The goal of this CF study is firstly to evaluate the importance of LUC and secondly to show consequences of modeling choices for by-product substitution. The functional unit is 1 ha of plantation area, using the average performance for a plantation lifetime of 25 years.

A state-of-the-art production system as described above is investigated, and Monte Carlo simulations with 1,000 runs are used to show system-inherent variability of FFB production. For the Monte Carlo simulations, triangle distribution of the two main parameters, yield and LUC, are used. The change in yield is associated with an increase or decrease of linked parameters, e.g. yield decrease also leads to higher land demand per product unit and higher fertilizer input and consequently higher GHG-emissions per product unit; the same applies to the application of pesticides and diesel consumed. Therefore, input parameters linked to the yield are not used for the Monte Carlo simulations. Waste management is discussed in detail^[9,32] and more recently also in^[30] and therefore not repeated here.

In 2020, approx. 21% of the cleared forest in Indonesia was used for oil palms, according to^[6]. The situation concerning LUC is complex. To account for the complexity and regional variability across Indonesia, 20% LUC was chosen as the upper level and 1% as the lower level.

Various LUC scenarios are investigated:

- Palm oil production without LUC (baseline), assuming an FFB yield of 20 t FFB ha^-1.

- 1%, 5%, 10%, and 20% LUC, equally converted from forest and peatland, i.e., 10% LUC = 5% from forest and 5% from peatland. The equal conversion from forest and peatland is arbitrary; therefore, Monte Carlo simulations are used to calculate the uncertainty. In addition, two scenarios for 10% LUC from forest and 10% from peatland are calculated separately.

- Annual GHG emission factors from direct land use change in tropical regions are taken from^[44].

LUC is of paramount importance with respect to the climate-relevant performances of oil palm plantation; therefore, 1 ha is used as a functional unit. Substitution is applied for by-products and surplus electricity. Moreover, the uncertainty due to substitution, is investigated by applying an approach used by^[45] and in addition, three options with and without LUC. Substitutes are identified based on function and existing markets.

Inventory data for oil palm production are taken from^[31] and the nutrient composition of fronds and EFB are taken from^[9].

The following assumptions are used for the CHP plant:

- For sake of simplicity the oil extraction rate (OER) is assumed to be 23%, although it can vary between 19% and 26%. However, the technical improvement of OER is not within the scope of this study

- 1 t FFB processed in the palm oil mill generates 130 kg mesocarp fibers and 54.6 kg shells, the ratio between the residues is assumed to be constant.

- The average low calorific value (LCV) of mesocarp fibers is 11 MJ kg^-1, while the LCV of shells is 13.4 MJ kg^-1. These values can vary, e.g.^[55] reported 16.3 MJ kg^-1kernel nut shells and^[56] 10 MJ kg^-1 for mesocarp fibers. The selected LCV values are at the lower end so that electricity surplus is not overestimated.

- Heat demand is 1,100 MJ t^-1 FFB and electricity demand is 22.5 kWh t^-1 FFB^[57].

- CHP efficiency in typical palm oil mills is 0.72 according to^[57] and can reach 88% in energy-optimized CHP plants. Just the lower CHP-efficiency is used, as it represents the current situation in palm oil mills better.

The life cycle stages and related mass flows for 1 ha plantation area are shown in Figure 3.

This study uses the CF as quantitative indicator. The CF is based on the Product Environmental Footprint methodology, version 3.1. The CF measures the potential contribution of a product to global warming. It considers all GHG emissions associated with the product throughout its life cycle—from raw material extraction to production, use, and end-of-life. The contribution of fossil-based emissions, biogenic-based emissions and emissions from land use change are separately shown for the oil palm plantation.

The CF of oil palm plantations is primarily driven by LUC and by biogenic methane emissions resulting from the treatment of POME. The CF varies significantly depending on the type of land converted. The same LUC from forest and peatland is arbitrarily assumed, although it varies between different regions of Indonesia. Therefore, two additional calculations are conducted to show the CF when 10% LUC comes either from forest or peatland. The CF is calculated for oil palm plantations having a yield of 20 t FFB ha^-1; corresponding CF results are shown in Table 1.

In the baseline scenario, 45% of CF originates from POME treatment, approx. 17.5% from N-fertilizer, 17.5% from diesel consumption, and 10% from potassium fertilizer. The CF nearly doubles if 10% of the area is converted from peatland or if 20% is converted from a mix of forest and peatland. Other environmental impacts are not considered in this study. In the LUC 20% scenario, CF originated from LUC (approx. 50%) and plantation activities, thereof 23% from storage of POME in lagoons. Biogenic methane emissions from POME treatment can be reduced by 2,600 kg CO_2eq per 20 t FFB when co-composted with EFB. The high CF of peatland conversion is due to the drainage of organic matter, which triggers CO₂ release through oxidation. While CH₄ and N₂O emissions occur, their contribution is small compared to the CO₂ released from peatlands.

