Oil palm modelling in the global land-surface model ORCHIDEE-MICT

. Oil palm is the most productive oil crop that provides ~40% of the global vegetable oil supply, with 7% of the cultivated land devoted to oil plants. The rapid expansion of oil palm cultivation is seen as one of the major cause for 20 deforestation emissions and threatens the conservation of rain forest and swamp areas and their associated ecosystem services in tropical areas. Given the importance of oil palm in oil production and its adverse environmental consequences, it is important to understand the physiological and phenological processes of oil palm and its impacts on the carbon, water and energy cycles. In most yields at multiple observation sites. Compared with the default ORCHIDEE-MICT version, ORCHIDEE-MICT-OP shows improved performance of GPP partitioning, NPP allocation and biomass components. The new oil palm version, parameterized with age- 515 specific parameters, generally captures temporal dynamics of oil palm biomass and yields. Implementation of more management practices (e.g., fertilization and irrigation) and parameterization of biophysical variables are further needed. Generally, our model improves the representation of oil palm in LSMs and further applications of ORCHIDEE-MICT-OP include but are not limited to regional carbon budget and water demand estimation, yield prediction and the sustainable development of oil palm industry.


Introduction 35
Oil palm is one of the most important vegetative oil crop in the world. It provides 39% of the global supply of vegetable oil and occupies 7% of the agricultural land devoted to oil-producing plants (Caliman, 2011;Rival and Levang, 2014). With the increasing demand for palm oil as a biofuel and a feedstock for industrial products, oil palm plantation continuously expanded from 5.59 to 19.50 million ha during 2001-2016 in the world's top two palm oil producers, Malaysia and Indonesia (Xu et al., 2020). This rapid expansion brought about high ecological and social costs. About half of the oil palm cultivation lands were 40 converted from biodiverse tropical forests during 1990(Koh and Wilcove, 2008, leading to losses of habitats (Fitzherbert et al., 2008), peatlands (Koh et al., 2011;Miettinen et al., 2016) and carbon emissions from land use change (Guillaume et al., 2018). Land use change (LUC) from peat swamp forest to oil palm plantation contributed about 16-28% of the total national greenhouse gas (GHG) emissions in Southeastern Asia (Cooper et al., 2020). A comprehensive understanding of fruit production, land use change, carbon emissions and other environmental consequences of oil palm is urgently needed 45 for guiding more sustainable management practices.
Many field-based studies underpinned the specific phenology and growth of oil palm and its key physiological processes (Ahongshangbam et al., 2019;Lamade and Bouillet, 2005;Noor and Harun, 2004;Sunaryathy et al., 2015). Models developed based on these field observations provide a useful tool for large-scale simulation of oil palm growth and yields and their impacts on the regional carbon, water and energy budgets. Oil palm growth models have been developed to simulate the 50 biomass yields of oil palm based on the physiological processes and phenological characteristics such as flowering and rotation dynamics (Van Kraalingen et al., 1989;Henson, 2009;Combres et al., 2013;Hoffmann et al., 2014;Huth et al., 2014;Paterson et al., 2015;Teh Boon Sung and See Siang, 2018). Although these models can generally reproduce the observed yields, they are usually applied for fruit production simulation without the whole carbon, water and energy cycle, do not allow to represent the land use changes, thus usually cannot be integrated for regional and global gridded simulations like land surface models. 55 Alternatively, process-based land surface models (LSMs) can simulate spatially explicit plant growth, biomass density and yield and a full set of carbon, nutrient, water and energy fluxes and storage pools (Fisher et al., 2014). Vegetation in most LSMs is represented by a discrete number of plant functional types (PFTs) and oil palm is approximated by tropical broadleaved evergreen (TBE) trees without a specific representation in LSMs, although the physiological characteristics of oil palm differ from generic TBE trees. For example, the maximum leaf area index (LAI) of oil palm is up to 6 m 2 m −2 according 60 to the genotypes and locations, which is lower than TBE (8 m 2 m −2 ) in Indonesia and other plantations such as rubber (9 m 2 m −2 ) (Vernimmen et al., 2007;Propastin, 2009;Rusli and Majid, 2014). The Maximum rate of carboxylation, Vcmax25 of mature oil palm, by contrast, is higher than in natural tropical forests (Carswell et al., 2000;Kattge et al., 2009;Teh Boon Sung and See Siang, 2018). Oil palm has a shallower rooting system and lower above ground biomass compared to forests (Carr, 2011), and its above and below ground biomass ratio is lower than the natural forests . To maintain a huge 65 fruit productivity with shallow roots, a large amount of water is required by oil palm for evapotranspiration (~4-6 mm d -1 ), typically 25% higher than in tropical forests in the same region (Manoli et al., 2018;Meijide et al., 2017). Ignoring those https://doi.org/10.5194/gmd-2020-252 Preprint. Discussion started: 22 October 2020 c Author(s) 2020. CC BY 4.0 License. differences in the parameterizations of LSMs would cause biases when simulating oil palm growth, yields and the biophysical processes in large-scale model application, which calls for new parameterizations dedicated to oil palm as a specific PFT in those models. 70 Oil palm has a specific morphology, phenology and management practice compared to other perennial crops and tropical evergreen forests. Oil palm has a solitary columnar stem with phytomers (palm branches supporting leaves and fruit bunches) produced in succession at the top of stem. Each phytomer carries a bunch of fruit and experiences a life cycle from leaf initiation, inflorescences and fruit developing to harvest and pruning (Fan et al., 2015). At the maturity stage, one oil palm tree holds ~40 visible expanded phytomers from the youngest to the oldest, and 40-60 initiating phytomers within the apical buds 75 (Combres et al., 2013). It takes about 3-4 years for the reproductive organ to develop before flower initiation and fruit harvest (Corley and Tinker, 2015). Currently, the biomass pool of phytomers is not included in the generic tree PFTs of the ORCHIDEE land surface model, which prevents modelling of phytomer-specific monthly harvest and pruning. In addition, the closest PFT of oil palm in the model, known as TBE, has a different leaf phenology -with a higher old leaf turnover and increased new leaf production in the dry season, based on the satellite and ground based observations (Wu et al., 2016). This 80 leaf phenology scheme was parameterized for leaf age cohorts in ORCHIDEE for Amazonian evergreen forest (Chen et al., 2020) but whether it can be adapted to the oil palm or not needs further investigations. At the productive stage, regular harvest and pruning are applied to maintain the optimal number of phytomer and maximize harvested yields. Also, oil palm is managed in a rotation cycle of 25-30 years (manually cut) due to the difficulties in harvesting and the potential decline of fruit production (Hoffmann et al., 2014;Röll et al., 2015). Thus, oil palm cannot be described as an annual crop, neither as a natural tree PFT 85 with a longevity of decades to centuries. Therefore, including forest age dynamics  is needed in a LSM to represent the management practice and cycle of growth, fruit harvest and rotation of oil palm at different age stages. CLM-PALM was the first LSM that distinguished specific palm plant from the generic PFTs using a sub-canopy framework from CLM4.5 (Fan et al., 2015). This work provides an important conceptual framework for implementing oil palm modelling in other land surface models. 90 In this study, we aimed to model oil palm growth from young to mature plants and the specific morphology, phenology and management characteristics in the ORCHIDEE LSM. The oil palm integration was based on existing leaf age cohorts-based phenology of TBE and distinct age classes of the model, but significant modifications have been made to accommodate the phenology, physiological and management characteristics of oil palm. The oil palm growth from leaf initiation, fruit development, maturity to tree cutting is represented. A sub-PFT structure-phytomer with leaf, branch and fruit for oil palm 95 was implemented in ORCHIDEE LSM. The plant carbon allocation scheme was modified to support the growth of branch, leaf and fruit component of each phytomer. Management practice of pruning, fruit harvest and rotation was also implemented.
The objectives of this study are to 1) implement growth (especially phytomer development), phenology and harvest processes for oil palm as a new PFT of the land surface model, 2) adjust physiological and phenological parameters using field https://doi.org/10.5194/gmd-2020-252 Preprint.

