Techno-Economic Assessment of Demand-Driven Small-Scale Green Hydrogen Production for Low Carbon Agriculture in Sweden

Wind power coupled to hydrogen (H2) production is an interesting strategy to reduce power curtailment and to provide clean fuel for decarbonizing agricultural activities. However, such implementation is challenging for several reasons, including uncertainties in wind power availability, seasonalities in agricultural fuel demand, capital-intensive gas storage systems, and high specific investment costs of small-scale electrolysers. To investigate whether on-site H2 production could be a feasible alternative to conventional diesel farming, a model was built for dynamic simulations of H2 production from wind power driven by the fuel demand of a cereal farm located on the island of Gotland, Sweden. Different cases and technological scenarios were considered to assess the effects of future developments, H2 end-use, as well as production scale on the levelised- and farmers’ equivalent annual costs. In a single-farm application, H2 production costs varied between 21.20–14.82 €/kg. By sharing a power-to-H2 facility among four different farms of 300-ha each, the specific investment costs could be significantly decreased, resulting in 28% lower H2 production costs than when facilities are not shared. By including delivery vans as additional H2 consumers in each farm, costs of H2 production decreased by 35% due to the higher production scale and more distributed demand. However, in all cases and technological scenarios assessed, projected diesel price in retailers was cheaper than H2. Nevertheless, revenues from leasing the land to wind power developers could make H2 a more attractive option even in single-farm applications as early as 2020. Without such revenues, H2 is more competitive than diesel where power-to-H2 plants are shared by at least two farms, if technological developments predicted for 2030 come true. Also, out of 20 different cases assessed, nine of them showed a carbon abatement cost lower than the current carbon tax in Sweden of 110 €/tCO2, which demonstrate the potential of power-to-H2 as an effective strategy to decarbonize agricultural systems.


INTRODUCTION
Renewable energy sources can be exploited in remote areas with limited interconnection such as islands and/or agricultural farmlands to increase energy independence and security. Recently, declining costs of solar and wind power combined with policies and incentives to tackle climate change have created favorable conditions to further expand renewable energy production in such regions. However, due to its intermittency and uncertainty (especially for wind), high levels of variable renewable energy (VRE) are challenging to integrate into current energy systems, frequently resulting in a mismatch between supply and demand. Such imbalances cause fluctuations in grid voltage and frequency, as well as curtailment of power production, considerably increasing the overall costs of the system.
For this reason, different energy storage technologies have been developed for several applications, in particular to avoid curtailment of power production, and to support stable operations of electric grids (Fischer et al., 2018b;Koohi-Fayegh and Rosen, 2020). To compensate for production fluctuations as well as providing benefits at the system level (e.g., control reserve energy), H 2 based storage systems have been proposed (Grueger et al., 2017). Also, as a clean and versatile energy carrier, H 2 may have an important role in future low-carbon pathways, for instance, to produce gaseous (e.g., CH 4 and NH 3 ) and liquid fuels (e.g., methanol, gasoline, and dimethyl ether), heat or even directly used as fuel for mobility (Hanley et al., 2018).
In grain-based agricultural systems, nitrogen fertilizers and fossil fuel consumption are responsible for the majority of the GHG emissions (Yan et al., 2015). Where VRE is deployed on farmland, an interesting concept to decarbonize agricultural activities is to also include H 2 storage. Thus, curtailment could be avoided and local renewable electricity could be used to produce H 2 to displace diesel as a fuel in tractors and/or used to make NH 3 for fertilizer via the Haber-Bosch process (Moreda et al., 2016;Allman and Daoutidis, 2018). While the latter may be restricted to large farming operations (minimum megawatt-scale equipment), H 2 as fuel could potentially be used in small-and mid-size farms since it has been successfully implemented at the kilowatt-scale in industrial applications such as welding and brazing, material handling vehicles (e.g., forklifts and airport towing trucks) as well as for mobility (e.g., golf cart and longrange passenger cars) (Allman et al., 2017;Apostolou et al., 2019). Thanks to its higher energy density compared to lithium-ion batteries, H 2 -based fuel cell agricultural machinery (FCAM) may be preferred to manned battery-electric since agricultural operations often require continuous hours of heavy fieldwork (NHA, 2012;Wu et al., 2019;Lagnelöv et al., 2020). In addition, during the conversion of electricity to H 2 via water electrolysis, oxygen (O 2 ) and low-temperature waste heat (WH) at 60-90°C are produced which could be valorized (Buttler and Spliethoff, 2018). For instance, WH could be used for drying grains or heating greenhouses, while O 2 could be used in aquaculture, in particular for sensitive species like salmon and trout (García et al., 1998;Mariani et al., 2016;Linde, 2017b). These applications should improve the sustainability of the concept as well as reduce costs associated with H 2 production. However, the on-farm production of H 2 to be used in FCAM is challenging: 1) farming is a highly seasonal activity in which typical operations like harrowing, sowing, fertilizing, plowing and harvesting occur over short periods of time, resulting in peaks of fuel demand; 2) VRE production is uncertain by nature increasing risks of mismatch between supply and demand; 3) Large gas storage to compensate for such seasonalities are capitalintensive, and 4) decentralized small-scale electrolysers have higher specific investment costs increasing production costs compared to larger facilities.
Small-scale H 2 production via water electrolysis has been investigated for different applications. For instance, Fischer et al. (2018a) developed a predictive control model for a 120 kW proton exchange membrane (PEM) electrolyser, injecting H 2 into the natural gas grid according to fluctuating electricity prices in the spot market and within network limitations. In an energy system dominated by hydropower production, Ulleberg et al. (2020) examined the deployment of small-scale electrolysers coupled to H 2 refueling stations for fuel cell electric vehicles. Similarly, Apostolou et al. (2019) further down-scaled the process proposing the use of a 50 kW wind turbine coupled to a 70 kW alkaline electrolyser to supply H 2 for fuel cell electric bicycles in a green urban mobility concept. Also, H 2 refueling stations with electrolysers smaller than 500 kW to supply the demand of H 2 cars, and the optimization of an electrolyser operation employing wind, electricity prices, and H 2 demand have been investigated elsewhere (Grüger et al., 2018;Grüger et al., 2019). Furthermore, the feasibility of stationary power-to-gas systems to store excess electricity from renewable sources in buildings with different heat and power requirements have been assessed combined with oxy-fuel boilers to produce concentrated CO 2 stream and facilitate further methanation of H 2 (Bailera et al., 2018;Bailera et al., 2019). Even though previous studies addressed H 2 production from solar PV to fuel an all-wheel drive vehicle on a winery Roda et al., 2018), to the best of the authors' knowledge, techno-economic assessments of on-farm H 2 production based on wind power to supply the fuel demand of heavier agricultural machineries like tractors and harvesters have never been reported. Such a concept could provide multiple benefits, curtailment could be avoided increasing the income of wind power project developers, locally produced clean fuel would be provided to decarbonize agricultural activities, and land leasing payments would be provided to farmers.
In Sweden, the local authorities of Gotland's island have committed to an ambitious plan to be self-sufficient in energy by 2025. For this reason, local wind power production is planned to increase 5-fold (to around 1,000 MW) while grid interconnection to the mainland will be restricted to 500 MW. Nowadays, major efforts are being made by different research initiatives to develop feasible options to store and manage excess electricity that may occur (GEAB, Vatenfall, ABB, and KTH, 2011;Byman, 2015;Mohseni et al., 2017;Wallnerström and Bertling Tjernberg, 2018). Our study differentiates from previous investigations by focusing on developing a modeling tool for discrete-event simulation of H 2 production according to the fuel demand of cereal-based farms located on Gotland. The model was used to find optimal plant configurations that minimized the levelized cost of H 2 (LCOH 2 ) according to the following cases: 1) single-farm H 2 production for FCAM; 2) shared infrastructure between two farms for FCAM and fuel cell minivan (FCMV); and 3) increased scale production by sharing the PtH 2 plant among four farms for FCAM and FCMV. Optimal plant configurations were used for further assessment of the equivalent annual cost (EAC) to compare the cost of ownership of FCAM and conventional diesel agricultural machinery in different technological scenarios (2020 and 2030). Additionally, to provide insights for policymakers on possible decarbonization strategies, the carbon abatement cost of each case assessed was also calculated.