Monte Carlo simulations were used to assess the variability of CF based on yield and LUC. Change in yield is associated with the increase or decrease of linked parameters, e.g., yield decrease also leads to higher area demand and higher nitrogen input and consequently higher GHG emissions per product unit; the same applies to the application of pesticides and diesel consumed. Therefore, input parameters linked to the yield are not used to avoid unrealistically high uncertainty due to the nature of Monte Carlo simulations. Typical yield ranges from 14 t to 25 t per productive area, besides four land use change scenarios (1%, 5%, 10%, and 20%) are defined. In the LUC scenarios, the contribution from forests and peatlands ranges from 0% to 100%. All Monte Carlo simulations are calculated with 1000 iterations. The CF results are shown in Table 2 and Figure 4.

In Figure 4 the upper and lower quartiles are shown, the median is indicated by a straight line, the average by crosses, and outliers are shown as circles; outliers are defined by 1.5 times the interquartile distance.

Land use change from peatland has a detrimental effect on climate change. The CF per ha doubles when 20% LUC occurs compared to no LUC. The uncertainty of CF increases with increasing LUC. The difference between the 5% percentile and 95% percentile increases from 2,382 kg CO_2e in the baseline to 10,248 kg CO_2e in the LUC 20% scenario. The fact that smallholders frequently achieve only 42% of attainable yields indicates a significant opportunity for improvement. Approx. 40% of 17 million ha of oil palm plantations are managed by smallholders^[58]. The average yield from 2017 - 2021 was 17.5 t FFB ha^-1, according to FAOSTAT^[59]. A yield increase of smallholders’ plantation from 11 t FFB to 20 t FFB ha^-1 would generate an additional 14 million t CPO, 2.0 million t PKO and 2.2 million t PKM without significantly higher GHG emissions. The assumed yield increase for smallholders equals a plantation area of 3 million ha.

Palm oil mills are inherently energy self-sufficient, utilizing residues (mesocarp fibers and shells) to heat and power. Processing 1 t of FFB generates 130 kg of fibers and 54.6 kg of shells. The total energy is calculated based on the energy content and the conversion efficiency of the CHP-plant:

$$ \text { Total energy }=\text { Shells* } \mathrm{LCV}_{\text {shells }}+\text { fibers*LCV }_{\text {fibres }}=43,233 * \mathrm{CHP}_{\text {efficiency }} $$

The total energy content of these residues (31,123-38,045 MJ ha^-1) comfortably exceeds the mill’s heat demand (22,000 MJ) and electricity demand (1,623 MJ) for processing 20 t FFB. At 72% efficiency, the system can export a surplus of 2,083 kWh to the national grid. Providing there is grid access, the CHP unit allows the mill to transition from a net GHG-emitter to a climate-beneficial system by displacing fossil-based electricity.

The mass and energy flows of the palm oil mill are shown in Figure 5.

The palm oil mill is divided into FFB processing (left side of Figure 5) and a CHP (right side); lagoons are associated to the plantation because the treated POME is used for irrigation. Environmental performance can be improved by treating POME instead of storing it in lagoons, as already demonstrated by^[32,60-62]. The oil mill is modeled using the Indonesian electricity mix and global heat mix based on natural gas due to data limitations. The residues are used as biogenic feedstock for the CHP-plant. The CHP is considered a waste treatment unit; hence it received the residues without upstream emissions. The generated energy in the CHP is credited by the same heat and electricity mix used in the palm oil mill. The treatment of 20 t FFB (yield per ha equals 4.6 t CPO) in the mill results in 2.0 t CO_2eq while the CHP credit would be 3.1 t CO_2eq, provided access to the electricity grid exists. The CHP covers the entire heat demand of the oil mill. The combined system (oil mill + CHP) is climate-friendly and saves fossil resources but produces harmful dust and other air emissions^[31]. The mill generates three products, CPO and kernels, which are further processed in specialized mills, as well as surplus electricity. Feed-in tariffs for electricity vary between different regions in Indonesia, as do prices for palm oil products. Hence, a simplified economic approach is applied, where the revenues from CPO, PKO, and PKM are used for the palm kernel mill operations. Recent revenue figures are taken from Malaysia^[63] due to a lack of data from Indonesia.

The CF results are highly sensitive to how PKO and PKM are handled. The most important application of PKO is in the oleochemical sector; 70% are used for oleochemicals, 28% for food, and just 2% for biofuels, according to^[64]. Table 3 shows that based on fatty acid composition, PKO serves as a direct substitute for coconut oil. The average price of PKO and coconut oil is almost the same. Therefore, 1 t PKO substitutes 1 kg coconut oil.