Observation Data
Data from 14 sites with reported coordinates were collected from published literatures for modelmodel validation (Table S1).
Since tropical humid climate is favourable for oil palm growth, most of in situ measurements are located in Indonesia (6 sites) 105 and Malaysia (7 sites) except one site in Benin ( Figure 1). The observation sites have high mean annual precipitation (MAP, 574.2-3598.8 mm yr −1 ) and high mean annual temperatures (MAT) between 24.3°C and 28.8°C throughout the year, which covers 97.27% and 85.14% range of MAP and MAT in the global oil palm plantation area respectively in 2010 (Cheng et al., 2018) ( Figure S1). The MAT, MAP and clay fraction (CF) for the global oil palm plantation area were based on the climate data from the CRUNCEP gridded dataset (Viovy, 2011) and the Harmonized World Soil Database (HWSD v1.2, (Nachtergaele 110 et al., 2010)). The observation sites include 4 smaller plantations (<50 ha, Site 1 for smallholders and Site 4, 5 and 12 for research sites, Figure 1) and 10 industrial plantations up to 23625 ha. Site 12 and Site 14 were covered by very deep peat soil before oil palm cultivation, where the former natural vegetation was peat swamp forest. The natural vegetation in other sites was dominated by tropical rainforest and clay fraction varies from 0-11% ( Figure S1). LAI, gross primary productivity (GPP), net primary productivity (NPP), fruit bunches (yield) and biomass at different ages including young and mature oil palms were 115 collected from these sites for model validation. Annual data of total biomass and yields were available for Site 3 and Site 12.
The biomass data at Site 3 was calculated by allometric equation using the measured diameter at breast height (DBH) and height of the stem (Corley and Tinker 2015), while yield data at Site 12 was obtained from measurements of harvested fruit bunch every time. Sites 1, 2, 12 and 13 provide observations of different NPP components by quantifying all the plant pools change for a specified time interval. Fractions of different biomass parts were collected by combining measurements of 120 biomass partition and calculations using empirical equations in Site 12 and Site 3 (see details in Table S1).

Organising Carbon and Hydrology in Dynamic Ecosystems (ORCHIDEE) is the land surface component of the French Institut
Pierre Simon Laplace (IPSL) Earth system model (ESM) and capable of simulating water, energy and carbon processes (Krinner et al., 2005). ORCHIDEE-MICT (aMeliorated Interactions between Carbon and Temperature) is a branch of 125 ORCHIDEE with a better representation of high latitude process with new vertical soil parameterization, snow processes, and fires (Guimberteau et al., 2018). The recent ORCHIDEE-MICT v8.4.2 also includes modifications in wood harvest, forest age class and gross land use changes . The need to represent age-specific physiological and phenological characteristics for young and mature oil palm can thus benefit from this pre-existing forest age dynamics representation. Therefore, our development of oil palm modelling started from ORCHIDEE-MICT v8.4.2.
Processes related to the carbon cycle in ORCHIDEE include photosynthesis, respiration, carbon allocation, litterfall, plant phenology and decomposition (Krinner et al., 2005). We added a new PFT for oil palm starting from the default setting of the closest PFT -TBE trees. The major modification brought was for the carbon allocation, by including a new phytomer organ for oil palm, and a new fruit harvest module for fresh fruit bunch harvesting ( Figure 2). The new model called ORCHIDEE-MICT-OP (oil palm) is schematized in Figure S2. Oil palm has a monopodial architecture and sequential phenology. The phytomers are produced in succession, each bearing a big leaf with a number of leaflets, rachis and a bunch of fruits (Corley and Tinker, 2015;Fan et al., 2015). To represent the major morphology and phenological process, we introduce a new phytomer structure in the model frame. Phytomers are 140 initiated successively and developed in parallel on the same tree. Each of them contains three components-leaf, branch and fruit bunch. Although each phytomer has its own sequence of initiation, allocation, fruit production and pruning, they share the same stem and root biomass and the same carbon assimilation process. In the default version of ORCHIDEE-MICT, there were eight biomass pools: leaves, sapwood above and below ground, heartwood above and below ground, roots, seed and carbon reserve pools. To simplify the modification and parameterization of phytomers and keep consistent with the model 145 structure, the leaf, branch and fruit bunch belonging to each phytomer were linked with the original leaf, sapwood biomass pools, although the fruit-bunch biomass pool was modified from the original model ( Figure 2). And the number of fruit and branch component was set corresponded to phytomer number but the leaf linked with leaf biomass pool was divided to four age classes without duplication in each phytomers ( Figure 2).

