Carbon-neutral power system enabled e-kerosene production in Brazil in 2050

Rich in renewable resources, extensive acreage, and bioenergy expertise, Brazil, however, has no established strategies for sustainable aviation fuels, particularly e-kerosene. We extend the lens from the often-studied economic feasibility of individual e-kerosene supply chains to a system-wide perspective. Employing energy system analyses, we examine the integration of e-kerosene production into Brazil’s national energy supplies. We introduce PyPSA-Brazil, an open-source energy system optimisation model grounded in public data. This model integrates e-kerosene production and offers granular spatial resolution, enabling federal-level informed decisions on infrastructure locations and enhancing transparency in Brazilian energy supply scenarios. Our findings indicate that incorporating e-kerosene production can bolster system efficiency as Brazil targets a carbon-neutral electricity supply by 2050. The share of e-kerosene in meeting kerosene demand fluctuates between 2.7 and 51.1%, with production costs varying from 113.3 to 227.3 €/MWh. These costs are influenced by factors such as biokerosene costs, carbon pricing, and export aspirations. Our findings are relevant for Brazilian policymakers championing aviation sustainability and offer a framework for other countries envisioning carbon-neutral e-kerosene production and export.

Table S2.Five preselected open-source modelling frameworks based on the 75 state-of-the-art models S3, a  a Even though FINE is not featured in Ringkjøb et al.S3 , we collect the information regarding it.Some of this information is updated based on the official documentation of the modelling framework.b Calliope is a model framework published by ETH Zürich (https://www.callio.pe/).c "UD" refers to User-defined.Developers can use the framework to build energy systems with multiple regions (also known as nodes) and time steps at their request.d "All sectors" indicates the common practice in the energy system model of combining demand/load based on power consumption across all sectors.e MILP is under development.f The Open Energy Modelling Framework, or Oemof, is an important element of this research published by Reiner Lemoine Institute/ZNES (https://oemof.readthedocs.io/en/latest/).g PyPSA stands for Python for Power System Analysis, published by FIAS S15 ( https://pypsa.org/).h OseMOSYS stands for Open Source Energy Modelling System, published by KTH Royal Institute of Technology (http://www.osemosys.org/).i While only "Electricity" is recognised in [S3, Table 3], the OseMOSYS online documentation indicates its application in the "Water-Food Nexus" project, suggesting the possibility of incorporating multiple commodities within OseMOSYS.j GNU MathProg is a mathematical programming language for describing linear mathematical programming models.In addition, the OseMOSYS model framework is implemented using the languages like GAMS and Pyomo.k The Framework for Integrated Energy System Assessment, or FINE, is published by Forschungszentrum Jülich GmbH (https: //vsa-fine.readthedocs.io/en/latest/index.html).

A.3.2 Framework comparison
The preselected frameworks are further compared based on model logic, techno-economic details, ease of use, popularity, and added value.
Model logic evaluations focus on the purpose and language of the frameworks that are compatible with our objective.Except for OseMOSYS, the other four frameworks meet the requirements for cost optimisation for operation and expansion (cf."purpose" under "Model logic" in Table S2).
Modelling frameworks can be scripted in mathematical programming languages (e.g., GAMS, GNU MathProg, and Pyomo) or general-purpose programming languages (e.g., Python, Julia).While mathematical programming languages offer a closer resemblance to the mathematical model, general-purpose languages are more accessible to non-programmers S13 .The latter also facilitates data processing and analysis through packages such as Pandas, Numpy, and Matplotlib.For this reason, in the "language" column under "Model logic" in Table S2, a framework based on a general-purpose language is considered more beneficial.
Techno-economic details are essential for the application of modelling frameworks to create energy system models.In modelling the impact of increasing the share of vRES, large-scale production of e-kerosene, and alternate kerosene supply options, key attributes to consider include grid development, energy storage, and demand-side management S3 .The modeller connects general functions (often called components) by specification and adds user-defined mathematical constraints where necessary.These attributes are the technical and economic parameters in the model components, including conventional or renewable generation technologies, energy storage, emissions, cost, and grid, which determine the level of technical-economic detail.A framework allowing for varying levels of techno-economic details is preferred for flexibility in balancing computational 3/29 resources and the consequences of errors due to low resolution of techno-economic detail S1, S2 .While most of the available model components are similar across the five frameworks, it is the grid modelling that is noteworthy (cf."Techno-economic details" in Table S3).OseMOSYS does not include grid modelling, Calliope and Oemof (SOLPH) use the Net Transfer Capacity (NTC) approach, while FINE builds on this supports linear power flows, as do power system analysis tools, and PyPSA further allows nonlinear power flows.Thus, PyPSA and FINE are preferred for their relatively great technical-economic details compared to other frameworks.
"Ease of use" is gauged by the availability of comprehensive documents and tutorials that enable entry-level modellers to easily understand the functionality and usefulness of the framework effectively S13 .From this perspective, the quality and understanding of the online documentation, formulations, and tutorials are characterised as high, medium, or low based on our practical experience (summarised in Table S3 under the "Easy of use").
Up next, the popularity of the framework indicates its sustainability and potential for widespread adoption (cf."Popularity" in Table S3).The popularity is measured by a combination of factors, including the number of projects, publications, and the presence of an open community.Building a model on a popular framework can enhance its maintenance and likelihood of adoption by other researchers.Choosing a framework with an open community is also beneficial for modellers as it allows for peer-to-peer support and direct discourse with the core developers of the framework S13 .Evaluating the number of publications, along with the geographic distribution of the most prolific contributors, aids in assessing the international dissemination of the framework.
"Added-value" in Table S3 refers to additional algorithms or embedded libraries within the framework, which relieve modellers of the burden of creating specialised functionalities, such as space and time aggregation.Apart from OseMOSYS, other frameworks provide added value to their frameworks by making extra efforts on additional algorithms or packages.The gathered information relies on examining framework documentation, relevant publications, and web sources, as outlined in the associated footnotes.

