The total system costs (C) are the sum of all costs across every time period. As expressed, the total system costs in each time period are divided into the five different costs associated at each phase (Feedstock-FS, Collection-Coll, Pretreatment-Pre, Conversion-Conv and Transport-Trans) in the supply chain plus the deficit cost (Deficit). Each cost is dependent on at least one of the four decision variables. C T = ∑ t ∈ T C t F S + C t C o l l + C t P r e + C t C o n v + C t T r a n s + C t D e f i c i t (1)

Starting from the purchasing of the feedstock to the conversion and transportation. C t F S = ∑ s ∈ S ∑ b ∈ B u b c s , b , t ⋅ B U s , b , t ∀ t ∈ T (2)

Total feedstock costs are measured as the sum of the total amount (UB) of each type of biomass (b) collected at each country (s) multiplied by the respective unit biomass price (ubc) (per biomass type) at each country in each time period (t). C t C o l l = ∑ s ∈ S ∑ b ∈ B B U b , s , t ⋅ w s , t + r c s , t + c t f s , t ⋅ f c l o a d e r ⋅ 1 0 3 ρ b w e t ⋅ L H V b w e t ⋅ β s l a b o r ⋅ r ∀ t ∈ T (3)

The collection costs can be summarized as the product of the total costs per hour (machine rental (rc), wages (w) and fuel cost (ct) times fuel consumption (fc)) multiplied by the total time needed to load all of the biomass (total biomass energy content divided by lower heating value (LHV), density (ρ) and loading rate in m³/h (r). This is all divided by the productivity of labor in each country (β). C t P r e = ∑ s ∈ S ∑ b ∈ B B U s , b , t ⋅ c e s , t ⋅ s e c b d r y i n g L H V b w e t + δ d r y i n g ⋅ s e c b m i l l i n g L H V b d r y ∀ t ∈ T (4)

Costs of pretreatment consist of two parts; drying and milling. The costs associated with these treatments are mainly those relating to electricity costs (ce) times the consumption of each proces (sec) multiplied by the mass (BU) and diveded by the LHV to convert to energy. Since material is lost during both drying and milling, two constants are added to reflect this; δ ^drying and δ ^milling . Due to the drying being performed before the milling, the amount of biomass that reaches the mills is reduced by the factor δ ^drying as can be seen on the right side of Eq. 4. As the costs are based on the mass input δ ^milling does not play a role here, it is only used to calculate the final mass at the end of both processes. This is the total efficiency of the pretreatment process which is the product of both pretreatment yields (δ ^drying and δ ^milling ), this is reflected in the constraints of the model. C t C o n v = ∑ i ∈ I ∑ r ∈ R ∑ p ∈ P ∑ f ∈ F u c c r , i , f , t ⋅ S F r , p , f , t β r R l a b o u r ∀ t ∈ T (5)

Conversion costs are calculated based on the average cost of producing a unit of the product. In other words, the cost (ucc in €/GJ) associated with producing one GJ of product from the given inputs (shipped amount to the refinery (SF)). Values are pulled from literature and averaged to approximate the operational costs of production. Again, the productivity of labor (β) is factored in to account for an additional or fewer amount of work depending on the country (or refinery) in which the process occurs.

Lastly, the transportation costs are described: C t T r a n s = ∑ s ∈ S ∑ b ∈ B B U s , b , t ⋅ d s O P r e ⋅ f c t r u c k ⋅ c t f s , t + w s , t T D / a s ⋅ 1 0 3 L H V b w e t ⋅ ρ b w e t ⋅ ω t r u c k + ∑ s ∈ S ∑ r ∈ R ∑ i ∈ I T I s , r , i , t ⋅ d s , r S R ⋅ f c t r u c k ⋅ c t f s , t + w s , t T D / a s ⋅ 1 0 3 L H V i d r y ⋅ ρ b d r y ⋅ ω t r u c k + ∑ r ∈ R ∑ p ∈ P ∑ f ∈ F S F r , p , f , t ⋅ c r ⋅ d r , p R P L H V f f u e l ⋅ a v ⋅ ω s h i p ∀ t ∈ T (6)

The transportation costs are divided into the three legs of transportation. In the first and second segment (the transport of biomass to the pretreatment facility and the transport of the intermediate to the refinery), the respective products (BU, TI) are carried by truck, for which the limiting carrying capacity is volume (ω). Therefore, the total energy content of the used biomass is transformed into total volume and with the carrying capacity and trip distance (d), the total amount of trips/number of trucks needed is calculated (using speed a). The costs of truck transport include the fuel costs as well as the truck driver wage, which is different in each country. The last leg of transport is calculated slightly differently. The costs are based on the daily chartering rate of tanker vessels. This includes operational costs such as crew, fuel and docking fees. In the case of a tanker vessel, the limiting factor in terms of capacity is the weight, therefore the deadweight tonnage is used in ω ^ship .