Phytomer phenology 150
Here we describe the phytomer dynamics related to planting, vegetative maturity and rotation at plant level and the sequential initiation and pruning at phytomer level. The modification of leaf seasonality is also presented. A schematic diagram of oil palm tree, phytomer and leaf phenology is shown in Figure 3. Since the phytomer phenology is closely related to the age of the tree, the age of the phytomer and the age of leaf, three temporal variables of tree age (the age of oil palm tree in years), the phytomer age (the age for each phytomer counted from its initiation, in days) and the leaf ages (the age of leaves in day) were 155 used to compute tree, phytomer and leaf dynamics ( Figure 3).
Based on the field evidence, there are three major phenological phases for phytomers during a tree life cycle. The first phase is the first three years between oil palm planting and the beginning of fruit-fill. In this period, leaf and branch begin flourish and expand without fruit production. The second phase is the fruit development phase when fruit begins to grow, while fruit and branch biomass continues to increase. The third phase is the productive phase with high and stable yields that will last 160 until the age of 25-30 years old. This phase ends up when the tree grows very tall (harvesting of fruit bunches becomes difficult) https://doi.org/10.5194/gmd-2020-252 Preprint. Discussion started: 22 October 2020 c Author(s) 2020. CC BY 4.0 License. and the fruit yield starts to decrease. The modified subroutines of phytomer dynamics are adopted from the forest age cohorts simulated in ORCHIDEE-MICT v8.4.2. The forest age cohort module was originally designed for modelling forest management such as wood harvest and gross land use changes . This module allows us to represent photosynthesis, allocation and harvest practice for different forest age classes (each tree PFT is divided into 6 age 'cohort 165 functional types' called CFTs) by setting CFT-specific parameters. This module is adopted to represent the rotation cycle of oil palm and the land conversion to or from oil palm. Detailed parameterization for the new oil palm CFTs is presented in Sec.

2.5.
For an adult oil palm tree, the number of newly produced phytomers is stable at around 20-24 per year (Corley and Tinker, 2015). Phytomers are manually pruned twice a month to keep a maximum number of 40 phytomers, while fresh fruit bunches 170 are harvested every 15-20 days (Combres et al., 2013;Corley and Tinker, 2015). Considering the regular development of phytomers and the periodic harvest and pruning practices, the initiation of new phytomers occurs every 16 days, and the phytomer longevity (640=16×40, Figure 3) is set by this fixed initiation interval and by the maximum number of expanded phytomers of 40 in the model. Thereafter, we introduce two temporal variables in unit of days, i.e., the critical phytomer age or phytomer longevity ( ℎ ) and the age of each phytomer ( ℎ , ℎ ). The former defines the time length between 175 phytomer initiation and pruning, while the latter records the age of each phytomer. When the phytomer age reaches the critical value, the pruning practice is triggered and the pruned phytomer (including the branch and leaf) goes into the litter pool of the model. Subsequently, another new phytomer is initiated to maintain the total number of phytomers. The carbon allocation and harvest related to phytomer dynamics is discussed in the Sec. 2.3.3 and 2.3.4.
Leaf phenology of TBE forest is important for seasonal carbon and water fluxes. In another version of ORCHIDEE-MIC, the 180 leaf phenology of TBE forests was implemented using four leaf age cohorts (See Figure 3) by Chen et al. (2020). Different photosynthetic efficiencies were used for leaf age cohorts to represent the leaf aging process. In this new canopy phenology scheme, NPP allocation to new leaves is driven by SWdown and the vegetation optical depth of old leaves (Eq. 1 in Chen et al. (2020)), and weekly VPD is used to trigger the shedding of old leaves (Eq. 3 in Chen et al. (2020)). Here, we adopted this leaf phenology scheme for oil palm modelling. 185