System Description
The power-to-hydrogen (PtH 2 ) plant refers to an electrolyser, compressor, storage system, and a dispenser located on a farm on the island of Gotland, Sweden (57°30′N 18°33′E/57°50′N 18°55′E). A proton-exchange membrane (PEM) electrolyser was chosen due to its suitability for small-scale applications. The overall reaction of H 2 production by water electrolysis is shown in Eq. 1: The electricity is primarily obtained from wind turbines located inside the farm boundary. However, during system downtime and for safety infrastructure, electricity is also obtained from the grid (regulated market) in small volumes. To allow storage at 500 bar, H 2 is compressed as soon as it is produced in the stacks (Linde, 2014). The H 2 is supplied according to the demand of FCAM and where applicable FCMV used for delivery (see Agricultural H 2 Demand). Additionally, the economic benefits of utilizing low-temperature WH at 60°C in a greenhouse for growing tomatoes (see Appendix A), and O 2 for on-site fish farming are considered (Linde, 2017a;Linde, 2017b;Törnfet and Nypelius, 2020).
In this system, farmers cooperate with wind power project developers in a business model where farmland is leased to wind power production securing additional revenues to farmers and improving wind power output, in case H 2 is produced at times of constrained power grid. Figure 1 and Table 1 show the technical system boundary and an overview of the characteristics of PEM electrolyser considered in the present study.

Dynamics of the Power-to-Hydrogen Plant Operation
The H 2 demand of FCAM and FCMV and the H 2 level in the storage tank determine whether the electrolyser should enter in operation. Therefore, whenever H 2 storage is low and wind power production is sufficient to run the electrolyser on full-load, H 2 is produced until the storage tank is full. At times of no wind power production or full H 2 storage, the system is put directly on cold standby since the time required to ramp-up a PEM electrolyser is negligible (Buttler and Spliethoff, 2018). Thus, cold standby solely defines the non-operating hours (NOH) of the system. The energy consumed during NOH and by the safety infrastructure is purchased from the regulated market with a fixed tariff of 100 €/MWh as the quantities are too low to qualify for a cheaper tariff (e.g., day-ahead spot market). A schematic diagram of the dynamic PtH 2 plant operation is presented in Figure 2 (additional information is provided in Power-to-Hydrogen Model and Optimization Procedure).