Additionally, another cost term is added to the equation called “Deficit Cost”. In order for the model to allocate fuel to the respective ports and meet the demand while also minimizing costs, it should receive a penalty for failing to do so. In optimization, these variables are called slack variables, and this one serves to ensure that the model is inclined towards coming as close as possible to the demand even when there is not enough supply. This variable is also present in the constraints, and without it, in cases where the demand is greater than the supply, the model would be unsolvable and be rendered infeasible. C t D e f i c i t = D T p , t ⋅ c p t (7)

The deficit cost consists of the decision variable DT _p,t which is the total amount of energy deficit at each port and time period. The cost associated with the energy deficit, cp _t , is the cost of replacing that energy with another clean fuel such as biodiesel or LNG. The value of cp _t is not fixed and will need to be altered for each demand case to a level where the model reaches full demand satisfaction. It is also important to note that although C t D e f i c i t is included in the total cost function for the optimization, it is not considered in the post-processing values of the cost.

The environmental impact of the supply chain is assessed on the basis of total greenhouse emissions released during the 5-year period. This includes all pollutants that can be associated under a CO₂ equivalence indicator. NO_x and SO_x emissions resulting from production and distribution activities will not be accounted for in this calculation, as they are not greenhouse gasses, and therefore cannot be associated under a common indicator.Throughout the emission calculations and for each part, an effort is made to stay within the LCA accounting rules of the RED II, though deviations sometimes have to be made from the guidelines (lack of specific data). For the considered feedstocks, annualized emissions from carbon stock changes caused by land use change are not accounted for in the calculations, as all biomass in question originate as biproducts and thus do not cause ILUC. Further, any emission savings from soil carbon accumulation, CO₂ capture and geological storage and CO₂ capture and replacement as expressed in RED II are not considered. Accounting for these emission factors is difficult and uncertain. Including them could lead to highly optimistic results, thus they have been excluded from the analysis. The effect of this decision on the final result will be a slight overestimation of total emissions (according to the RED II emission calculation rules). Further, emissions from the manufacturing of machinery are not taken into account. In other words, the focus is on operations, as they constitute the bulk of emissions over the long-term.

The total emissions (E) in the system: E T = ∑ t ∈ T E t C o l l + E t P r o c + E t T r a n s ∀ t ∈ T (8)

Compared to the total cost function, there are two main differences in the calculation. First and most apparent, the lack of feedstock emissions, which was already explained. Secondly, the pretreatment and conversion phases are grouped into one stage and are called emissions from processing, E ^Proc . This is done in order to be able to use the RED II values as a reference when finding/calculating the total emissions related to the pretreatment and conversion of the biomass.

The emissions related to the biomass collection are proportional to the amount of fuel burned (and the emission factor of diesel (ef)) during the loading process. E t C o l l = ∑ s ∈ S ∑ b ∈ B B U s , b , t ⋅ f c l o a d e r ⋅ e f f ⋅ 1 0 3 L H V b w e t ⋅ ρ b w e t ⋅ r ∀ t ∈ T (9)

The processing emissions (upe) are specific to the type of input biomass and fuel output (f) as well as the refinery (r) E t P r o c = ∑ r ∈ R ∑ p ∈ P ∑ b ∈ B ∑ f ∈ F S F r , p , f , t ⋅ u p e b , f ∀ t ∈ T (10)

Lastly, the transport emissions are set up similarly to the transportation costs. The emissions at each leg are related to the distance travelled, number of trips (or number of trucks/ships needed to transport all of the product in consideration) and the emission factor of the fuel being used. E t T r a n s = ∑ s ∈ S ∑ b ∈ B B U s , b , t ⋅ d s O P r e ⋅ f c t r u c k ⋅ e f f ⋅ 1 0 3 L H V b w e t ⋅ ρ b w e t ⋅ ω t r u c k + ∑ s ∈ S ∑ r ∈ R ∑ i ∈ I T I s , r , i , t ⋅ d s , r S R ⋅ f c t r u c k ⋅ e f f ⋅ 1 0 3 L H V i d r y ⋅ ρ b d r y ⋅ ω t r u c k + ∑ r ∈ R ∑ f ∈ F ∑ p ∈ P S F r , p , f , t ⋅ d r , p R P ⋅ e f f s h i p L H V f f u e l ∀ t ∈ T (11)