Allocation
In ORCHIDEE-MICT, carbon is allocated to leaf, sapwood and root in response to water, light and nitrogen limitation, except for a fixed allocation of 10% to reproductive plant tissues (Krinner et al., 2005). The allocation of carbon to phytomers was simulated following this framework. The allocation to fruit and branch sub-component for each phytomer was calculated as a fraction of aboveground sapwood allocation, whereas the allocation to leaves was unchanged. For each phytomer, the fraction 190 of aboveground sapwood allocated to branch and fruit components ( + , ℎ , where nphs is the total number of phytomers and https://doi.org/10.5194/gmd-2020-252 Preprint. Discussion started: 22 October 2020 c Author(s) 2020. CC BY 4.0 License.
i is the index of phytomer) is a function of phytomer age as follows (Eq. 1). This fraction is further adjusted by the oil palm tree age to account for yield increase with tree growth ( + , ℎ Eq. 2).
where + , and + , are prescribed values of minimum and maximum aboveground sapwood allocation fractions to branch and fruit, which is increased with tree age. ℎ (day) is the age of phytomer, and (yr) is the age of the oil palm tree. 1 , 2 and 3 are empirical coefficients (set at 0.265, 2 and 0.8 ; unitless), respectively, based on yield calibration against observations). All abbreviations and parameter values are shown in Table S2.
After fruit initiation started (second phase), the allocation strategy changes with more resources shifted to fruit than leaf and 200 the rate of fruit assimilation is accelerated (Corley and Tinker, 2015). This is represented by Eq. 1 with more carbon allocated to old and ripening phytomers to achieve the largest amount of yield. The further separation of branch and fruit ( + , ℎ ) and fruit fractions ( , ℎ ) follows a similar scheme, i.e. an increase with phytomer age to accelerate fruit accumulation (Eq. 3).
where , and , is tree age-specific value of minimum and maximum fruit allocation. , ℎ stands for the branch fraction in the total branch and fruit fraction ( + , ℎ ), and 1 is an empirical coefficient, set at 0.02 (unitless). The change of + , ℎ and , ℎ with phytomer age is shown in Figure 3.
The total phytomer allocation fraction is a sum of leaf, branch and fruit allocation: where and is the leaf and aboveground sapwood allocation fraction, respectively.

Fruit harvest
The default wood harvest in ORCHIDEE-MICT is based on the different forest age classes (implemented as CFTs). For each CFT, when the stem biomass reaches the prescribed maximum woody biomass of current CFT, it will move to the next CFT.
Wood harvest can start from any CFT by user's choice, and the default wood harvest sequence starts from the second youngest 215 CFT to the oldest one and back to the youngest until reaching the required harvest amount . Unlike wood https://doi.org/10.5194/gmd-2020-252 Preprint. Discussion started: 22 October 2020 c Author(s) 2020. CC BY 4.0 License.
harvest, oil palm fruit is produced in sequence and harvested regularly. Here we assume the harvested fruits were taken from the oldest phytomer before pruning. The duration between fruit initiation and harvest is prescribed ( (day), Table   S2), and fruits will be harvested after the phytomer age in the oldest phytomer reaches the . The harvested fruit biomass is then added to a new separate harvest pool. 220

Parameter calibrations for oil palm
Since most parameters vary across different PFT types, we systematically adjusted parameters related to photosynthesis, respiration, carbon allocation and morphology for oil palm according to the observed values from field measurement literature.
Some parameters are CFT-specific values in accordance with the tree age cohorts in the model. Details of the parameters for oil palm are summarized in Table S2. 225

Photosynthesis parameters
The photosynthesis module of ORCHIDEE-MICT is based on an extended version (Yin and Struik, 2009) of the Farquhar, von Caemmerer, and Berry model (FvCB model;Farquhar et al., 1980). Leaf age class is introduced to take into account the fact that the photosynthetic capacity depends on leaf age (Ishida et al., 1999). The maximum rate of Rubisco activity (Vcmax) is defined by the prescribed Vcmax25 and weighted leaf efficiency (erel, unitless: 0-1). The relative leaf efficiency (erel) is a function 230 of relative leaf age (Arel) where Arel is the ratio of the leaf age to the critical leaf age (Ageleafcrit), also known as leaf longevity which increases from a low initial value to 1 (reaching the prescribed optimal Vcmax25) for a given period and then decreased to a low level for the old leaves. This was modified by setting the minimum efficiency to 0 and at both leaf flushing and longevity based on observations of the leaf phenology of Amazonian TBE forest in another ORCHIDEE-MICT version with leaf cohorts, 235 ORCHIDEE-MICT-AP (blue dashed line) (Chen et al., 2020). However, unlike the natural TBE forest, the old leaves in the old phytomers of oil palm are probably more productive to sustain the high fruit amount because of the sequent growth, phytomer pruning and fruit harvest. Thus, erel for the old leaves of oil palms is maintained the same as the value in the default ORCHIDEE-MICT version (red line in Figure 4). We also adjusted Vcmax25 for each tree age class of oil palm according to the experimental evidence (Fan et al., 2015;Meijide et al., 2017;Teh Boon Sung and See Siang, 2018) (Table S2). Vcmax25 for oil 240 palm increases with tree age (from 35 to 70 μmol m −2 s −1 ) correspondingly to the increase of gross assimilation (Breure, 1988).
Another two important parameters for photosynthesis are maximum leaf area index (LAImax, controlling the maximum carbon allocation to leaf biomass) and specific leaf area (SLA). The observed maximum LAI varies from 4 to 7 m 2 m -2 across different genotypes, plant densities and soil types (e.g., peat) according to nine observation-based publications listed in Table S2, and LAImax was found to increase with oil palm tree age (Kallarackal, 1996;Kotowska et al., 2015;Legros et al., 2009). SLA, by 245 contrast, generally decreases with oil palm tree age from 0.0015 to 0.0008 m 2 g -1 C Legros et al., 2009;https://doi.org/10.5194/gmd-2020-252 Preprint. Discussion started: 22 October 2020 c Author(s) 2020. CC BY 4.0 License.
Van Kraalingen et al., 1989). We thus used a CFT-specific value which is close to the median values of LAI and SLA obtained from observational data (Table S2).