4/29
b The provided details are sourced from the S3 and FINE's documentation.While there is more additional beyond what is presented in Table S2, one can refer to S3 .
The evaluation is conducted by examining the project's source code and documentation. d The data collection phase lasted until October 2021.
e Algorithms available to modellers, aiding in workload reduction, are taken into consideration. f The value presented reflects the number of open projects available. g The number is identified through the citation of publications from the Scopus database, which are recommended in the online documentation. h The "all" indicates that power generation technologies, whether conventional or renewable, can be modelled using the provided general functionality. i The "all" signifies that all energy storage technologies, such as batteries, hydrogen, and thermal energy storage, can be modelled with the general functions provided.j The NTC is a popular approach used in energy system models due to its simplicity and high accuracy.k Calliope's clear explanation of its general functionality is particularly useful for developers in technology modelling.
l Power flow modelling proves more capable than the NTC approach in grid modelling, as it adheres closely to grid principles.Offering both power flow and NTC approaches is seen as a model selection advantage, providing modellers with more options S15 .m The web documentation indicates the reference to more than five countries.
n Of note is the extensive use of OseMOSYS by a great academic community for teaching purposes.

A.4 Selection outcome
Depending on the information collected and shown in Table S2 and Table S3, each framework is scored according to its level of satisfaction with the underlying research criteria.Consequently, PyPSA S15 emerges as the chosen framework due to its highest score (cf.Table S4).
Table S4.Results of the model frame selection.The orange cell is the preselection step (cf.Table S2), the green cell is the second step of comparison (cf.Table S3), and the obtained scores are listed in the "Total" column.The rating is given according to the level of satisfaction of the criteria, +2 for satisfied, +1 for partially, and 0 for rarely.As shown in the blue cells, PyPSA is the selected model framework.

Open source model
High resolution

Model logic
Techno-economic details

Subjective perception
Popularity Addedvalue Total

B.1 Optimisation problem
The objective is to minimise the annual system costs.This includes the annualised capital costs -attributed to the expansion of generation capacities, transmission capacities, storage capacities, and energy conversion capacities -and the variable costs associated with the dispatch of generation, storage, and energy conversion.Equation (1) presents the mathematical expression for this objective: where c * represents the capital costs, while o * signifies variable costs.The indices r, s, l, and k label the generation technologies, storage technologies, transmission lines, and energy conversion technologies, respectively.The symbols G, F, E, and P correspond to the expanded capacity of generation, transmission, storage, and energy conversion technologies, respectively.Additionally, g, f , e, and p indicate the dispatch of the respective elements across various time snapshots, denoted as t and expressed in hours.The index n stands for each node within the system with n ∈ N , N = 1, 2, . . ., 27. Transmission lines serve as bus connectors, while specific nodes are equipped with converters that are capable of converting one energy carrier into another.Both transmission lines and converters are modelled using Link component, as illustrated in Main Text Figure 5.
The cost of the entire system is derived by summing the annualised capital expenditures, encompassing the annuity payment required and fixed operating expenses, in addition to variable costs, which include variable operating expenses and fuel costs.The equivalent annuity payment required is ascertained by employing the capital investment and the capital recovery factor, in alignment with the method demonstrated by Hörsch et al.S16 , which follows the formula stipulated in Short et al.S17 , taking into consideration the lifetime and discount rate of each technology.
where A : equivalent annuity payment I : capital investment i : annual discount rate N : lifetime of the technology in years.

6/29
Modelling the physical process entails adopting linear eligibility as a standard assumption.

B.2 Constraints
The system cost minimisation is subject to a set of applied constraints.

B.2.1 Energy balance
The fundamental constraint for optimisation is ensuring that the federal state's energy supply covers the local demand -power and kerosene -for every hour of the year.
where ) λ n,t : Karush-Kuhn-Tucker (KKT) multipliers associated with the equality constraints of the supply-demand balance.The value of λ n,t at the optimal point is an output of the optimisation.
The KKT multiplier represents the marginal price of the respective energy carrier at which the node n fulfils more demand at time t, also known as the local marginal price S18 .
The model's hourly resolution significantly lengthens the computation time needed, but it allows for a more accurate system description that considers the synergy effects of various system components or sectors.

B.2.2 Transmission
Between the nodes, the energy can be transferred from one to another through transmission lines.The maximum power flowing through the links is limited by the maximum physical capacity F l at any time: where the f l = −0.7 and f l = 0.7 denote an additional unit security margin for line capacity.These values, on the one hand, secure the approximate N-1 security and reserve capacity for lines, and on the other hand allow both import and export between the neighbouring nodes S16, S18 .
The physical capacity of the line F l is constrained (F l ⩽ F l ⩽ F l ) and can be expanded during the optimisation, depending on the cost-effectiveness.In this study, the lower bound F l is the nominal transfer capacity of lines obtained from Deng et al.S19 .The upper bound F l is set to infinite.However, the expansion of the transmission lines is limited by a global constraint: where Γ line volume is the sum of transfer capacities F l of the line l multiplied by their lengths l l (referred as line volume and measured in MWkm withing the model) for existing lines with a variable multiplier.Should Equation (5) be binding, the KKT multiplier µ line volume is expected to be positive, which represents the marginal value of the increase in line volume to the system.Otherwise, it indicates the cost per MWkm essential for the optimal solution to have the line volume Γ line volume S18 .For the analysis at hand, the multiplier is set to an infinite value, thus making Γ line volume infinite as well.
Linear optimal power flow is applied using Kirchhoff's formulation, which ignores the effect of impedance on the flows and solely requires the nodal power balance according to Kirchhoff's Current Law (KCL) S20 .