The first constraints are the non-negativity constraints. These ensure that all decision variables take on positive values, otherwise the system would minimize the objective functions by allowing the largest negative decision variables, which would yield infinitely low emissions and costs. The lower bound on all decision variables is zero. (Φ) is a binary for the selection of which pretreated biomass (i) to convert into which fuel (f) at each refinery (r) and time (t). B U b , s , t ≥ 0 ∀ b ∈ B , s ∈ S , t ∈ T (12) T I i , s , r , t ≥ 0 ∀ i ∈ I , s ∈ S , r ∈ R , t ∈ T (13) S F r , p , f , t ≥ 0 ∀ r ∈ R , f ∈ F , p ∈ P , t ∈ T (14) D T p , t ≥ 0 ∀ p ∈ P , t ∈ T (15) Φ r , i , f , t ≥ 0 ∀ r ∈ R , i ∈ I , f ∈ F , t ∈ T (16)

Each biorefinery has a certain capacity of fuel production for each type of fuel. For this to be reflected in the model, a capacity constraints should be made for each refinery (rcap is this capacity). The same goes for the ports. However, it is assumed that ports have unlimited storage capacity when it comes to these three fuels. In reality, a real representation of port LNG capacity can be obtained. Port-specific capacity data for methanol and ethanol however, is not available due to the novelty and lack of current use. Regardless, port capacities for the considered fuels are assumed to be unlimited. This way the fuels will flow to the most economic/least environmentally harmful ports without the constraints of present-day infrastructure, which will be useful for the determination of future infrastructure investments. The following constraint ensures that the maximum fuel production level for each fuel and refinery during every time period is maintained at all times. ∑ p ∈ P S F r , p , f , t ≤ r c a p r , f , t ∀ p ∈ P , f ∈ F , t ∈ T (17)

To ensure that the same amount of product (in terms of energy and including losses and efficiencies) flows through the nodes, energy balancing constraints should be placed at each node. To start, the amount of biomass collected should be equal to the amount of intermediate product trucked for each type of biomass and from each country. ∑ b ∈ B B U s , b , t ⋅ δ d r y i n g ⋅ δ m i l l i n g − ∑ r ∈ R ∑ i ∈ I T I s , r , i , t = 0 ∀ t ∈ T , s ∈ S (18)

The amount of intermediate product being transported to the refineries should also be equal to the amount of fuel leaving the refineries, accounting for the conversion yield, and efficiencies. δ i , f c o n represents the yield of converting the intermediate product into a biofuel. ∑ i ∈ I ∑ s ∈ S T I s , r , i , t ⋅ δ i , f c o n ⋅ Φ r , i , f , t − ∑ p ∈ P S F r , p , f , t = 0 ∀ f ∈ F , t ∈ T , r ∈ R (19) Since some feedstocks have multiple conversion pathways and can be converted into different fuels, a constraint is needed to allow for the selection, but also to prevent double (or triple counting). ∑ f ∈ F Φ r , i , f , t = 1 (20)

That is, each inflow of a certain type of biomass from any country at any refinery can only be turned into one type of fuel (otherwise the model would turn it in two all three, which would go against the laws of physics).

The demand constraint ensures that the total demand for each biofuel type at each port is met by the conversion facilities and the transport network. However, the production of a surplus in line with fuel availability is allowed. The purpose of this constraint is to ensure that at least each port is met with the suffice amount of fuel. The slack variable DT _p,t is present again here to allow for a deficit when needed. D F p , t − D T p , t ≤ ∑ r ∈ R ∑ f ∈ F S F r , p , f , t ∀ p ∈ P , t ∈ T (21)

For the supply, the total amount of used biomass of type b in country s during time period t should not exceed the available supply. The following equation ensures this. S B s , b , t ≥ B U s , b , t ∀ s ∈ S , b ∈ B , t ∈ T (22)

The model was programmed in Python and Gurobi was used to solve the MILP model.

For the future supply prediction of the different biomasses, three separate scenarios are developed to encapsulate the inherent uncertainty faced when modelling future conditions. Starting with a baseline scenario based on current/past data and a continuation of historical trends, the supply is also looked at from both high and low availability scenarios. The main difference in the calculation of the three schemes lies in the assumed future biomass competition and specific industry growth.

The baseline availability of each biomass type is calculated in a separate way depending on the available data. For homegenity sake, the data collected is taken primarily from Eurostat () and FAOSTAT (). The maritime industry will face direct competition from other industries for the procurement and utilization of the available biomass. In order to capture this competition, data on current industry energy use will be compiled and compared using an energy balance. The main sources of competition and final end users of renewables include the industry sector, the transport sector and commercial and public services. Current biofuel and energy production is centered around first generation biomass, and will slowly shift to second generation sources over time. Third and fourth generation biomass sources could change near future biomass competition and use depending on their rate of development and adoption. The data of the baseline (current demand and growth) was taken as the basis. In the high competition scenario, the growth of demand was doubled, while in the low competition scenario the growth was the negative of the current growth.

