen is assumed to be operated on surplus renewable electricity [7,[27][28][29] and existing studies have typically focused on the potential use of hydrogen for system integration of intermittent renewables. A limited amount of studies have assessed the production profile of renewable hydrogen production for industrial purposes [30][31][32]. None of the studies, however, assessed how climatic variations impact the long-term variability of hydrogen output and how existing flexibility measures can be used to stabilize hydrogen generation.
We therefore assess here in how far a high utilization rate of electrolyzers can be guaranteed in an almost fully renewable electricity system if additional electricity for electrolysis comes on average only from VRE, in particular wind power. We study the case of Sweden and assess the effect of gradually restricting the flexibility of hydropower and thermal power generation on the utilization of electrolyzers.
We chose Sweden as a case to study, as the country is well positioned to take a lead in the production of low-carbon hydrogen. Sweden has a power system with a very low CO 2 -emission footprint, and strong policies in place to support full decarbonization, including a goal of 100 % renewable electricity production by 2040 [33]. The above-mentioned trade-off between increased hydropower production to meet future needs in the energy system and reduced environmental impact on rivers is evident in the Swedish system [34], and hydropeaking has been observed as both high and increasing in Nordic regulated rivers [13,35]. Hydrogen is also being outlined as a potential key technology in the future Swedish energy system, both as a flexibility technology to balance a high VRE share in the power system, for production of biofuels, and as reductant in the steel industry, where hydrogen is currently considered as the main track for decarbonization of primary steel-making [36][37][38]. While the results of our case study cannot be directly transferred, our conclusions do well apply also to other hydro-thermal systems with hydropower shares larger than 35% such as Brazil, Canada, New Zealand, or Austria [39].
Hydropower dominated electricity systems are prone to large inter-annual variations in water availability [40,41]. To a limited extent, wind power systems also show inter-annual variability [41,42]. A multi-year approach to assessing energy systems with high shares of renewables is therefore necessary [43]. In order to be able to realistically capture inter-annual variations and extreme weather events, we use simulations of time series of VRE and electricity demand in a dispatch model for the Swedish power system for 29 different weather years, at hourly temporal resolution. We can thus assess how both short-term (hours to days) and long-term (days to months to years) variability of climatic variables drive power production patterns. The model was developed by Höltinger et al. [41] and is extended to allow simulation of electrolyzer technology and a more detailed representation of thermal power generation and hydropower operational restrictions.

Material and methods
We assess here how different scenarios of thermal and hydropower flexibility affect the utilization of electrolyzers using a generation dispatch model for our case study of Sweden. In the following section, we present the general model structure and most relevant parts, as well as the major changes compared to the work by Höltinger et al. [41], with a more comprehensive and detailed model description given in Appendix A.1. The optimization program (written in GAMS and controlled by Python), the data necessary to run the simulations, the result of the simulations, and the R Code for result analysis can be found on Zenodo [44].

Model description
The optimization model is based on hourly data for natural river runoff, load, and wind generation and aims to minimize total variable system cost less the revenue from hydrogen production.
Residual demand, i.e., the mismatch of wind power generation and load, has to be balanced by thermal and hydropower plants, and curtailment of wind power. Potential further balancing needs are provided from further backup measures. These backup measures were assumed to be available at very low investment but high variable costs, as they are used with low frequency. Potential candidate technologies are additional thermal peaking plants, demand side management measures, or imports from neighboring countries. These are, however, not modelled in detail.
To be able to account for climate variability, the model was run for 29 different weather years, which were used to simulate temperature dependent load, hydropower, and wind power generation in the model. The model optimizes a single year of dispatch, i.e. inter-annual water storage was not considered.
A temperature dependent load profile of electricity demand was derived from a regression model based on reanalysis temperature data for 29 years and gridded population raster data (for details, see Höltinger et al. [41]). The modelled annual load is on average equal to the average annual observed load in the period 2010 -2018 (approximately 130 TWh a −1 , excluding transmission losses). In addition, we have considered increased power demand from electrolysis (see section 2.2.1), but not from other uses such as increased demand by industry, data-centers, or from electrification of transportation. Table 1 summarizes the characteristics of the included power plant types and other model components.