B.2.3 Generation
The dispatch of the generators for every hour g n,r,t is constrained by: where G n,r : optimum installed capacity of generators g n,r,t : lower bound of the availability.For all types of generators, this value is set to zero (g n,r,t = 0), which indicates that must-run operation is not required.g n,r,t : upper bound of the availability, [0, 1].For wind and solar energy, the value is time-and location-dependent.This value refers to its availability per unit capacity.The multiplication value (g n,r,t • G n,r ) indicates the maximum producible energy per hour, which can be derived from the reanalysis weather data.In the context of fossil-fuelled power plants, a default value of 1 signifies high flexibility, devoid of any ramp-up, ramp-down, start-up or shutdown costs.
The installed capacity for generators can be expanded and is limited by the following: where G n,r : lower bound of capacity expansion.For onshore wind, PV, and biomass thermal power plants, the value is set to be the installed capacity of the base year 2019 provided by Deng et al.S19 .G n,r : upper bound of capacity expansion.The value for wind and solar energy is the technical generation potential estimated by geometric and environmental constraints obtained from Deng et al.S19 .For biomass thermal power plants, it is the sum of existing and planned capacity for the base year 2019 and the economic potentials presented in S19 .
The PyPSA-Brazil model does not permit the expansion of fossil power plants, and the calculated capital cost is set to zero.However, fossil and nuclear power plants are excluded from our scenario analysis.

B.2.4 Storage
The state-of-charge of the storage should equal the dispatch at each hour: where soc n,s,t : state-of-charge of every storage s : storage technology, s ∈ {battery, reservoir hydropower plant, e-kerosene tank} η s,0 : standing loss per hour to the state-of-charge.For battery, η s,0 is constant, while for reservoir hydropower plant and e-kerosene tank, no standing losses are assumed (η s,0 = 0).e inflow n,s,t : natural inflow to the storage.Only reservoir hydropower plant has time-dependent inflow for each node (cf.S19 ).determine losses and signify that storage is only charged during periods of excess system power supply and depleted during periods of insufficient power production by generators and import options S18 .
In PyPSA-Brazil, the storage energy capacity, represented by h max s • E n,s , is optimised depending on the storage power capacity E n,s .The Energy to Power (E2P) ratio, denoted as h max s , is the fixed duration during which the stored energy can be fully charged or discharged at maximum power S21 .
The power capacity of the storage E n,s can be expanded but should be within the upper and lower limits: where E n,s : lower bound.For battery and e-kerosene tanks, E n,s = 0 is set.For hydro reservoirs, however, E n,s signifies the installed capacity of all sizes of hydropower plants in the base year, derived from Deng et al.S19 .E n,s : upper bound.E n,s = ∞ is designated for battery and e-kerosene tanks, while for reservoir hydropower, E n,s represents the sum of the installed and planned capacity from the Brazilian Ten-Year Energy Plan S22 (cf.Deng et al.S19 ).
Due to the annual periodicity of demand and seasonal generation patterns, it makes sense to assume cyclic states of charge when optimising a full year S18 .In this way, storage can be used efficiently at the beginning of the modelled time horizon and avoid the depletion in the end, soc n,s,t=0 = soc n,s,t=T ∀ n, s.

B.2.5 E-kerosene generation
The modelling of e-kerosene production employs the Link component in PyPSA framework.It is assumed that the capacity expansion of the e-kerosene production unit relies exclusively on a cost basis, setting 0 The dispatch of e-kerosene generation is not only constrained by its rated capacity but may also be limited by conditions under which it must operate and the availability: where k : energy conversion technology, k ∈ {e-kerosene production unit} P n,k : capacity of energy conversion p n,k,t : dispatch of energy conversion p n,k,t : must-run factor.It is set to 0, p n,k,t = 0, implying that the conversion from electricity to e-kerosene is a unidirectional process, without any mandatory operational levels.p n,k,t : availability factor.It is set as p n,k,t = η k , where η k denotes the conversion efficiency of the e-kerosene production unit.

B.2.6 Supply of biokerosene and conventional kerosene
In the model, the supply of biokerosene and conventional kerosene is represented using the Generator component in PyPSA framework.It is assumed that there are no capacity limits, hence, setting 0 ⩽ G n,r < ∞ (cf.Equation ( 7)).This implies that the supply amount per hour depends entirely on the marginal cost of providing biokerosene and conventional kerosene, which is measured in e/MWh.

B.2.7 Additional constraints
The constraints in Equations ( 3) to (11) primarily represent technical restrictions.However, to ensure that the optimisation problem produces feasible solutions, additional constraints can be implemented.One such constraint involves limiting the total CO 2 emissions ensuring that they do not exceed a specified budget, denoted as where 9/29 r : technology, r ∈ {fossil generators, conventional kerosene supply}.η n,r : generator efficiency at node n for technology r g n,r,t : generator dispatch at time t ρ r : fuel-specific emissions, measured in units of CO 2 t/MWh.This emission rate is assumed to apply only to the supply of conventional kerosene.Γ CO 2 : predetermined budget of CO 2 emissions µ CO 2 : KKT multiplier, also referred to as the shadow price.It indicates the marginal cost of emitting an additional tonne of CO 2 .It can also be interpreted as the additional cost required to achieve the CO 2 emissions reduction target.
In our scenario analysis, there is an option to limit the contribution of e-kerosene to a specific fraction of the total kerosene demand.This results in a constraint where the combined supply of biokerosene and conventional kerosene must exceed a certain proportion (γ) of the total kerosene demand, d kerosene where r ∈ {conventional kerosene supply, biokerosene supply}.For instance, in the "100% e-kerosene supply" scenario, γ = 0 is set.