To start, the amount of agricultural residues potentially available for bioenergy production is calculated as the difference between the total produced residues (total amount of crop produced times the residue-to-crop ratio) and the sum of residues left on the field and those used for other purposes. In a similar way, forestry residues are the sum of the total woodchips produced from forestry activities and the recoverable wood products from wood processing. Animal wastes are calculated on the basis of total livestock heads for each animal type per country, the average daily manure production of each animal, and the manure’s moisture content. Organic and biowaste potentials are taken as the recoverable fraction of the total amount produced in the in-scope countries (see Figure 2).

Additional to the supply of biomass, the distribution of biofuel refineries across Europe is needed. Based on data from Concawe (ESER, 2021), the locations along with the respective capacities and type of fuel produced are mapped out. These are presented in Table 3.

The refineries are classified into two primary groups; Bio-LNG and Bio-Liquid (ethanol and methanol, but also others) facilities. The bio-LNG refineries are well documented and straightforward. Data on other bio-refineries, especially the type and outputted product, is more difficult to assess. Different sources show different numbers for different types of refineries. In order to tackle this, the total amount of bio-liquid refineries are calculated in each country and are multiplied by the proportion of output capacity of a certain bio-liquid to total bio-liquid output in Europe. This can be though of as an averaging technique. This figure is multiplied by the average output of a medium sized refinery (specific to each fuel and feedstock). Further, the pilot plants under construction are considered for expansion of capacities in the coming years. The refineries are then grouped and their capacities are assigned to the closest port. In other words, the refineries are assumed to be at the ports and the total outputting capacity of a port is proportionate to the capacities of the refineries in its vicinity.

In the case of the biofuels in general (bio-LNG, bio-ethanol, bio-methanol and others) in the marine industry, the main two initiatives for the pursuit of environmentally-friendly innovation are operational drivers in the form of cost reduction and regulatory policy (Aronietis et al., 2016). Detailed demand per fuel is not available, hence this section focusses on the general demand.

The general trend of rising oil prices in combination with the volatility of the market has created a market of new fuels to dampen the effects of crude-oil price uncertainty (Ciria and Barro, 2016). Assessing and comparing the costs of advanced biofuels is an objective of this report, however, current and future regulations are unresolved and not binding. The current imposed European mandate from RED II sets the goal of having at least 40% of the EU’s gross final energy consumption be renewable by 2030. However, the RED II also outlines a separate sub-target that relates to transport, although it does not mention the maritime sector. Within the 14% transport sub-target, there is a dedicated target for advanced biofuels produced from feedstocks listed in Part A of Annex IX. The contribution of advanced biofuels and biogas produced from the feedstock listed in Part A of Annex IX as a share of final consumption of energy in the transport sector shall be at least 0.2% in 2022, at least 1% in 2025 and at least 3.5% in 2030 (see also Table 4). As the fuels considered in this report are considered advanced, the targets used for the future demand are these.

In order to convert this percentage into an amount of energy, the total current and predicted consumed energy in the maritime sector should be calculated. It is assumed that each energy sector complies with the above target and thus only the bunkering energy is relevant for the maritime industry.

Total marine bunker fuel demand is considered in the IEA’s World Energy Model, the OPEC World Energy Model and the EIA’s (Energy Information Administration) World Energy Protection Model at both the collective and regional levels. The IMO also publishes a recurrent study on fleet emissions to gain insight into GHG emissions, fleet development, fuel use and bunker demand. The most recent study, the Fourth IMO GHG study 2020 builds on past developments and uses new data to produce more reliable GHG inventories. It is also the first study to distinguish between international and domestic shipping on a voyage basis (IMO MEPC, 2021). The study identifies two main factors in the transport demand projection, namely, the long-term socio-economic scenarios. GDP and population growth are the two main indicators in this, the higher the predicted growth of these two variables, the higher the projected transport work for products that are positively correlated to them. Second, the long-term energy demand scenario. Higher consumption of fuels leads to higher transport work. To account for different recovery scenarios from corona, GDP growth, population change, etc., the latest IMO report has come up with a set of different scenarios that depict multiple future possibilities of worldwide marine energy demand. Three of those scenarios are used in this report as benchmarks for low, medium and high demand cases.