Cases Assessed
To assess the influence of H 2 demand on the economic performance of the agricultural PtH 2 plant, different cases were investigated under the current and future technological scenarios (2020 and 2030). A single-user PtH 2 plant was used as a reference in case 1, while the option of sharing the PtH 2 with more than one farm was investigated in Cases 2a and 3a. The influence of having FCMV for delivery in addition to the H 2 demand of FCAM was investigated in Cases 2b and 3b. In all cases, wind power per land area was calculated as an average value according to local conditions found in Sweden (approx. 6.5 MW/ km 2 ) (Stadkraft, 2020). Thus, wind power capacity was used to determine land lease revenues and, combined with the specific wind power production (see Wind Power Production), it was also used for modeling wind power production/availability for PtH 2 applications. A summary of all cases assessed is found in Table 2.

Wind Power Production
Historical wind speed measurements were used to simulate wind power production. Hourly values for 2017 were obtained from the meteorological station at the Visby Airport (57°66′N 18°34′E) on Gotland, Sweden. The station is located at 42 m above sea level and measures wind at 10 m high from the ground (SMHI, 2017). The wind speed and wind direction are shown as a wind rose in Figure 3A as well as the wind speed frequency ( Figure 3B). Wind speed was extrapolated to the turbine hub height of 95 m using the power law with an exponent of 0.13 . The power curve of the V90 2.0 MW wind turbine (Vestas, Denmark) was used to convert wind speed into power for parks with 10-40 wind turbines depending on the case assessed (see Agricultural H 2 Demand). Such turbine cuts in at 4 m/s, is rated at 13 m/s, and cuts out at 25 m/s (Vestas, 2019).

Price of Electricity Used for H 2 Production
Hourly values from the day-ahead spot market of the Nord Pool power exchange were used to calculate the electricity costs to run the PtH 2 plant (NordPool, 2019). Thus, the electricity used is accounted for as an opportunity cost if the wind power operator would have the option to sell electricity to the grid. Even though electricity prices can vary significantly when different years are compared (Janke et al., 2020), 2017 was chosen since the average price found in this year is representative of historical values between 2013-2018 in Sweden. The price distribution in the dayahead market of the Nord Pool power exchange for the SE3 region in 2017 is shown in Figure 4. We do not consider discounting the electricity price as the benefits offered to the wind farm developer (reduced curtailment, system flexibility) are captured in the land leasing payment made to the landowner/ farmer (see Power-to-Hydrogen Model and Optimization Procedure).

Agricultural H 2 Demand
In the present study, H 2 demand is modeled according to the requirements of two different consumers, namely FCAM (agricultural machinery) and FCMV (minivan). FCAM H 2 demand was estimated for a cereal farm in Sweden according to the model described by Lagnelöv et al. (2020) based on dynamic discrete-event simulation with embedded state-based logic for decision making. The simulated farm encompassed 300ha equally distributed between barley, oats, spring wheat, and winter wheat crops. The model used a conventional cropping system with work beginning in mid-March and ending at the start  of November (Lagnelöv et al., 2020). One of the main aspects of this model is a workability control based on weather conditions and soil moisture content, which considered the water balance model described in (Witney, 1988) and tested by (Nilsson and Hansson, 2001). In the present study, weather and soil data from the island of Gotland (Sweden) were used instead of the values in the original report. The soil type on Gotland is mainly sand or sandy loam (Lundblad, 2015;Paulsson et al., 2015) and the soil density, field capacity, saturation, permanent wilting point, and plastic limit of sandy loam described in (Witney, 1988) was assumed adequate and used. Weather data on monthly mean air temperature, number of daily sunshine hours, and hourly precipitation were obtained from the meteorological station at the Visby Airport on Gotland, Sweden (57°66′N 18°34′E). Even though for wind power production and electricity prices data from 2017 was considered, to model the agricultural H 2 demand, data of precipitation, air temperature, and sunshine hours from 2016 were considered since they better represented average values in the region for the period 1989-2018 (SHMI, 2020). The H 2 demand during crop harvesting was modeled based on 28% of total fuel demand in a cereal farm according to Safa et al. (2010). The instantaneous power demand for the FCAM and the refuelling station was measured separately and were both assumed to be linear average values. The average refuelling time considered was 0.32 h and the refuelling station was assumed to have a constant H 2 flow for the duration. The input values used to simulate the H 2 demand of FCAM are shown in Table 3. H 2 demand of FCMV was estimated for an average driving of 35,000 km/year with a diesel-equivalent consumption of 4.43 L/100 km. The total consumption of 1,825 L/year (8.86 L/day) was equally distributed throughout the year and added to the H 2 demand of FCAM when applicable (see Agricultural H 2 Demand).