Wind power
For wind power, we used time series for potential future wind power production in Sweden, as modelled by Olauson et al. [45,46]. The authors generated a range of different production scenarios, based on bias corrected wind speed data for Sweden. The assumed expansion of wind power is based on the assumption that nuclear power production is fully phased out and replaced by wind power, i.e. 60-65 TWh a −1 from 8.6 GW of installed capacity [47]. We also assumed a reduction of power exports to zero, which compares to observed net exports in the range of around 7.2 and 23 TWh a −1   during the years 2011-2018 [47]. Further, we accounted for minimum thermal production from CHP plants (see section 2.1.4). This gives a total average annual wind power production of 49 TWh a −1 from 14 GW of installed capacity, in the base scenario without hydrogen production. This compares to the current annual production of 17 TWh a −1 from 7.4 GW of installed capacity (in 2018) [47].
The higher utilization of additional wind power can be partly attributed the addition of offshore wind capacities, and partly to larger rotor sizes.
In the modelled electrolysis scenarios, wind generation was scaled to match the increased electricity demand from electrolyzers, as described in section 2.2.1.

Hydropower
We followed the model formulation given in Höltinger et al. [41], which assumed a simplified hydropower model aggregating all hydropower plants to one reservoir and one plant. Simulated time series for river discharge from the hydrological catchment model S-HYPE were used [48,49] and translated into power generation with a simulation model that takes the characteristics of all Swedish hydropower plants into account. A total reservoir capacity of 33.7 TW h was modelled, which needs to be kept between minimum (5 %) and maximum (98 %) observed levels during all hours [41,50].
The reservoir level at the start of each year, as well as the required minimum level at the end of the year, were both set to 62 %, which corresponds to the average reservoir filling level for the time period of 1960 to 2016 [50]. This represents a rather conservative approach which limits the possibility for hydropower to act as inter-annual storage, as extreme weather years with, e.g., low production of both hydro and wind power, cannot be dealt with by running down the levels of hydropower reservoirs at the end of the year. Historically, end-of-year reservoir levels have seen variations between 43 % and 86 % [50].
Hydropower operational limits, i.e., minimum flows and maximum ramping rates, were then assessed in two different scenarios (see section 2.2.3).

Thermal power generation
Höltinger et al. [41] did not differentiate between plant and fuel types; instead, thermal power generation was defined as one plant. We have here improved the original thermal generation model by disaggregating it into different plant technologies and fuel types.
Data on existing thermal power generation was compiled from the World Electric Power Plants database [51], and complemented by national statistics [52,53]. Thermal production was categorised based on fuel, production scale and production technology, with each plant type defined by individual marginal generation cost and capacity (Table 1). Power production costs were based on projections for technology efficiency and fuel costs. Production costs further include a CO 2 -charge (50 e /ton CO2 ), operation and maintenance cost, and heat credits, where applicable. Appendix A.2 provides the details on assumed cost and plant efficiencies.
Additionally, we have defined must-run conditions for different seasons of the year, as many plants serve heat demand from the residential and industry sectors. Generation in CHP plant was assumed to follow a monthly pattern throughout each year, with higher minimum production during the winter months, and lower minimum production during the summer. Additionally, maximum production was restricted during the summer period to account for maintenance and limited operation of CHP plants due to heat load restrictions. Annual production profiles were developed based on statistics of installed capacity and annual production per fuel type, in district heating systems and industrial back-pressure systems, respectively [50]. Expected annual full-load hour equivalents amount to 7500 h a −1 for industrial biomass CHP and waste CHP, and 5000 h a −1 for biomass CHP plants [54]. The load profiles applied in the optimization model are shown in Figure 1.
Maximum combined hourly ramping rates for the sum of thermal power generation were derived from historical observations of maximum ramping rates, and therefore set to 1.5 GW.