The consequences of modeling choices will be demonstrated using publications from^[45,66], although they aggregated CPO and PKO hence do not use substitution for PKO despite both oils ending-up in different markets^[64].

PKM is used for animal feed, but identifying a single substitute is difficult because PKM provides protein, energy, and fiber. The composition of relevant animal feed ingredients is shown in Table 4.

Schmidt and Weidema^[45] used the Feeding Component Table of^[68] in order to calculate the substituted products for 0.122 g PKM, which is generated when 1.0 kg palm oil is produced. PKM replaced 0.035 kg of soybean meal (SBM) and 0.066 kg barley and generated additional 0.007 kg of soybean oil (SBO).

Alternatively, just SBM can be substituted by PKM based on the protein content, 0.122 kg PKM would substitute 0.045 kg of SBM and 0.011 kg of SBO. Using the feed composition figures from Table 4 results in the figures shown in Table 5.

Neither soybean meal (SBM) alone nor a mixture of SBM plus barley provides the same function as PKM, e.g. the fiber content is substantially lower, while the starch content is higher, as shown in Table 5.

The cereal unit (CU) is a key figure, which reflects the energy supply capacity of a product in relation to the calculated energy supply capacity of feed barley, depending on the structure of use of the agricultural product in feeding, according to BMELH^[69]. The CU of PKM is 0.41 CU dt^-1 and of soybean meal 0.95 CU dt^-1. Hence, 0.122 g PKM would replace 0.91 g SBM.

Depending on whether PKM is assumed to replace SBM, barley, or a “cereal unit”, credits can deviate by up to 100%. This illustrates the importance of modeling choices and illustrates also the uncertainty related to product substitutions. Therefore, the assumptions for product substitution need to be transparently described.

To show the consequences of the modeling choices for by-product substitution, the same amount of PKM is used for the following analysis. Based on the economic revenue the kernels account for 10% of the upstream emissions, i.e. are 0.35 t CO_2eq per t kernel. The kernels are transported to kernel mills and processed to PKO and PKM. Treating the kernels causes another 0.062 t CO_2eq t^-1 and generates 0.53 kg PKM and 0.47 kg PKO. Using market-based substitution ratios of 78.5% for PKO and 21.4% for PKM results in 0.75 t CO_2eq t^-1 PKO and 0.15 t CO_2eq t^-1 PKM.

For simplicity, it is assumed that the same amounts of PKM and SBO are produced as reported by^[45]. The system is credited for by-products using data from^[34], consequential module, market-datasets. Results are shown in Table 6 and compared with^[45].

The credits vary substantially. The most important assumption by [66] is that PKO equals CPO, which is an oversimplification. For each ton CPO, approx. 0.14 t PKO is produced. Globally 15% of CPO is used as biofuel and 77% as food, while 70% of PKO is used for oleochemicals, 28% for food, and just 2% for biofuels, according to^[64]. The choice of the most appropriate substitution for PKM (and PKO) is context-specific and limits the comparability of results from different studies and other data sources, respectively.

In addition to the comparison above, different approaches used by^[28-30] are applied to reveal the consequence of modeling choice for by-products for this analysis. Six scenarios are defined, i.e., baseline without LUC, and 5 % LUC, each with three approaches. Coconut is used as a replacement for PKO and soybean for protein feed from Ecoinvent 3.11, consequential. Results are expressed per ha and subsequently converted per t CPO, or t (CPO + PKO) in Table 7. The scenarios are defined below:

(1) Baseline, CPO as product PKO and PKM as distinguished products

(2) Baseline CPO + PKO as combined product and PKM as by-product (similar to^[45])

(3) Baseline, CPO and kernels as negligible by-product (similar to^[29])

(4) With 5% LUC and as 1

(5) With 5% LUC and as 2

(6) With 5%LUC and as 3

The CF of CPO ranges from 0.26E3 to 1.4E3 kg CO_2eq t^-1 without LUC and 0.85E3 to 1.9E3 kg CO_2eq t^-1 depending on the modeling choice for by-products. These variations demonstrate that a direct comparison of results from different studies is often impossible because substitution can introduce remarkable uncertainties. Therefore, it is highly recommended to check the underlying assumptions and relationships used in the CF models, as also suggested by^[70-72].

This study aims to estimate the carbon footprint variability due to LUC in Indonesia. Regional variables, such as peatland depth in Kalimantan or local feed-in tariffs, may lead to different localized results. In this study it is assumed that access to the electricity grid exists.

Due to data limitations, revenue figures from Malaysia (2023-2024) were used as a proxy for the Indonesian system. Given the similarity of palm oil production systems in both countries and the global market price fluctuations, it is assumed to be a reasonable approach. Although Indonesian average data over the same data collection period as for the palm oil system would be the preferred option.

Substituting single ingredients for PKM is not fully representative of real-world animal feed markets; economically optimized feed mixtures would provide a more realistic baseline.