Respiration parameters
Autotrophic respiration (AR, including maintenance and growth respiration, MR and GR) in ORCHIDEE-MICT is based on 250 the work of Ruimy et al. (1996). MR is a function of the temperature and biomass for each plant part (Eq. 6-7) whereas GR is prescribed as 28% of the allocable assimilates for TBE tree PFT (Krinner et al., 2005). Field evidence shows that MR in gross assimilation of palm increases with oil palm tree age but MR per unit of tree biomass decreases (Breure, 1988). In total, AR represents 60-75% of GPP for oil palms (Henson and Harun, 2005). Based on this prior knowledge, we adjusted both the constant 1 in Eq. 7 and the fraction of GR in GPP ( ). The former parameter ( 1 ) increases with age and the latter does the 255 opposite ( ) (Table S2). The parameter values were calibrated to match the observation of GR/MR, AR/GPP and GPP.
Where j is the different plant part. 0 is prescribed for each plant part for each PFT. is the 2-m temperature/root temperature for above/belowground compartments. is the long-term (annual) mean temperature. is the second-260 degree polynomial dependency of . 1 , 2 and 3 are empirical coefficients.

Carbon allocation parameters
Carbon allocation to new leaves in the ORCHIDEE-MICT-OP was modified following the ORCHIDEE-MICT-AP by Chen et al. (2020) as described in 2.3.2. The leaf allocation ( ) is both related to the amount of sunlight available at the top of canopy and the light transmission of old leaves so that the is expressed as a function of higher shortwave downwelling 265 radiation ( ) and LAI of the old leaves as followings: where , and , are the prescribed values for minimum and maximum leaf allocation. 4 is the LAI of the oldest leaf age cohort 4. 1 , 2 and 3 are empirical coefficients, setting to be 0.45, 100 and 3 (unitless), based on the calibrations using observed NPP allocation among leaf, sapwood and fruit (Henson and Dolmat, 2003;Van Kraalingen et al., 270 1989).
The original leaf ( Where , and , is the prescribed values of minimum and maximum root allocation according to Kotowska et al. (2015). 1 is an empirical coefficient (= 0.95).
NPP partitioning between above and belowground sapwood biomass is a function of tree age. Older trees get more allocation to aboveground sapwood than younger ones (Krinner et al., 2005). In the default ORCHIDEE-MICT version, the values of 280 minimum and maximum NPP partitioning to aboveground biomass are constant. By contrast, observed oil palm gross assimilation increases with age (Breure, 1988), and most of the assimilates go into phytomer to sustain fruit production. In ORCHIDEE-MICT-OP, we adopted the original model equation of sapwood allocation to aboveground ( , ) increasing with age (Eq. 9) but adjusted parameters to match the observations.
Where , and , are prescribed tree age-specific values of minimum and maximum sapwood allocation to aboveground, which increases with tree age. is the oil palm tree age, and is the empirical CFT-dependent coefficients (Table S2).