Hydrogen production
We have assumed that electrolyzer technology will advance beyond the current state-of-the-art (alkaline electrolysis), to proton exchange membranes (PEM). This has been reflected in our assumptions on hydrogen production efficiency, which was set to 72 % (system efficiency, defined as hydrogen output on LHV basis divided by electrical input to the electrolysis system), corresponding to 46 kW h of electricity per kg hydrogen [55][56][57]. The electrolyzer was sized according to an expected electrolyzer utilization of 90 %. As described in section 2.2.1, the utilization of the electrolyzer is actually a model output, which means that hydrogen demand in the different scenarios will not always be exactly met.
As PEM electrolyzers in general are highly flexible, with start-up times and ramping in the range of minutes or even seconds and very low minimum part-load operation (<10 %), neither ramping nor electrolyzer part-load restrictions were considered in the model [57].

Scenarios
We defined increasing levels of demand for hydrogen (4 scenarios) and tested these demand scenarios under gradually more restricted flexibility in the system. First, we restricted thermal generation for hydrogen production by either allowing or not allowing the dispatch of thermal power generation for hydrogen production. Subsequently, we restricted the hydropower flexibility by limiting ramping and minimum river flows in two scenarios. This results in a total of 12 scenarios with hydrogen production plus two baseline scenarios without hydrogen demand (and therefore no thermal dispatch for hydrogen scenario). Additionally, we ran a sensitivity analysis on the model for a wider set of hydrogen demands (see section 2.2.1).

Hydrogen demand
The hydrogen production scenarios were chosen so that they are able to deliver different amounts of hydrogen in order to satisfy different levels of future projected demand. We limited the hydrogen This results in different electrolysis loads on the system, as outlined in Table 2.  The estimates regarding biofuel production build on the ambitions announced by Preem, Sweden's largest fuel producer, who has a goal of producing 3 million m 3 of biofuels by 2030 in their two refineries in Sweden [58]. Judging from their announced projects and plans, all biofuels will be drop-in fuels produced via hydroprocessing of various bio-crudes. Hydrogen is currently produced from refinery off-gases or via steam reforming of natural gas, but Preem has also expressed strong interest in hydrogen produced via electrolysis [58]. We have based our scenarios on the assumption that electrolysis-based hydrogen will be the main pathway in the future, and implemented two different annual demand levels; 5 and 10 TWh a −1 hydrogen, respectively. The lower level represents a scenario in which a large share of the biofuel feedstock has a relatively low oxygen content (e.g., used cooking oils or bio-crudes produced via hydroypyrolysis or hydrothermal liquefaction), while the higher scenario assumes a large share of biofuel feedstock with a higher oxygen content (e.g., lignin oil or fast pyrolysis oil) [59,60].
In the Large scenario, we also assumed that the HYBRIT route will be fully implemented in Sweden, and that all primary steel-making via the blast furnace-basic oxygen furnace route will thus be replaced with hydrogen based direct reduction followed by electric arc furnaces. We only considered the projected additional electricity demand to cover the required hydrogen production In addition to the scenarios based on actual announced plans by different industrial actors, we applied a sensitivity analysis, where we tested extended capacities of electrolyzers -and associated increases in wind power generation -on a wide range of scenarios (5000 MW to 100,000 MW).

Thermal power flexibility
Additionally to varying the hydrogen demand, we also assessed the impact of dispatching thermal power plant for hydrogen production. Technically, we did so by varying the value of hydrogen in the objective function so that it was either below most (18 e MWh −1 ) or above all (160 e MWh −1 ) marginal costs of thermal generation. Thus, thermal power was either never or always dispatched to produce hydrogen. This represents two extreme settings, for which reason the results cover a wide range of possible outcomes. We refer to these scenarios as No thermal and Thermal, respectively.