B.3 Assumptions about e-kerosene production route
The e-kerosene production chain has a wide choice of technologies that correspond to each processing step, namely, the provision of H 2 and CO 2 , the synthesis and upgrading S23 .This section outline herein the assumptions and simplifications employed in the modelling of e-kerosene production within the PyPSA-Brazil model, which we refer to as the "e-kerosene production link" in Main Text Figure 5.
We utilise a simplified representation of the e-kerosene production plant, as opposed to the detailed component modelling delineated in Sherwin S24 .The model relies on conversion efficiencies and techno-economic parameters derived from Schmidt et al.S25 to emulate the technical behaviour.The plant features a conversion efficiency from electricity to e-kerosene of 0.42 and includes a plant configurations of low-temperature electrolysis -Alkaline Electrolysis (AEL) or Proton Exchange Membrane Electrolysis (PEM) -hydrogen storage, CO 2 sourced from Direct Air Capture (DAC), and a Fischer-Tropsch (FT) pathway comprising FT synthesis, Reverse Water-gas Shift (rWGS), and hydrocracking/isomerisation.
In the provision of CO 2 , fossil CO 2 using carbon capture and utilisation is elected not to used, as it falls short as a long-term, carbon-neutral solution S26 .Additionally, while CO 2 could be obtained from concentrated sustainable sources such as biomass combustion, organic residues, and bio-ethanol production from sustainably produced sugars or starches S27 , these options call for access to extensive biomass data.As such, we assume that CO 2 is sourced exclusively from the atmosphere via DAC and is universally available in Brazil.
The assumption is made that high-purity water needed for FT-based e-kerosene production is readily available in Brazil and may not pose a technological challenge in Brazil.This is due to the relatively lower water demand compared to biokerosene production by Hydro-processed Esters and Fatty Acids (HEFA) process or Alcohol-to-jet (ATJ) S23 .The ready availability of both groundwater S28 and surface water S29 , coupled with Brazil's status as one of the world's leading nations in terms of seawater availability S30 and the successful water scarcity management S31 , implies that the pre-treatment process of water or the cost thereof is deemed negligible.
The focus of our research is the FT synthesis pathway, a prevalent process in large-scale industrial applications for producing liquids from natural gas or coal S32 .Although methanol synthesis is capable of yielding e-kerosene, so far, the first-of-its-kind aviation fuel specification -namely, the American Society for Testing and Materials (ASTM) D7566 standard -exclusively specifies this synthetic kerosene derived from the FT process S33 .A noteworthy assumption in our study is the potential for altering future blending ratios of e-kerosene with crude oil-based kerosene S32, S34 , speculating that they may no longer be limited to the 50%, currently prescribed by ASTM regulations.
We explicitly model the energy system with generation, storage, transmission and demand for electricity supply for e-kerosene production using PyPSA-Brazil model.By considering the dynamics of the system, the cost of electricity supply is calculated endogenously at the federal-state level in Brazil.
The PyPSA-Brazil model posits that the e-kerosene production results in an Emissions Reduction Factor (ERF) of 100%.ERF measure the net life-cycle CO 2 benefits of Sustainable Aviation Fuel (SAF) by accounting for the CO 2 savings from feedstock production or growth, and incorporating the emissions incurred during fuel production S35 .100% ERF indicates that there is no net carbon loss between the stages of emissions and capture, which is also referenced by Micheli et al.S36 .However, Micheli et al.S36 adopt a more conservative stance, assuming 5% loss within the closed carbon cycle.Such a loss translates to emissions ranging from 0.9-4.0g CO 2 e/MJ, on the condition that electricity is sourced from wind or solar energy.In addition, 10/29 the Air Transport Action Group explores sustainable trajectories for aviation by assuming ERF of 70-100% for SAF S35 .This is indicative of the industry's direction towards mitigating carbon emissions.
It is imperative to underscore that focusing solely on achieving a closed carbon cycle with regard to CO 2 emissions does not address the broader environmental impacts.Specifically, aircraft emissions in the upper atmosphere, including water vapour, aerosols, and nitrogen oxides (NOx), have a significantly more detrimental effect on the climate compared to CO 2 emissions at lower altitudes S37, S38 .These non-CO 2 impacts, often overlooked, merit greater consideration.

C Data C.1 Power sector
The inputs and the assumptions underlying the model of the Brazilian power system are well documented in Deng et al.S19 , where the open, spatially resolved, harmonised data set of the Brazilian energy system is introduced.PyPSA-Brazil adopts 2019 as the base year in this study, ensuring that all input data are from that year.

Stored capacity for reservoir hydropower plant
We assume that the hydropower plants are of reservoir type, with their energy capacity determined on the E2P ratio and power capacity (cf.Equation ( 9) and Deng et al.S19 ).We derive the E2P ratio from the data set of stored energy capacity for each electric region in the National Interconnected Network (Portuguese: Sistema Interligado Nacional, SIN), released by National Electricity System Operator (Portuguese: Operador Nacional do Sistema Elétrico, ONS) S40 .The stored capacity of hydropower is about 357,548,409 MWh.This figure, accumulated daily across the four electric regions, is presented in MWmês.For clarity, we apply the conversion: 1 MWmês = 720 MWh/month.This regional storage energy capacity then undergoes conversion to the federal-state level: where  S19 ) S R,s : stored energy capacity at the region R, in unit of MWh.

C.2 Cost assumptions
The financial and technical assumptions used in this study, such as investment costs, Variable Operation and Maintenance (VOM) costs, Fixed Operation and Maintenance (FOM) costs, efficiency, and lifetime, are based on insights from the scientific literature and are presented in the Table S5.The capital investment of each technology is annualised with a discount rate of 8% over the economic lifetime used in Equation ( 2).
Capital investment in transmission lines comes from historical auctions with units of R$/MW/km S41 .The capital investment costs are derived by interpolating representative line lengths for Alternating Current (AC) and High Voltage Direct Current (HVDC) lines, considering the cost of converters and transformers for each historical auction. 2 The value represents the lower limit of aviation fuel density, ranging from 775-840 kg/m 3 . 3The 2019 annual average is used. 4The value comes from "sheet 22 Photovoltaics medium". 5The value results from dividing "Fixed O&M" by "Specific investment, total system" in the raw data. 6A value of 0.01 adjusts the curtailment order of renewable technology S16 . 7 The value originates from "sheet 21 Large wind turbines offshore". 8This is a ratio of "Fixed O&M" to "Variable O&M" derived from the original data.The model assumes flights departing from Brazilian airports are refuelled within Brazil, thereby constituting the kerosene demand for each airport.As a result, four attributes of the original air transport data set (as detailed in Table S6) are used.By accounting for the location of the refuelling airport, the kerosene demand is aggregated by each federal state.The model presumes that the departure time of the aircraft corresponds to the refuelling date, which yields a daily kerosene demand pattern.This daily time series is then averaged out to derive an hourly demand profile (presented in Figure S3).In 2019, the kerosene demand is 4,286.6ML (approximately 39.5 TWh), with the highest demand found in the federal state of SP(39.9%),followed by RJ and DF (7.9% each), PE (5.2%), BA (4.8%), CE (4.5%), MG (4.4%), RS and PR (3.2% each).Historical kerosene demand is available for 2000-2020, but the model only employs the time series for the base year 2019 (cf. Figure S4).