Other parameters
Other adjustments of parameter values include morphological, phenological and turnover parameters. The maximum number 290 of phytomer (nphs) is set as 40 according to observations (Combres et al., 2013;Corley and Tinker, 2015). Given the phytomer initiation rate of 20-24 per year, the pruning frequency of twice a month and the number of phytomer (Combres et al., 2013;Corley and Tinker, 2015), the critical phytomer age ( ℎ ) is estimated to be around 600 to 720 days. Based on previous studies (Corley and Tinker, 2015;Fan et al., 2015;Van Kraalingen et al., 1989), the leaf longevity for oil palm is 600-700 days, shorter than the 730 days used for the default TBE tree PFT in ORCHIDEE-MICT. As a result, both the critical leaf age 295 (leaf longevity, ) and the critical phytomer age ( ℎ ) are set to be 640 days. The critical fruit age ( ), defined as the duration between the fruit initiation and harvest, is set as 600 days, that is, shorter than the critical phytomer age, allowing leaf senescence after fruit harvest.
After pruning, cut branches and leaves in a pruned phytomer are transfered to the litter pool. The additional leaf loss ( ) during the phytomer pruning was taken into account in the old leaf age class as follows: 300 Where is the leaf biomass for leaf cohort m, 1 is an empirical leaf loss coefficient.
In cycle. The natural tree mortality is thus not applicable for oil palms. In ORCHIDEE-MICT-OP, we assumed that there is no 305 natural mortality for the oil palm, and the mortality of oil palm is only controlled by the artificial management (clear-cutting every 25 years, Figure 5).

Site simulation setup
The 6-hourly 0.5° global climatic data, CRUNCEP v8 and the 0.08° global soil texture map were used as forcing data in the simulations (Reynolds et al., 2000). The vegetation cover of the 14 sites (Figure 1 and table S1) was all set to the oil palm PFT 310 with a coverage of 100%. Biomass boundary value for each age classes ( Figure 5) are prescribed for oil palm based on the prior knowledge from observation (Tan et al., 2014). When the total biomass reaches the lower boundary of the oldest tree age class (CFT6, Figure S3) and moves to CFT6, wood harvesting will be performed, and oil palm trees will thus be cut down.
New oil palms will be established in the youngest tree class (CFT1) for the next rotation cycle. Site simulations were run for 30 years which is consistent with the rotation duration of ~25 yrs and the climatic forcing for the period between 1986 to 2015 315 were used. Spin-up simulation was not performed since we didn't focus on the soil organic carbon and there is no feedback of soil carbon to plant growth in the model. Oil palm yields at maturity were calculated using the average values during 11-20 years for comparison. Fruit yields are converted to kg DM ha -1 yr -1 using a carbon ratio of 0.45. Figure 6 shows annual dynamics of observed and simulated LAI vs. tree age averaged over the 14 observation sites (black line). For each age, we collected observational LAI values from different field measurement studies and presented the medians and ranges (the red marker and error bar) in Figure 6. Since there are no continuous LAI measurements available (to the best of our knowledge), we combined single LAI measurements at a certain age from different studies. The simulated LAI increases from 0.3 to 5.3 in the first ~10 years, and then stays stable at the maximum value (5.5, Figure 6). The simulated LAI trajectory 325 can generally reproduce the trend from observations. Although simulated LAI ranges overlap with the ranges of LAI observations at most ages, some observations are not reproduced at Age 13 and Age 19 when the model achieved a stable and maximum LAI (Figure 6). This variability of LAI measurements reflects the use of different sites with different oil palm species and management practices. In the model, however, genotypes and practices are uniform. The detailed intra-annual variations of LAI, combined with leaf biomass and Vcmax for each leaf age cohort are shown in Figure S3 with significant 330 seasonality after merging the leaf phenology scheme from Chen et al. (2020). Compared to the ORCHIDEE-MICT version with no seasonality in LAI (dashed line in Figure S3a), the LAI of young leaves increases but decreases for old leaves during the canopy rejuvenation period (January to May, solid line in Figure S3a). The opposite behavior is shown in the rest of the year. Similarly, the default ORCHIDEE-MICT version shows no seasonality of leaf age and leaf photosynthetic efficiency in https://doi.org/10.5194/gmd-2020-252 Preprint. Discussion started: 22 October 2020 c Author(s) 2020. CC BY 4.0 License. different leaf age classes (dashed line in Figure S3b and c), while the seasonality of leaf age and leaf efficiency is successfully 335 captured in this version (solid lines in Figures S3b and c).