Hydropower flexibility
We assessed two different scenarios regarding the impact of seasonal ramping restrictions and seasonal flow thresholds in the hydropower system, that represent two different hydropower reg-  The resulting monthly limits are displayed in figure 2, where a shows the minimum flow, and b the maximum ramping rates. These limits were derived from the simulation of potential natural flows (i.e., without human intervention) from the S-HYPE model [48,49]. We observe the highest monthly median flow in June at about 18,000 MW, and the lowest flow in March, at only 3600 MW, which leads to a high variation throughout the year of both minimum flows and maximum ramping rates. Figure 2 a also shows the historical daily mean flows of all the 29 modelled weather years (blue) and the daily mean of 16 modelled weather years with regulated (human interference) flow (yellow).

Model runs and performance indicators
We ran the optimization model for all scenarios outlined in the previous sections. Each scenario was evaluated for the 29 different weather years, at hourly temporal resolution. Our evaluation focused on the following performance indicators: • Utilisation of electrolyzers (% of 8,760 hours) • Inter-annual variability of hydrogen production (variance in utilisation across years) • Thermal power generation (MWh a −1 ) • Required additional backup capacity (MW) • Required additional backup energy (MWh a −1 ) • Variability in hydro flow (maximum hourly ramping) (MW) • Minimum hydro flow (MW)

Results
The results focus on the utilization of electrolyzers, thermal generation, backup generation and capacity, and the changes in flows and minimum flows as induced by hydropower operation. The results of the base scenario without electrolyzer operation are shown where applicable.

Electrolyzer utilization
We first discuss here the electrolyzer utilization in the different scenarios. Figure 3 shows the annual utilization rate in all 29 simulated years for all scenarios. The average utilization rate is significantly lower for those scenarios which do not allow for the dispatch of thermal generation for hydrogen production. Additionally, lower hydropower flexibility decreases the utilization rate. For the No thermal scenarios with high hydropower flexibility, the average utilization rate is mainly lowered by some extreme years, as can be observed from the difference between the median and the mean of the distribution. The average utilization rate of the electrolyzer increases with higher electrolyzer and wind power capacities in the No thermal scenarios, while it decreases slightly for the Thermal scenarios. The reason is that in the No thermal scenarios, first wind power substitutes thermal power generation compared to the baseline scenario, as this reduces operational cost. However, this substitution effect is limited by the required minimum generation of thermal power production, as implied by heat demand. At higher electrolyzer and wind capacities, this causes an increase of excess wind power to use in hydrogen production. This effect is explored in more detail in section 3.5. The inter-annual variability of hydrogen production is very high for the No thermal scenarios and the electrolyzer utilization drops to even below 25% for single years. This inter-annual variability is mainly driven by the variability in the availability of hydropower, and less so by variability in wind power generation or temperature dependent demand (see figure 4).
This implies that single bad hydropower years have to be accounted for in the long-term planning of hydrogen supply, in case that the dispatch of thermal capacities for hydrogen production should be prevented.

Thermal generation
The dispatch of thermal capacities in all scenarios is shown in figure 5. As could be expected, thermal generation is lower in the No thermal scenarios, with the obvious exception of the No hydrogen capacity scenario. Thermal generation is, correspondingly, higher for the Lo hydro flex scenarios. Limiting hydropower flexibility increases the annual thermal power generation, on average, by 3 TWh in the No scenario. It can be observed that for those scenarios, where thermal generation is not dispatched for electrolyzer operation, the total thermal generation even falls because additional wind power capacity is added to the system. This is, conversely, not the case for the Thermal scenarios, where thermal generation is not replaced by additional wind power.
In the Large scenario, the difference in annual thermal generation increases to on average 8 TWh when comparing the two most extreme scenarios, i.e., the No thermal -Hi hydro flex and the Thermal -Lo hydro flex scenarios. In single years, this difference can amount to up to 20 TWh. Again, this shows that while thermal power production is not essential to achieve sufficient electrolyzer utilization on average, in single years the dispatch of thermal power for electrolyzer operation will allow for significantly higher full load hours.
The share of thermal generation in total generation is between 11% and 17%, depending on the scenario. This, however, is mostly due to must-run conditions of thermal power plants, which have to generate at least 15 TWh of power to provide sufficient heat to heat consumers. Of course, in future systems with lower heat consumption and other heat sources, thermal generation may be reduced further.
Appendix A.3 shows details on how different thermal plant types are dispatched.