By-Products Recovery
As the primary aim of the current study is to investigate H 2 production on-demand, by-products production is not optimized. However, it is nevertheless possible to recover the produced WH and O 2 for individual end-use applications. To valorize the WH stream, the size of the greenhouse was varied from 1,000 to 10,000 m 2 for tomato production according to the rated power of each simulated electrolyser size (50-500 kW). A water tank (heat capacity of 70 kWh/m 3 ; 5% whole system thermal losses assumed) with up to 24 h of full load capacity is used to account for short-term imbalances between heat supply and demand such as daily fluctuations during summertime (Novo et al., 2010;Guelpa and Verda, 2019). Along with the varying greenhouse size, different sizes of water tanks were also considered from 3 to 30 m 3 with a specific investment cost of 40 €/m 3 (Guelpa and Verda, 2019). More information about the heat demand estimation can be found in Figure A1 in Appendix A.
For the on-site use of O 2 , it was considered that all assessed farm configurations are combined with a 2,500 m 3 tank for rainbow trout cultivation where O 2 is injected for controlling the dissolved oxygen levels in the water. A stock density of 15 kg/m 3 was applied with an average specific O 2 consumption of 350 mg O 2 /kg/h (Boyd, 2011;Woynarovich et al., 2011). As the produced O 2 from the electrolyser can only offset 23-27% of the total demand for fish farming, each tank is equipped with a dedicated O 2 generation system based on pressure swing adsorption (PSA) technology with a power consumption of 1.1 kWh per m 3 of O 2 (85% v/v) (Aquaculture Technology, 2020). Given an electricity tariff of 100 €/MWh (regulated market), those characteristics result in an O 2 production cost of around 0.19 €/kg. Hence, this value is further used to monetize the O 2 production from the electrolyser.

Power-to-Hydrogen Model and Optimization Procedure
The PtH 2 model was implemented in the Matlab-based Simulink environment version R2019b (MathWorks, USA). Individual equations are discretized for a fixed step size (sampling time) of 1 h. It is based on variable hourly values of wind power production, day-ahead spot market price, and fuel demand. The PEM electrolyser was modeled in combination with a compressed gas storage system to assist H 2 production and delivery on-demand. The model calculates H 2 , WH, and O 2 production as well as run hours and total electricity cost. The decision whether the electrolyser should enter into operation is dependent on the amount of H 2 available in the gas storage and the availability of wind power to run the electrolyser on full-load as described below (Eq. 2): where E i -electrolyser operation mode (binary), V H 2 ,i -gas storage volume in each hour i (m 3 at 500 bar), V H 2 ,maxavailable gas storage size (m 3 at 500 bar), W wind,i -wind power production in each hour i (MWh), W elec -hourly power consumption of the electrolyser on full load (MWh). H 2 production in each hour (m H2,i ) is calculated based on the power consumption of PEM electrolysis in 2020 and 2030 (Eq. 3): where ρ H2 -H 2 density (0.08988 kg/m 3 at STP), W H 2 ,i -specific power consumption during operation mode (4.6-4.9 kWh/m 3 H 2 at STP). O 2 production (m O 2 ,i ) is calculated based on hourly H 2 production and the molar mass of H 2 O, H 2, and O 2 (4 and 5): where r O2 -molar mass ratio of O 2 /H 2 O (0.888093). Waste heat production in hour i (W heat,i ) is calculated as a fraction of the consumed power during operation mode (Eq. 6): where f heat -fraction of electrolyser's power consumption that becomes available heat (0.171). The run hours of the system per year (R ON ) is defined as the sum of hourly events that satisfies the condition needed to the electrolyser enter on operation mode (Eq. 7): The costs associated with electricity use during the electrolyser operation (C elec,i ) are based on wind power consumption (W H2,i ) and from the grid for safety infrastructure (W safe ) as follows:   where T spot,i -is the day ahead spot market price in each hour i of the electrolyser operation (€/MWh), T grid -is the fixed tariff for grid-based power (100 €/MWh), P N -electrolyser's nominal rated power (MW) The yearly costs to keep the electrolyser on cold standby (C cold ) during non-operating hours is described in Eq. 9 as follows: The power consumption during cold standby and for safety infrastructure is based on a 1.074 MW plant and is proportionally adjusted to each size of electrolyser assessed (Frank et al., 2018). To allow gas storage at 500 bar, H 2 is compressed requiring 2.2 kWh/kg H 2 (W comp ) (Linde, 2014). The costs associated with H 2 compression (C comp,i ) in each hour i are described in Eq. 10 below: Finally, the total electricity cost of the PtH 2 plant (C total ) is based on costs associated during the electrolyser operation (C elec,i ), H 2 compression (C comp,i ) and to keep the electrolyser on cold standby (C cold ) (Eq. 11): To determine the optimal plant configuration a total number of 256 simulations were run for each case assessed. Each simulation corresponded to a combination of electrolyser size between 50 and 500 kW (30 kW increments) and gas storage capacity between 10 and 50 m 3 (2.66 m 3 increments). For each plant configuration, specific CAPEX (€/kW), capacity factor, the average price paid for the electricity, and the LCOH 2 were calculated and used for assessment. To verify whether the PtH 2 plant configurations were fulfilling the consumers' fuel requirement, the delivery of H 2 on-demand was considered a mandatory criterion. The characteristic dependencies of different plant configurations on each performance indicator were visualized using Matlab function contour 3-days plot (MathWorks, USA). For each case assessed, the combination of electrolyser size and gas storage capacity that resulted in the lowest LCOH 2 and simultaneously fulfills H 2 demand was considered the optimal plant configuration.