15/29
Figure S3.2019 kerosene demand by federal states.The left y-axis is in units of the original data, and the right axis is in units converted to GWh according to Table S5.S5.
The ANAC releases official annual demand forecasts for domestic and international flights up to 2050, distinguished by two scenarios -"without mitigation" and "with mitigation" S61 .The projection for the period 2019-2050 on total annual national kerosene demand is determined using monthly Revenue Tonne Kilometer (RTK) and kerosene consumption statistics from 2010-2018, as well as the anticipated progress in fuel efficiency S62 .In the "with mitigation" scenario, Brazil's total kerosene consumption is expected to reach 13.2 million tons in 2050 S61 , representing an overall growth of 217.5% and an average annual growth of 3.7% from 2018 to 2050.This projection of kerosene consumption of 13.2 million tons (around 157 TWh) for 2050 is integrated into the PyPSA-Brazil model, which is approximately four times higher than the consumption level in the base year of 2019.

C.4.2 Kerosene supply
According to Figure 5 of the Main Text, the aviation sector consumes kerosene that is synthetically produced with e-kerosene, fossil origin, or biokerosene.

Supply of conventional kerosene
The supply of conventional kerosene depends purely on the cost of supply.
The National Agency for Petroleum, Natural Gas and Biofuels (Portuguese: Agência Nacional do Petróleo, Gás Natural e Biocombustíveis, ANP) regularly updates the fuel distribution prices at a national, regional, and federal state level S63 .The state-level data are incorporated into PyPSA-Brazil, using regional data when values are absent for certain states such as SE and AP.It is assumed that the conventional kerosene price (R$/L) in 2019 is the marginal production cost in each state in 2050.An average hourly price is then determined from the monthly prices (cf. Figure S5) since PyPSA-Brazil considers a 16/29 time-dependent marginal cost.With the carbon emissions from conventional kerosene generation, a carbon price is introduced for Brazil to achieve carbon neutrality (cf.Section C.6).

Figure S5.
Box plot of monthly kerosene distribution price across federal states in 2019: input data of PyPSA-Brazil model processed from original source S63 , with the raw and converted units of e/MWh (cf.Table S5).
Supply of biokerosene PyPSA-Brazil model posits that the biokerosene potential for 2050 is measured by its production cost, as appraised by Cervi et al.S64 .This assessment employs techno-economic assessment and spatial data analysis to examine the potential of 13 distinct production routes in 2030 at a spatial resolution of 5 km (cf. Figure S6).These routes comprise eight first-generation biomass -maize, sugarcane, sorghum, eucalyptus, soybean, sunflower, palm, and macaw -and five conversion pathways, including ATJ, HEFA, FT, Direct Fermentation of Sugars to Hydrocarbons, and Hydrotreated Depolymerized Cellulosic Jet.Cervi et al.S64 assume that the land used for biokerosene production is leftover -it is not utilised for other purposes, such as forest plantations and urbanisation, and it is not set aside for conservation areas.
Cervi et al.S64 provide data in $/t, which is converted to e/MWh for integration into PyPSA-Brazil.The resolution provided by Cervi et al.S64 for the cost of biokerosene production is 5 km, whereas PyPSA-Brazil utilises a coarser spatial resolution corresponding to federal states.In PyPSA-Brazil, the production cost of biokerosene for each federal state is determined by calculating 10%, 25%, 50% percentile of the cost within 5 km × 5 km cell, as depicted in Figure S7.For scenario analysis, the input to the model indicates that the production cost at the federal-state level ranges from 69.5-149.2e/MWh at a low level, 104.2-234.6 e/MWh at a medium level, and 147.8-725.6 e/MWh at a high level.
As the biomass used in Cervi et al.S64 (maize, sugarcane, sorghum, eucalyptus, soybean, sunflower, palm and macaw) is of the biogenic type, the life-cycle GHG emissions of biokerosene are set to be carbon neutral S65 , signifying its environmental sustainable character.

C.5 Carbon emission cap
In PyPSA-Brazil, total carbon emissions across sectors are limited so as not to exceed a predetermined budget, represented by Γ CO 2 as shown in Equation ( 12).In compliance with Brazil's Intended Nationally Determined Contributions, the nation is committed to cutting GHG emissions to 50% below the levels recorded in 2005 by the year 2030, and further attaining climate neutrality by 2050 [S66, p. 1].This commitment translates into an average annual reduction rate of 2%.The scarcity of sector-specific statistics and emission budgets, however, necessitates certain assumptions for modelling purposes.
As a reference, in 2006, the power sector was responsible for the emissions of 26.42 Mt CO 2 e S67 .Since data for 2005 is not available, we assume a consistent 2% reduction rate from 2006 to 2050.This assumption yields an 88% reduction by 2050 based on 2006 emissions, setting a budget of 3.17 Mt CO 2 e.This budget represents GHG emissions, and in the context of PyPSA-Brazil, it is regarded as the carbon emission budget for the power sector.
Regarding the aviation sector, the ANAC has estimated the carbon emissions for the period 2019-2050, drawing upon the projected kerosene consumption for domestic and international flights.This calculation, underpinned by the assumption of kerosene demand being exclusively met by conventional kerosene, applies a carbon emission factor of 3.16 kg CO 2 /kg Jet [S61,  p. 32].Consistent with the global mitigation measure to secure carbon-neutral growth in aviation from 2020, and net-zero carbon emissions by 2050 [S68, p. 1-2], PyPSA-Brazil model postulates that Brazil adopts an emissions budget for 2050 equivalent to the projection for 2020 S61 , amounting to 14.2 million tonnes.This budget of 14,221,089.4 tonnes pertains to carbon emissions.Consequently, the model permits up to 34% of the demand to be satisfied by conventional kerosene, assuming a total kerosene consumption of 13.2 million tonnes in Brazil by 2050.