Backup generation and capacity
Backup generation and capacity is very low in all scenarios. If one extreme weather year, i.e.
1996, is removed from the data set, annual backup energy utilization is lower than 82 GW h in all remaining years (see figure 7), with a required annual backup capacity lower than 4 GW in all scenarios (see figure 6). Both the backup energy and the capacity fall with increasing electrolyzer capacity and corresponding wind power generation in the system, as electrolyzers are ramped down in hours when backup operation would otherwise be necessary. Electrolyzers, in tandem with the additional VRE generation capacities required to fuel them, can therefore provide crucial value to the system by reducing the amount of backup capacity necessary. In the Large scenario, backup capacity requirements fall to less than 2.5 GW. The difference between the hydropower flexibility scenarios is very low. Lower hydropower flexibility increases backup capacity and energy, but mostly due to one extreme year.
All scenarios run on at least 83% of wind and hydropower and only existing thermal generation capacities excluding nuclear are considered. The resulting backup requirements are low, and in particular the energy provided by those backup capacities is negligible, even if high variable costs are assumed. Providing, in total, on average 0.080 TWh of annual backup in the form of, e.g., demand response in a system that has a total demand of at least 130 TWh seems to be on a realistic scale.

Hydropower ramping and minimum flows
The distribution of hourly river flows and ramps in the system are shown in figure 8  Higher electrolyzer capacities slightly increase the spread in the river flows and in the magnitude of ramping events. Hydropower is therefore operated slightly more flexibly with increasing hydrogen production and added wind power capacity, as increased system variability is partly balanced by hydropower. There is almost no difference between the No thermal and Thermal scenarios if the same hydropower flexibility rules are applied, i.e., using thermal generation for hydrogen production does not affect the extreme conditions of hydropower operation. Interestingly, extreme ramping events are, although allowed, rare. The maximum ramping allowed in the scenario with low flexibility is just above 1 GW. However, this ramping capacity is required at most in 0.3% of all hours in all scenarios, while ramping above 0.75 GW is only necessary in at most 1.9% of all hours. Likewise, a ramping capacity above 2 GW is only required in at most 0.4% of all hours in the scenario with high hydropower flexibility, where around 5 GW of ramping is allowed. This indicates that rapid ramping of hydropower is beneficial to the system in some moments, but is not massively required to balance the system. Figure 9 shows the resulting utilization of electrolyzers when extending the capacities to very high levels, also including a corresponding expansion of wind power capacity.

Sensitivity analysis
The figure shows that an increase of the electrolyzer capacity up to around 8 GW of capacity has a perhaps counter-intuitive consequence for the No thermal scenarios: at the lower end of electrolyzer capacities, increasing capacities increase the electrolyzer utilization for the No thermal scenarios, up to somewhere in the range of 5-10 GW of capacity, depending on the scenario. Above that point, adding more electrolyzer (and wind) capacity, reduces the utilization. The cause is that at lower wind capacities, wind is used to cover residual demand in hours when there is a lot of wind as it is assumed to have zero marginal cost. Less wind is thus available for surplus hydrogen production due to that substitution. Increasing the wind power capacity in the system further will reduce the thermal generation to the defined minimum load, thus making no more substitution of thermal generation possible and instead releasing a higher share of wind power generation for hydrogen production. Therefore, rapidly increasing utilization of electrolyzers can be observed at the lower capacity range in the No thermal scenarios. At some point, negative residual demand will exceed the electrolyzer capacity for some hours. Once the latter effect grows stronger than the former, the utilization starts decreasing again. This effect, however, applies only if no extra thermal power is dispatched for hydrogen production. In the Thermal scenario, where thermal power is dispatched for hydrogen production, the utilization starts falling with increasing electrolyzer capacity immediately.