Economic Assessment
The economic performance of the system was assessed based on two economic indicators, namely the LCOH 2 and equivalent annual cost (EAC). While the LCOH 2 is used to optimize the PtH 2 plant configuration in terms of electrolyser size and H 2 storage capacity, the EAC is used to compare the H 2 system with a conventional diesel-fueled one. To determine the EAC, the net present value (NPV) is first calculated as follows (Eq. 12): where CAPEX is the capital expenditures of the PtH 2 or diesel system; k is the discount rate estimated at 6.5% per year for onshore wind projects in Nordic countries (Thornton, 2019); y is the 25 years lifespan of the project. The net cash flow (NCF) is the operational expenditures subtracted by the land lease over the lifespan of the project as per Eq. 13: where Heat use is annual savings produced by utilizing the electrolyser's waste heat as opposed to traditional heating at a conservative value of €75/MWh (Wiederholm et al., 2018); O 2 use annual savings produced by utilizing the electrolyser's waste O 2 at a saving of 0.19 €/kg; Land lease is the yearly income by leasing the land for wind power production (6,000 €/MW per year) (McGreevy, 2013); OPEX y is the yearly operational expenditures of the PtH 2 or diesel systems. Here, fixed as well as variable costs such as electricity and water for the PtH 2 plant and fuel consumption for the diesel system are considered. The latter is assumed to be equivalent to the H 2 demand in energy units multiplied by the tank-to-wheel efficiency ratio between fuel cell and diesel vehicles (50%/ 30%) (Moreda et al., 2016). Different diesel prices are considered to depict the influence of agricultural diesel tax relief as well as a future fuel prices in Sweden. A summary of the different prices considered in the present study are shown in Table 4.
Finally, the EAC is calculated as the cost per year of owning and operating the PtH 2 and diesel-fueled agricultural systems over the lifespan of the project as follows (Eq. 14): The LCOH 2 is the breakeven selling price of the H 2 produced and is given by Eq. 15 below: LCOH 2 n y 0 costs in year y (1+k) y n y 0 kWh of H2 produced in year y (1+k) y All indicators are calculated in 2018 euros. The timeline for relevant calculations includes a 3-year commissioning phase, 25 years of operation, and one-year decommissioning. Also, additional costs, such as land, permits, transport, site preparation, engineering, and design costs, grid connection as well as contingency were assumed to be equivalent to 10% of the electrolyser's CAPEX (Benjaminsson et al., 2013).The economic model does not consider reductions in electrolyser performance over time, however, component replacement costs are included in economic assessment (2 replacements over project's lifetime). Even though our study uses the most recent literature available, unavoidable uncertainties exist in capital expenditures (Schmidt et al., 2017). CAPEX and OPEX values of PEM electrolyser used in this study are shown in Table 5.

RESULTS AND DISCUSSIONS H 2 Demand
As described in Agricultural H 2 Demand, H 2 demand was modeled for different farm cases which include FCAM, some also include FCMV. As an example, Figure 5 shows the demand profile at the dispenser for case 2b where two farms share a PtH 2 plant to fuel their agricultural machinery and one FCMV each. As expected, H 2 demand for FCAM is highly seasonal, there is no demand during winter, extended parts of the summer, and some interim periods when no fieldwork is required. In contrast, H 2 demand of FCMVs occurs on a year-round basis, however, requiring much less energy than FCAM per refill. In fact, the total H 2 demand of the FCMVs in case 2b was just 28% of the total fuel demand.
When the PtH 2 plant is scaled-up to fulfill the H 2 demand of four farms including one FCMV each (case 3b), the fuel demand is double that seen in Figure 4 with the same demand profile. Conversely, where a single farm operates a PtH 2 plant (case 1) FCAM demand is halved and FCMV is disregarded.