C.6 Carbon pricing range
Carbon pricing is an instrumental policy tool that places a charge on GHG emissions, thereby encouraging governments and industries to modify their production, consumption, and investment habits in favour of low-carbon growth S69 .As of 2022, Brazil has been evaluating the introduction of carbon pricing, though it has not yet been established [S69, p. 12, 55].Notably, the current carbon pricing in effect regulates only the CO 2 emissions from aviation S70 .For instance, the EU's Emission Trading System (EU ETS) focuses on CO 2 emissions because the scientific understanding of non-CO2 effects (such as NOx, water vapour, soot, sulphates, and contrails) is not yet sufficiently mature to formulate comprehensive policies S70, S71 .For this reason, PyPSA-Brazil limits the application of carbon budgets and carbon pricing to CO 2 emissions.
In our scenario analysis, carbon pricing is primarily applied to the combustion of jet fuel, which leads to CO 2 emissions.The analysis begins with the implementation of a carbon emissions budget (cf.Section C.5), and in PyPSA-Brazil, this corresponds to an initial carbon price of 160 e/t.
The analysis then explores a range of carbon prices for the year 2050, extending from 160 e/t to 1000 e/t.This range includes specific price points such as 200 e/t, 260 e/t, 320 e/t, and 500 e/t.This range is chosen to account for potential extreme shifts in carbon pricing, possibly due to drastic policy changes or unforeseen global events impacting the carbon market in 2050 S72 .The assumed range of carbon prices is in line with the IPCC's mitigation pathway of limiting the median warming to below 1.5 °C by 2100.The selected carbon prices fall within the range specified by two IPCC scenarios: the "low overshoot pathway", which holds a 50-67% likelihood of a temporary early overshoot of 1.5 °C, and the high overshoot pathway, having an over 67% probability of an early temporary overshoot S72 .

D Equations for analysis D.1 Average system cost
In our study, the Average System Cost (ASC) is employed as a metric that quantifies the average optimised total system cost per unit of energy generated within an energy system [S18, Fig. 2].Employing ASC as a metric enables the insightful comparison of the relative costs associated with energy supplies across various system designs through a given equation: where C * : optimised total system cost, e, computed as outlined in Equation (1) Γ * : optimised total dispatch, MWh.It includes the amount of energy produced by renewable generation technologies and the supply of kerosene in the system.

D.2 Levelised cost of energy
The Levelised Cost of Electricity (LCOE) assesses the costs associated with electricity generation from a single technology, with the costs being the net present sum of investment, fuel, operational and maintenance costs S17, S73 .We use LCOE to compare the generation cost for each technology among studies through the equation:

D.3 Levelised cost of fuel
The Levelised Cost of Fuel (LCOF) refers to the supply of e-kerosene, which is essential for determining its economic viability: : total electricity consumption for e-kerosene production at node n Γ e-kerosene n : total e-kerosene supply at node n.
: total system cost of the reference scenario d kerosene β ,δ : total kerosene demand of export scenario given β , δ , d kerosene ,kerosene : total kerosene demand in the reference scenario, i.e., 157 TWh.

D.5 Export amount of the e-kerosene
The amount of the e-kerosene for export is the absolute difference in the amount of e-kerosene supply between the export scenario and the reference scenario.

Figure S8
. Breakdown of annual total system costs in Brazil's carbon-neutral power system scenarios without ("only power system") and with e-kerosene production integration ("100% e-kerosene supply").

E.1.1 Annual total system costs
The PyPSA-Brazil model outcomes, displayed in Figure S8, indicate a rise in total system costs, which is driven by the additional provision of e-kerosene (increasing from 59.1 to 76.3 billion e/ year, by 29%).The main contributor to this increase is the further investment in PV installations, which provides 317 TWh.In contrast, hydropower remains relatively stable, while wind onshore (additionally 35 TWh) and biomass (additionally 38 TWh) energy see an uptick.E-kerosene production offers some load-balancing flexibility, leading to a slightly reduced investment in battery storage.

E.1.2 Annual electricity generation
Figure S9 illustrates the additional electricity generation required for producing e-kerosene in a carbon-neutral power system.Specifically, the results accentuate the substantial contribution from PV generation.This is complemented by 35 TWh from onshore wind and 39 TWh from biomass thermal energy.While hydropower experiences a slight rise, contributing to 2 TWh, it pales in comparison to the aforementioned technologies.Offshore generation also contributes but has a minimal impact in both scenarios.

E.2.1 Feasible e-kerosene production in Brazil from abundant renewable potential
As shown in Table S7, São Paulo emerges as the dominant federal state in terms of electricity and kerosene demand, given its stature as the country's largest economy and the most populous state.In contrast, Rio de Janeiro and Distrito Federal have relatively high population densities.São Paulo's low population density might contribute positively in lessening social acceptance issues linked to land use for PV expansion.Distrito Federal, characterised by the highest population density, predominantly relies on importing electricity from neighbouring states.Adjacent to Distrito Federal, Goiás displays moderate electricity demand and relatively low kerosene demand.The lower population density in Goiás positions it favourably for PV expansion and for acting as an electricity conduit or even a powerhouse to other federal states.Minas Gerais follows São Paulo in electricity and kerosene demand, and its low population density might result in fewer social acceptance issues regarding new PV installations.Lower population density can be a factor in diminishing local acceptance conflicts, albeit not a definitive solution S74, S75 .