Discussion
We have assessed multi-annual variability of hydrogen production in almost fully renewable energy systems, considering flexibility from thermal and hydropower generation. While this has, to the best of our knowledge, not been done before, we discuss here some limitations of our analysis. The three hydrogen demand scenarios are marked with vertical dotted lines, the curved lines shows the adjusted electrolyzer capacity to meet the expected yearly production (estimated at 90 % for each scenario).
The interaction between electricity and hydrogen markets was not modeled. In bad weather years, electricity prices will increase, which will also drive up the hydrogen prices. Depending on the relative effect on the two, markets will either favor thermal dispatch for hydrogen production, thus limiting the negative impact of single bad weather years on hydrogen production, or hydrogen production will decrease.
We also assumed that Sweden has no international interconnections and that internally, there are no transmission bottlenecks. The first assumption makes our results conservative, as the existing interconnections clearly could provide backup capacities. At the same time, market prices and dispatch would change significantly, if interconnections would be taken into consideration. Depending on the thermal flexibility scenario, thermal power generation in neighbouring countries would be replaced by Swedish wind power generation up to the interconnector capacity, before generating surplus electricity for hydrogen production within the country. Disregard of international interconnections also implies that Sweden has net zero exports and imports, which can be compared to current annual exports of around 10-20 TWh of electricity. The second assumption, which neglects internal restrictions in transmission, is an obvious simplification. Here, we assume that the transmission grid is reinforced to accommodate additional wind power and prevent frequent large-scale curtailment.
Land and sea availability for placing wind-turbines is another major issue. In the scenario with the largest electrolyzer capacity (3610 MW), around 100 TWh of wind power is generated, requiring an installation of around 30 GW of wind power capacity. This compares to an installed capacity of above 50 GW of wind in Germany, a country with only 80% of the available land area of Sweden. Such an expansion should thus in principle be possible. It may, however, cause regional environmental impacts and land conflicts, and mitigation measures must be taken seriously.
We here only considered increased load from electrolysis -other scenarios show a potential increase from electrification of, e.g., transport and (other parts of) industry where total power demand may increase to over 200 TWh a −1 , including demand from electrolysis [63]. This compares to a load of about 130 TWh a −1 in our scenario without electrolyzer and to around 160 TWh a −1 in our largest electrolysis scenario. At the same moment, we assume that around 60 TWh a −1 of nuclear are phased out. Keeping nuclear in the system would therefore be able to at least partly cover additional demands from other sectors.
Further, we modelled hydropower operation on an aggregated level for the whole of Sweden.
Therefore, detailed assessments of environmental impacts are not possible. Also, some of the dispatch schedules may be physically impossible once down-scaled to the level of river basins. We recommend further work here.