Optimization of H 2 Production
For the optimization of the PtH 2 plant, the electrolyser and storage capacity sizes were varied to find plant configurations that resulted in the lowest possible LCOH 2 . This procedure was performed for each farm case as well as for different technological scenarios assessed ( Figure A2 in Appendix B). Again, case 2b (2020) is used as an example ( Figure 6).
Economies of scale are significant in the ranges examined and heavily influenced the economic performance of the agricultural PtH 2 plant (Zauner et al., 2019). However, as shown in Figure 6B, increasing electrolyser sizes also led to lower capacity factors. Such behavior is explained by the plant being driven according to the specific H 2 demand, thereby increasing electrolyser capacity did not necessarily result in higher H 2 production. Previous studies on electrofuels production showed that the number of running hours of the plant and the price paid for the electricity were the most important factors to minimize the production costs for a fixed capacity Janke et al., 2020). In the present study, as the average price paid for the electricity varied less than 10% among all simulated conditions, it was indeed the capacity factor that most influenced the H 2 production costs. For instance, in case 2b (2020) the lowest LCOH 2 was found for a plant with a 140 kW of electrolyser size and 15 m 3 (500 bar) of storage capacity (i.e., equivalent to 11 days of full-load operation). This plant configuration resulted in 3,060 h/year of operation and it was able to produce H 2 at a cost of 15.87 €/kg. When the H 2 demand of FCMN is disregarded (case 2a-2020), a comparable PtH 2 plant would operate 11% less (2,715 h/year), which in turn results in around 6% higher H 2 production costs. In contrast, if the electrolyser size would be reduced to lower than 140 kW, the number of operating hours would increase, which in theory could potentially reduce the production costs. As observed in Figure 6B, however, if smaller electrolysers are used H 2 is not delivered ondemand, thus excessively small electrolysers are not considered suitable for farm operations even if coupled to large storage capacities (expensive option).
In fact, due to the highly seasonal fuel demand and relatively high cost of additional storage capacity, it is challenging to design a PtH 2 plant with sufficient run hours able to truly minimize the LCOH 2 . A previous study on PtCH 4 showed that at least 5,000 operating hours per year (57% capacity factor) would be required, and values lower than 4,000 h/year would likely result in prohibitive production costs . For case 2b (2020) the LCOH 2 of 15.87 €/kg is equivalent to a diesel price of 4.02 €/L which is indeed prohibitive when compared to the assumed diesel retail price of 1.35 €/L. Such diesel price includes a carbon tax of 110 €/tCO 2 applied for fossil fuel consumption. In countries like Sweden where farmers pay less for consuming fossil fuel due to the relief on the existing carbon tax, the adoption of alternative fuels by farmers becomes even more challenging since the real diesel price paid by farmers is around 1.17 €/L.
In case farmers organize themselves in a small cooperative where four farms share the same PtH 2 infrastructure to supply fuel for their agricultural machinery and one FCMV in each  (Skatteverket, 2020). farm (case 3b), the system is up-scaled to an optimal configuration of 290 kW electrolyser and 26 m 3 (500 bar) of storage capacity. Even though this higher H 2 demand does not necessarily result in major changes in the capacity factor of the plant, the specific CAPEX is reduced by 17% compared to sharing the infrastructure with just two farms (case 2b), which in turn proportionally reduces the H 2 production costs (Table 6). New composite materials for compressed H 2 storage and reduced use of noble metals like platinum and titanium in PEM electrolysis will result in lower costs in the future (Schmidt et al., 2017;Moradi and Groth, 2019). As no changes in optimal plant configurations were found in 2030 compared to 2020, these technological developments are considered the main reason for the 30% reduction in LCOH 2 observed. Interestingly, a previous study from our group based on H 2 production without demand constraints, showed a lower reduction of 18% in production costs when comparing 2020 and 2030 technological scenarios (Janke et al., 2020). In that case, the higher capacity factor of the electrolyser (≥75%) increased the total expenses with electricity purchase, thereby reducing the effect of CAPEX on the LCOH 2 .

Effect of By-Products Recovery
The PtH 2 plant produces and delivers H 2 according to the demand of FCAM and FCMV, however, the process of water electrolysis also results in O 2 production mediated by an exothermic reaction (Eq. 1 described in System Description). As PEM electrolysers are operated under controlled temperature (50-80°C), a water-based cooling system needs to be integrated to avoid overheating of the cell (>100°C), thereby also allowing the recovery of low-temperature waste heat (Buttler and Spliethoff, 2018). The feasibility of valorizing these byproducts depends on local demand. For instance, our farm includes intensive tomato cultivation in a greenhouse, which requires temperature control for year-round production. In this case, it is assumed that the electrolyser's cooling system could be integrated to the heating system of the greenhouse, offsetting the heat required from conventional sources (Wiederholm et al., 2018). Furthermore, O 2 use in aquaculture has gained attention in recent years, in particular in recirculating aquaculture systems that require high levels of dissolved oxygen to allow high production densities. As O 2 would be usually generated on-site via energy-intensive PSA systems, water electrolysis could partly supply O 2 to aquaculture offsetting costs associated with the oxygenation process. Both WH and O 2 valorization would positively impact the economic performance of the PtH 2 plant. Such benefits in terms of LCOH 2 reduction are shown in Figure 7.
Independent of the case and/or year assessed recovering O 2 showed to be more valuable compared to WH. On average, a reduction by 12% on the LCOH 2 was possible by recovering O 2 , while WH was able to reduce the production costs by approximately 5%. This is mostly explained by the large quantities of O 2 generated by the water electrolysis process, i.e., 88% of H 2 O mass becomes O 2 . Thus, assuming a price of 0.19 €/kg, O 2 recovery substantially improves the economic performance of the process. When both O 2 and WH are valorized, the LCOH 2 is reduced on average by 17%.
However, diesel is still cheaper than H 2 in all cases and years at the given prices. For instance, in 2020 the lowest LCOH 2 found in case 3b (11.44 €/kg) was between 2.14 and 2.47 times higher than diesel with and without carbon tax respectively. In 2030, when diesel prices are expected to be 20% higher and H 2 production costs 23% lower than in 2020 (case 3b with WH and O 2 recovery), such differences are reduced to 1.15-1.33 times higher than diesel depending on the carbon tax scheme considered.
As clearly observed, purchasing diesel is cheaper than on-farm H 2 production, for all PtH 2 cases and technological scenarios considered. However, due to significant differences in terms of tank-to-wheel efficiency and purchase costs between FCAM and conventional diesel agricultural machinery, further analysis is required to understand the competitiveness of small-scale H 2 production for farming activities.