Figure S9.
A comparison of annual electricity generation in Brazil's energy system under carbon-neutral scenarios without ("only power system") and with the integration of e-kerosene production ("100% e-kerosene supply").The economic practicability of e-kerosene supply, as portrayed in Main Text Table 3, might lead one to expect its extensive incorporation in the fuel mix.However, an unexpected observation emerges from Main Text Figure 2, which reveals that even amidst high carbon pricing and biokerosene cost, e-kerosene accounts for a maximum of 51.1% of the total supply in the scenarios studied.For clarification, Figure S10 illustrates the allocation of biokerosene and e-kerosene across federal states, considering high biokerosene costs and carbon prices.The data in Figure S10 indicate that, in the majority of federal states, e-kerosene assumes the larger portion of the fuel supply.São Paulo, however, deviates from this trend, exhibiting a marked inclination for biokerosene, thus magnifying its cumulative share in fulfilling the kerosene demand.
A deeper examination of costs in São Paulo sheds light on prominence of biokerosene.Within PyPSA-Brazil, the biokerosene production cost in São Paulo is assumed to be 184.8e/MWh, lower than the e-kerosene LCOF (about 215.4 e/MWh).An essential point of note is that the production cost of biokerosene is considered constant, whereas the cost of e-kerosene varies over time as calculated by PyPSA-Brazil.Consequently, e-kerosene's contribution is optimised during periods wherein its supply proves more economically efficient compared to biokerosene, which explains its modest share.São Paulo's reliance on biokerosene has an overall impact on the fuel landscape in Brazil.The economic barriers that e-kerosene faces in competing with biokerosene are illustrated by the entrenched dominance of biokerosene in São Paulo, the state with the highest kerosene demand.To strengthen the market position of e-kerosene, efforts should be made to reduce its production costs or policy reforms should be made to create an enabling environment for e-kerosene.

Figure S10.
Supply of kerosene at federal state given the carbon price of 1000 e/t and biokerosene production level at high.

E.2.3 Biokerosene's impact on Brazil's role as an exporter of carbon-neutral kerosene
Main Text Section Export costs of carbon-neutral kerosene from Brazil explores Brazil's potential to become an exporter of carbon-neutral kerosene.As exhibited in Main Text Figure 3, the contribution of e-kerosene appears relatively modest when biokerosene production costs are low to medium.In those scenarios, biokerosene remains the primary export due to its affordability and presumed carbon-neutrality in PyPSA-Brazil.However, as the costs of biokerosene production increase, ekerosene becomes more noticeable.A corresponding marked influence of biokerosene costs on the export costs and e-kerosene's export prominence is observable in Main Text Section Export costs of carbon-neutral kerosene from Brazil.
Assessing the likelihood of biokerosene production becoming high-cost within Brazil's geographic context is intricate due to various dynamic factors S79 .However, biokerosene production demands substantial water and land resources, which could contribute to cost augmentations in the future S44 .
Additionally, the environmental sustainability of biokerosene is called into question by the risk of indirect GHG emissions arising from land-use changes.For instance, the HEFA pathway using cooking oil as feedstock, despite being cost-effective S23 , may emit as much as 27-67.4g CO 2 /MJ S80, S81 .Although such levels of emissions are compliant with the sustainability criteria for SAF -which aim to reduce life-cycle GHG emissions by 10% compared to conventional kerosene S82 -they cast doubts on the complete climate neutrality of biokerosene.The imperative to restrain biomass cultivation to safeguard biodiversity S79 , mitigate indirect GHG emissions S69 , and ensure responsible land use S79 makes the presumption of limitless biokerosene production improbable.These factors could subsequently lead to a more conservative deployment of biokerosene at scale and, conversely, higher cost levels.
In light of these considerations, the ambition for Brazil to establish itself as an exporter of carbon-neutral kerosene may act as a catalyst for heightened investments in e-kerosene production.This would signify a progression from mere pursuit of a self-sufficiency in kerosene supply in 2050 to engaging proactively in global kerosene markets towards carbon-neutral aviation.The implications of such a transition could extend beyond economic considerations to involve broader sustainability objectives S83 .Harnessing the potential of e-kerosene could position Brazil as a leader in the transition towards carbon-neutral kerosene, while concurrently supporting its commitments to climate change mitigation.This strategic approach would enable Brazil to strike a balance between economic, environmental, and societal interests, contributing significantly to the global endeavour of creating a sustainable aviation sector S83 .