To assess maritime energy demand in Europe per port, the proportion of fuel consumed at European ports is taken as a fraction of total world consumption. As a rough estimate, the share of bunker energy demand in Europe lies between 20 and 25% of all worldwide bunker energy demand depending on the year in question (IEA, 2020). This figure might seem large, however, a small number of ports account for the majority of the worldwide sales, of which Algeciras, Rotterdam and Antwerp are significant (Lin, 2021).

The OPEC released a World Oil Outlook in 2014 outlining several supply and demand scenarios for the oil industry. In it, key figures were provided on the bunker sales of key worldwide ports (OPEC, 2014). By using a historic approximation of the share of worldwide bunker sales of the main European ports from the IEA 2021 and normalizing them with respect to the total European marine energy demand, one can estimate the percentage of bunker energy demand for each of the in-scope ports (with respect to Europe). Three scenarios are developed based on the IMO GHG scenarios and the RED II advanced biofuel targets. In the high scenario, the RED II targets are taken as twice the nominal values. These scenarios are depicted in Figure 3. For readability, only the top five ports are shown.

To start, one of the most apparent and clear outputs of the model in the various scenarios was the choice of biomass. Unmistakably, the model showed an overall inclination towards forestry residue biomass in every scenario. This choice was due to a balance of biomass physical properties, price and source location. The conversion and transportation advantages of this feedstock proved to outweigh the costs of the biomass and relative transport distance to conversion plants. It was also noted that the option to use forestry residues was more prevalent in scenarios with low or medium demands. When the demand was increased over a certain threshold (around 100 PJ per year, but also depending on the supply scenario), the model ran into an availability constraint with respect to forestry residues and opted for different feedstock choices. For options in which only one specific fuel was used, the choices differed. In the case of ethanol production, forestry residues still reigned supreme. For methanol, agricultural residues and livestock manure also made up a large quantity, and for LNG a significant portion of the feedstock was manure, on top of forestry residues.

The biomass suppliers were rather consistent throughout all run scenarios. The vast majority of biomass energy was supplied from Poland in the form of forestry residues. France was the runner up, followed by the Netherlands, which mainly supplied manure. Romania, Slovenia, Estonia and Spain all supplied a much lower quantity of biomass energy. The rest of the countries supplied biomass to a much lower extent. The countries with the lowest use included Czechia, Finland, Ireland, Sweden, Switzerland and United Kingdom. Factors such as biomass availability, geographical location, proximity to refinery, fuel yield, feedstock prices and available trade routes were all determinants of a specific feedstock uptake. It was discovered that physical properties bore more weight in the selection than originally believed. On top of this, since the pretreatment of biomass was assumed to take place within the country of origin, and the pretreatment cost was largely a function of electricity cost and moisture content as well as LHV, countries with lower electricity costs were more suitable for sourcing. The preferred bio-types were those with low moisture contents and high lower heating values (such as wood), or those that were located relatively close to the refineries (small countries with ports, i.e. Netherlands, Belgium).

In terms of fuel production, the choice to produce one over another turned out to be largely dictated by the available processing capacities. There was a stronger inclination towards ethanol and methanol than LNG, mainly due to the installed capacities of both.

The trade routes were also very centered around a few strong ones, given that the biomass sources remained mostly constant in all scenarios. The routes between countries and ports were fairly straightforward, with a propensity towards the shortest path while still meeting refinery capacity constraints. Poland was a main exporter to the two German ports; Bremmerhaven and Hamburg, but also Antwerp and Gdansk. France mainly exported to Le Havre, but also Rotterdam and Antwerp when local capacities were used up. Insofar as the shipping routes, the majority of the product transport was centered around the northern European ports, specifically Rotterdam, Antwerp, Hamburg and Bremmerhaven (See Figure 4). Though the capacities of Bremmerhaven and Hamburg were equal, Hamburg in general had more fuel outflows, since it is located more towards the east, where the majority of biomass originates from. Most of the biofuel produced in Rotterdam remained at Rotterdam to meet the energy demand quota there. In cases where the demand was low however, Rotterdam became a major exporter to all ports (depending on the scenario).

It was also noticed that in times when an energy deficit was necessary (the high demand cases, ethanol, LNG), the largest deficits would generally appear in the southern-Atlantic ports of Algeciras and Sines, as well as Barcelona. Algeciras by far had the largest deficits in terms of total energy. However, as a percentage of local demand, the deficits in Spain and Portugal were quite substantial. Moreover, a very large amount of fuel is shipped to these locations from the northern seaports. It is not the case that Spain and Portugal do not have the necessary supply of biomass, however, there is a strong absence of refining capabilities in those countries. Algeciras would greatly benefit from a larger installment of refineries, and could possibly become a large exporter to the Mediterranean ports if this was the case.