Equivalent Annual Cost
The equivalent annual cost (EAC) was assessed as an additional economic indicator to understand the H 2 system from the farmers' perspective. The EAC is used to compare the cost of owning fuel cell or diesel vehicles over the lifetime of the PtH 2 plant. In addition, a scenario where farmers finance H 2 production and use by means of leasing land to wind power project developers is also considered. Such a business model is considered advantageous for both parties: 1) farmers obtain additional revenues by leasing their land for wind power production; 2) wind power project developers potentially enhance their wind power production by selling curtailed electricity to farmers; 3) farmers can locally produce clean fuel to decarbonize their activities, and 4) support for the wind farm will likely be much greater with local involvement. The EAC according to the different farm cases and technological scenarios assessed are found in Figure 8.
Important differences were observed among the farm cases, in which sharing the PtH 2 plant between two farms reduced the EAC by 21% on average, and sharing the PtH 2 plant among four farms reduced annual costs by 27%. In contrast, no major benefits were found if farmers share the same diesel refueling infrastructure since the CAPEX of the diesel system is considerably lower than the H 2 one. Nevertheless, EAC values associated with H 2 production and use were always higher than conventional diesel farming for the period 2020 unless land lease revenues are counted.
Similarly to the LCOH 2 , the EAC of the H 2 system will be considerably lower in 2030. In this case, annual costs would be around 30% lower compared to 2020 values. In the meantime, diesel prices are expected to increase by around 20%, reaching values between 1.40 and 1.62 €/L depending on the carbon tax scheme considered. Due to these factors, the EAC of the H 2 system becomes competitive with diesel, except for case 1 which  still will be more expensive. For case 2a, H 2 becomes cheaper than diesel if farmers are not entitled to carbon tax relief on diesel consumption. For all remaining cases in 2030 H 2 shows lower or equal EACs than diesel. Unsurprisingly, the case that presented the lowest EAC (3a-2030, no FCMV) was not the same case that showed the lowest LCOH 2 (3b-2030, inc. FCMV). This is explained by the LCOH 2 being inversely proportional to the amount of H 2 produced (Eq. 5) which increases with the inclusion of FCMV demand, while the EAC is only marginally influenced by production via the NPV (Eq. 2-4). Thus, by adding the H 2 fuel demand of FCMV, the increase in cost is greater than the savings produced from having H 2 production, however, we did not compare this to diesel minivans as FCAM was the focus of this study.
When the revenues for leasing the land to wind power project developers are taken into account (6,000 €/MW/year), a major impact on the EAC is observed in favor of the H 2 system. In this case, instead of having costs associated with agricultural machinery, farmers would have annual gains by operating the PtH 2 plant in all farm cases and technological scenarios assessed. In cases where H 2 production and use is less competitive than diesel, only a minor share of the land lease revenues would be required to make H 2 competitive with diesel. For instance, under the current technological scenario, between 10-26% (600-1,560 €/MW/year) is needed to finance H 2 production and use. In the 2030 scenario, only case 1 requires additional assistance from land lease revenues to make it competitive with diesel. In this case, the fraction of land lease revenues needed would be reduced from 26% to just 8% (498 €/MW/year).

Carbon Abatement Cost
The implementation of a farm-based PtH 2 plant results in carbon emission reductions from different sources, namely direct fossil fuel displacement by H 2 , power consumption from the grid by recovering O 2 from the electrolyser, and reductions in district heating use also by recovering WH from the electrolyser. As the latter two are dependent on local characteristics such as variable emission factor from the grid and use of fossil fuel in district heating systems, a simplified approach to calculate the cost of carbon mitigation was performed solely focusing on diesel displacement by the produced H 2 . Considering a diesel consumption of 19,446 L/farm/year without FCMV, 27,131 L/farm/year with FCMV and the diesel emission factor of 2.64 kgCO 2 /L, the carbon emission reductions provided by the PtH 2 plant could be estimated. In addition, the difference in EAC between H 2 and diesel with and without carbon tax relief was used to calculate the carbon abatement cost of each case in different technological scenarios (Figure 9).
It is possible to observe that under the current technological scenario without land lease revenues the carbon abatement cost is considered high, with values above 100 €/tCO 2 . The case 3b showed, however, the lowest carbon abatement cost in 2020 with values close to the current carbon tax in Sweden (110 €/ tCO 2 ), in particular if the diesel tax relief entitled to farming activities would be excluded. In fact, state subsidies and taxes often influence positively or negatively the cost efficiency of carbon abatement costs of different mitigating measures (Eory et al., 2018). For instance, incentives for the production and use of H 2 could reduce its carbon abatement costs, but the existing tax relief on fossil fuel consumption prevents the adoption of low carbon fuels by the agricultural sector in Sweden.
In 2030, the carbon abatement costs are negative in most farm cases examined. Negative carbon abatement costs have been previously reported for different activities such as lighting switch, methane recovery from landfills, retrofit insulation in buildings, among others (McKinsey and Company, 2009). They owe negative values due to the advantage of having higher economic benefits than their implementation costs. In our case, this is translated by lower annual costs than diesel farming in most of 2030 cases. Such a favorable situation is not only due to the expected technological developments but also due to the 20% increase in diesel prices in the future. Thereby, emphasizing the importance of the price paid for diesel on the development of efficient climate protection strategies by policymakers.