E.2.4 Comparison to related literature
Electricity supply cost in carbon-neutral scenarios This section evaluate carbon-neutral Brazilian energy systems from our study and contrasts them with alternative designs presented in existing literature, focusing on electricity supply cost, represented by ASC (cf.Equation ( 15)).
The results from the "only power system" scenario in our study indicate an ASC of 50.3 e/MWh.This is lower than the 70.6 e 2019 /MWh posited by Dranka and Ferreira S9 for a 100% renewable power system across four Brazilian regions in 2050.The 70.6 e/MWh is converted from the original 78.99 $/MWh using 2019 exchange rates as indicated in Main Text Table 2.The discrepancy in ASC between our study and Dranka and Ferreira S9 can be explained by differences in system designs.First, Dranka and Ferreira S9 assume a higher rate of electrification, resulting in an electricity demand of 1,571.5 TWh, while we estimate a more conservative demand of 1,164.8TWh (COPPE lowBECCS scenario presented in Main Text Table 2).Secondly, Dranka and Ferreira S9 suggest a total power generation capacity of 623 GW, much larger than the 523 GW identified in our study.Furthermore, their study estimates hydropower to contribute 50% of energy supply, approximately 800 TWh, which is almost twice as much as Brazil's hydropower production capacity in 2019.In comparison, our study suggests a smaller share of 32.8% for hydropower (386-388 TWh, akin to the levels in 2019, cf.Main Text Table 1) and a higher share of 35.5% for PV.The capital costs and lifetime assumptions for PV also differ.Dranka and Ferreira S9 lean less on solar power due to a permissive technology and economic parameterisation -their annualised capital costs of 187,357.6 e/MWh, nearly three times higher than the 48,931.5 e/MWh observed in our study.
Comparisons can also be made with Barbosa et al.S11 , which examines potential fully decarbonised Brazilian power systems with applications of renewable electricity for other end use.They report a decrease in ASC from 61 e/MWh to 53 e/MWh when adding 25% more electricity demand for water desalination and synthetic gas production.In a similar vein, we include e-kerosene production, which leads to a 32% increase in additional electricity demand under the "100% e-kerosene supply" scenario.Consequently, our ASC of the system dips from 50.3 e/MWh to 44.2 e/MWh.One of the key factors for the lower ASC in our study is the inclusion of interstate transmission in the PyPSA-Brazil performed, in contrast to the coarser spatial resolution in four regions adopted by Barbosa et al.S11 .Furthermore, while Barbosa et al.S11 include pumped and run-of-river hydropower plants and allows hydropower capacity to expand significantly -200% of the installed capacity, we assume reservoir types with limited expansion capabilities, in line with Brazil's National Ten-Year Energy Plan S19 .This assumption, although seemingly conservative, is based on Brazil's objective to diversify its energy generation portfolio, mitigating risks associated with dependency on hydropower, especially considering the environmental sensitivities in the Amazon region S6 .
In summary, our study unveils carbon-neutral power system designs for Brazil, which largely concur with those found in the existing literature.Notably, it employs the high-granularity PyPSA-Brazil model, integrating interstate transmission benefits even in the face of hydropower expansion constraints.Through this approach, a refined analysis is enabled, capitalising on Brazil's geographically diverse renewable resources.Consequently, our study presents a more cost-effective system than previously estimated in alternative literature, reducing the average system cost and affirming carbon neutrality as an attainable and economically viable goal for Brazil's power sector, whilst incorporating e-kerosene production.
Capacity factor of e-kerosene production unit Apart from securing a low electricity supply cost, a high capacity factor for the e-kerosene production unit is indispensable to ensure its economic feasibility S84 .Echoing this perspective, Agora Energiewende S85 present a comparison between scenarios with varying operation hours of electrolysers and synthesis, specifically 2,000 and 8,000 hours.In addition, Breyer et al.S34 highlight the considerable impact that the operation hours of FT synthesis have on economic viability as they assume baseload-like DAC operating for 8,000 hours.Our study aligns with these findings, demonstrating that the capacity factor for e-kerosene production fluctuates between 0.7 and 0.9 -a range that supports its economic feasibility.188.4 -416.4This range [S86, Figure 7] relies on the electricity source (renewable or grid) and the incorporation of emissions costs.Utilising renewable electricity, the LCOF is estimated at 2.3 e/kg without emission cost and increase to 4.95 e/kg when considering the emission costs.Grid-sourced electricity results in a smaller increment due to emissions costs (from 2.24 e/kg to 2.8 e/kg).The study utilises a MILP approach to optimise the total system cost of the e-kerosene supply chain at 36 nodes.a Unit conversion details are elaborated in Table S5.b The formulation details can be found in Equation ( 17).c The data on supply costs of electricity originate from the US.d Sherwin S24 state that grid power not necessary for green power generation.e The data are derived from the supplementary materials.

Levelised cost of fuel for e-kerosene supply
. A reservoir hydropower plant operates under the setting of η charge s = 0, while a constant value applies to both battery storage and the e-kerosene tank.

Figure S7 .
Figure S7.Input assumptions of biokerosene production costs for the federal states used in PyPSA-Brazil.Each marker represents the input assumption -the percentile of biokerosene production costs for pixel points within the federal state.The lower the percentile, the lower the presumed production cost.The range of variation orders the values. .

Table S3 .
Comparison between the five preselected open-source modelling frameworks a .
⊆ reservoir hydropower plant n : node of the model -Brazilian federal state, n ∈ N , N = 1, 2, . . ., 27 E n,s : power capacity at federal state n, in unit of MW R : electric regions defined by SIN (cf.Deng et al.

Table S6 .
Attributes used in PyPSA-Brazil from the air transport statisticsS60.

Table S8 .
LCOF for e-kerosene in 2050 as reported in the literature.Scenarios excluding exports, considering variable biokerosene production costs (low, medium, and high) and carbon prices (160 e/t, 200 e/t, 260 e/t, 320 e/t, 500 e/t, 1000 e/t).The value, initially 2186 e/t, is based on electricity supply cost at 43 e/MWh [S84, Table13], employing the identical e-kerosene plant setup as the dissertation.The value ranges from 2.91 e/kg to 3.38 e/kg.Power-to-liquid process with lowtemperature electrolysis, CO 2 from ambient air and renewable electricity cost varying from 35 e/MWh to 55 e/MWh[S32, Table 7].This cost is the cost of electricity used to meet a specific e-kerosene demand, driven by the specific process, process efficiency and carrier requirements (e.g.CO 2 ).This range, originally 2.24 e/kg to 2.7 e/kg)[S32, Table 7].Similar to the previous entry but using high-temperature electrolysis.
S84Germany186.9 S86 Spain Based on FT synthesis with DAC and electricity supply cost at 35 e/MWh, the original value is 1762 e/t [S84, Table 13].Similar to Spain but with electricity supply cost at 32 e/MWh.The original value was 1735 e/t [S84, Table 13].The range, 75 e/MWh to 137.5 e/MWh [S85, Figure 21, 2050, 100%], indicates the site of liquid synthetic fuel production and import to Germany, which includes transport costs.Note that e-kerosene is not explicitly mentioned as the final product.27 75 In this study, FT-derived kerosene is produced using DAC with electricity sourced from hourly PV and wind energy generation.Electricity costs are calculated without grid expenses.The valued is from [S34, Table 11].The assumptions mirror the EU-27 scenario, but this scenario is conducted for the US, which has a different renewable energy generation potential.The raw value is converted from 0.99 $/L_gasoline equivalent to $/MWh using density and heat value of gasoline of 0.755 kg/L and 13 kWh/kg) from [S24, page F].The study employs grid electricity at 65 $/MWh d , wind supply at 19 $/MWh, and solar supply at 11.8 $/MWh e .The range (2-4 e/L) in the study represents a scenario in 2050 where CO 2 from DAC and H 2 from water electrolysis are combined to produce synthetic fuels using renewable electricity, supplied based on LCOE at 61 e/MWh [S87, Figure 11, 2050 DAC-CCU scenario].