Conclusions
We have assessed how electrolyzer operation evolves if a continuous stream of hydrogen should be produced and if power demand of electrolyzers is met on average by new wind power capacities for our Swedish case study. We have shown that while in all scenarios, the average annual utilization of electrolyzers is above 60%, the inter-annual variability of hydrogen production is high unless thermal power is dispatched for electrolysis.
Furthermore, if hydropower flexibility is additionally restricted to reduce hydropeaking, interannual variability is increased further. As the maximum constraints in hydropower generation are, however, only met rarely, one important policy conclusion is that allowing for high, yet rare, extreme operation of hydropower, can make the whole system more resilient. This calls for more research on the ecological impacts of rare hydropeaking events and for a detailed, river-scale assessment of hydropower generation under large penetration of variable renewables.
Due to high inter-annual variability, either long-term storage of hydrogen, backup hydrogen sources or a dispatch of thermal capacities in extreme years is therefore necessary to maintain a stable hydrogen flow to the industry. This means that on average hydrogen costs can be low, but extreme years with high costs have to be expected.
Adding more wind power to the system while also adding large electrolyzer capacity as main consumer of the wind power makes the system more stable, if electrolyzers ramp down in rare hours of extreme events with low availability of renewable generation. The need for additional backup capacities in a fully renewable Swedish power system are reduced in such a system.
We did not explicitly assess the costs of hydrogen production, however, our results, openly available, can provide fundamental input to such kind of analysis by others.
x thermal h,type ≤ thcap type Additional constraints for hydro flow, ramping and the storage level are controlled with the following equations: Equation A.5 ensures that the reservoir level x reservoir level h of one specific hour equals the reservoir level of the previous hour, plus the natural inflow natural inf low h , minus the outflow (hydropower production x hydro h and spill x spill h ).
The reservoir level x reservoir level h at the end of each simulated year is controlled by equation A.6, which ensures that the level is above min endstorage.
x reservoir level h min endstorage h ∈ {8760} (A.6) The mimimum flow in each hour is controlled by equation A.7, which ensures that hydropower production x hydro h and hydro spilling x spill h is larger than the minimum flow min f low h .
x hydro h + x spill h min f low h ∀h (A.7) Likewise, the maximum possible flow max f low h is defined by equation A.8.
x hydro h + x spill h max f low h ∀h (A.8) The maximum change in total flow per hour (maximum ramping, MRR) max hydro ramp h is defined by equations A.9 and A.10.
Maximum ramping max hydro prod ramp of hydropower production x hydro h is controlled by equations A.11 and A.12.
x hydro h − x hydro h−1 max hydro prod ramp ∀h (A.11) x hydro h−1 − x hydro h max hydro prod ramp h ∀h (A.12) Restrictions of thermal generation x thermal h,type are defined by the following equations. Minimum min thermal h,type and maximum max thermal h,type thermal generation by type are controlled by equations A.13 and A.14.
x thermal h,type min thermal h,type ∀h, type (A.13) x thermal h,type max thermal h,type , ∀h, type (A.14) Thermal ramping, summed over all types, is constrained by equations A. 15   Maximum hydro flow min f low MW see figure 2 Maximum hydro production ramp max hydro prod ramp MW 4000 Minimum thermal production by type min thermal h,type MW see figure 1 Maximum thermal production by type max thermal h,type MW see figure 1 Capacities, Ramps & Restrictions Maximum thermal ramping max thermal ramp h MW 1500 Appendix A.2. Input data for thermal power production costs Power production costs were set based on bottom-up technology and fuel specific projections of electricity generation costs [54,64], and were adjusted to e 2018 using the average 2018 currency exchange rate of 1 e = 10.3 SEK [65] and updated fuel prices. The production costs include fuel costs including a CO 2 -charge (50 e /ton CO2 ), costs for operation and maintenance (O&M), and heat credits, when applicable.  Appendix A.3. Detailed results on thermal power production and need for backup capacity The left column always shows lower thermal generation than the right column, indicating that decreasing hydropower flexibility will increase thermal dispatch. The first row shows a situation without electrolyzers. Thermal power production is dominated by biomass generation, with a small base-load production from waste incineration. Here, biomass plants and partly peaking plants from natural gas and oil are dispatched to balance wind power in the system.
The mid row shows the No thermal scenarios for a Large electrolyzer capacity. It is a situation where thermal generation is at its almost constrained minimum, as it is not dispatched for hydrogen production and wind power capacities in the system are high. Natural gas and oil plants are almost not dispatched.
In the bottom row the Thermal scenarios for the Large electrolyzer capacity scenarios are shown. The dispatch of biomass capacities is significantly increased here, as well as peaking natural gas and oil plants. This indicates that to achieve high utilization of electrolyzers in some years fossil generation has to be extensively dispatched (up to 5 TWh) to guarantee hydrogen supply.
However, as a share of total thermal generation, fossil generation is very low in all scenarios and never ultrapasses 1 TWh of annual generation on average over all 29 weather years.