Alternative Demand Profiles
As discussed in H 2 Demand and Optimization of H 2 Production, H 2 demand has a major impact on the optimal plant configuration and performance of the H 2 system. Where additional H 2 consumers could be integrated, resulting in alternative demand profiles, significant improvements in terms of production costs could be achieved. For instance, in case 2b (2030) the LCOH 2 of 11.03 €/kg is equivalent to a diesel price of 2.79 €/L, which is 2.4 times more expensive than currently found in retailers, including diesel tax relief. Such high production costs can be largely attributed to the low capacity factor of the PtH 2 plant. If the H 2 demand of FCMV in case 2b were to be multiplied by 10, i.e., 88.6 L of diesel eq. per day, a PtH 2 plant with an electrolyser size of 200 kW and 18 m 3 (500 bar) of capacity storage would be able to fulfill the demand of FCAM and FCMV, and at the same time operate during 4,241 h/year (48% of capacity factor). The PtH 2 plant, thus, could lower H 2 production costs to 7.04 €/kg in 2030, reaching a diesel equivalent price of 1.78 €/L (without by-products recovery). However, such a case is more akin to a small commercial filling station forecourt than a farm-based system and would require significant conversion of the local fossil fuel fleet to hydrogen fuel cells, or a medium-sized captive fleet.
Beyond sharing facilities across multiple farms as examined in this study, other farm types could be investigated for suitability for conversion to FCAM. In this case, ley crops for a dairy farm could show a more distributed H 2 demand throughout the year, reducing gas storage requirements as well as allowing the PtH 2 plant reach higher capacity factors. In addition, if these type of crops were integrated into a small pool of cereal-based farms sharing the same H 2 production infrastructure, the seasonality of fuel demand observed in the current study would certainly be reduced, potentially resulting in better economic performances. Ultimately, strategies similar to a demand-side management approach could be applied even to farmers sharing the same PtH 2 plant with the same rotating crop system (e.g., present study). In this case, farmers could slightly adapt their agricultural operations to the availability of H 2 , in particular during fall for plowing operations. Such strategy is considered a key aspect to improve the economic performance and it should be addressed in future studies on small-scale green H 2 production for agricultural applications.
Alternatively, H 2 surplus to FCAM demand could also be injected into agricultural biogas plants in a so-called in-situ biomethanation concept (Voelklein et al., 2019). Such synergies with agricultural biogas plants could be explored in different ways: 1) to increase biomethane output of biogas plants by reacting H 2 with CO 2 ; and/or 2) to use H 2 to partly displace costly energy crops as substrate like maize silage while keeping the same energy output of the biogas plant. Both concepts would increase the capacity factor of electrolysers and potentially decrease the costs of biogas production. However, care must be taken to ensure that the value of the methane-based H 2 and the economies of scale it allows for are greater than the sum of the additional costs.
Small-scale Haber-Bosch process (minimum of 1.5 MW) for ammonia fertilizer production could also be explored to provide an alternative demand for H 2 in the agricultural sector (Proton Ventures, 2018). This could quickly become the main demand for H 2 and would provide the required economies of scale to result in a more competitive H 2 either as fuel or platform for PtX processes.

CONCLUSIONS
This study examined the potential costs of an optimized system designed predominately to replace diesel-powered agricultural machinery with that powered by hydrogen (H 2 ) fuel cells. Several scenarios or cases were examined which included the addition of fuel cell light-duty vans, the sharing of H 2 facilities across neighboring farms, and valorization of the by-products (oxygen and waste heat). Results are presented in terms of levelized cost of hydrogen (LCOH 2 ), equivalent annual cost (EAC) to the farmer (consumer), and carbon abatement cost.
Even though sharing the same H 2 facility among four farms decreased the LCOH 2 by 28% and by adding fuel demand for delivery vans further decreased production costs by 35%, given the current cost of diesel and associated carbon taxes, H 2 is not competitive in 2020. However, anticipated reductions in H 2 costs coupled with increases in diesel prices mean that by 2030 H 2 fuel cells may represent an economic option in many cases. Therefore, the carbon abatement costs varied drastically from −145 €/tCO 2 when H 2 becomes competitive with diesel in 2030, up to 646 €/ tCO 2 in 2020. Nevertheless, when a PtH 2 plant is financed by the land lease revenues from a wind farm, H 2 becomes more competitive than diesel in all analyzed scenarios. Managing the demand profiles to decrease H 2 storage requirements and/or introducing an additional demand like for ammonia fertilizer production are effective strategies to reduce costs and should be addressed in future studies on H 2 production for low carbon agriculture.

DATA AVAILABILITY STATEMENT
All relevant data is contained within the article: The original contributions presented in the study are included in the article/ supplementary material, further inquiries can be directed to the corresponding author.