Productivity and fruit yield
The simulated GPP, NPP and fruit yield in comparison with field measurements are shown in Figure 7. For fruit yields, we collected six single-year observations at different sites for oil palm plantations aged from 10-15 yrs, expect 345 for one site where yield data cover ages 4 to 16. The observed oil palm yields at maturity vary from 13.0 to 22.1 t DM ha -1 yr -1 with a median of 15.0 t DM ha -1 yr -1 , and the simulated yields show a similar range of 12.2-21.4 t DM ha -1 yr -1 with a median of 16.9 t DM ha -1 yr -1 . Thus, simulated fruit yields show an overall good agreement with site observations with a NMBE of 6.1% (Figure 7c). There is only one site (Site 3) with available yield estimates for successive years (Figure 7d). It should be noted that it is not real observations but a fitted curve with oil palm age of yield data provided by the Malaysian Palm Oil 350 Board (MPOB) research station at Keratong (Tan et al., 2014). This yield-age curve shows a strong yield increase after Age 10 and even Age 25 (Figure 7d), which is against the field evidence that fruit yields for oil palms reach maximum at ~10 yr, stay relatively stable, and decrease after ~25 yr (Boo et al., 1994;van Ittersum et al., 2013). The reduction in yields after ~25 yr is also one of the reasons for clear cutting for next rotation. Still, we compared our simulated yields with that yield-age curve (Figure 7d). Simulated annual fruit yield at Site 3 is generally consistent with data during the first 9 years but lower than 355 the curve in the subsequent years, probably due to the uncertainties in the yield-age curve. Figure 8 shows the comparison of simulated biomass and time series with observations. The biomass here includes the developing fruit but exclude the harvested fruit biomass. Note that some sites have several observed values (Site 1, 2, 9 and 10 in Figure 8a) at different age and for biomass components e.g., total biomass (TB), above ground biomass (AGB) and below 360 ground biomass (BGB). A total of 13 biomass observations were collected at different age groups (3 in the young age group, 8 at maturity and the remained 2 for averaged biomass among several years, Table S1). Compared to the default ORCHIDEE-MICT version, simulated biomass by ORCHIDEE-MICT-OP is more consistent with observations ( Figure 8a). 10 out of 13 sites, are distributed close to the 1:1 line except Site 2 (TB at age 10), Site 9 (AGB at age 16) and Site 10 (AGB at maturity).

Biomass
The NMBE of oil palm biomass is 10.4% after excluding Site 9 with the largest bias, compared with the 156.7% by the default 365 https://doi.org/10.5194/gmd-2020-252 Preprint. Discussion started: 22 October 2020 c Author(s) 2020. CC BY 4.0 License. ORCHIDEE-MICT. We further compared the simulated above, below ground biomass and their ratio with observations ( Figure 8b). Similarly, the ORCHIDEE-MICT-OP version can better reproduce the observations than the default ORCHIDEE-MICT version. The NMBE for above and below ground biomass between ORCHIDEE-MICT-OP and observations are 12.1% and 55.3%, respectively. The ratio of AGB and BGB is reproduced of 1.7, which is much closer than the observation (1.1-3.0) compared with that of default ORCHIDEE-MICT (0.7-0.8). 370 There are only two sites (Site 3 and 12, Figure 8c and d) with time series of biomass. Similar to the fruit yields (Figure 8d) simulated biomass by ORCHIDEE-MICT-OP generally agrees with observed values but is higher in the first 18 years and lower afterward (Figure 8c). At Site 12, ORCHIDEE-MICT-OP simulated biomass is higher than observations for the whole oil palm life cycle. This is probability because Site 12 was covered by very deep peat soil (>3m) with a high soil water table and high C density and the potential impact on the oil palm production is not considered (e.g., different nutrient availability in 375 peat and mineral soil and palm leaning in peat soil which may cause the decline of yield). Also, the calibration is based on the observations from all sites and no calibration was applied for this site, which may cause the higher estimation. The NMBE is 16.2 % and 15.5% at Site 3 and Site 12. The default ORCHIDEE-MICT version largely overestimated the biomass at both sites (dashed line in Figure 8c,d).

Partitioning of GPP, NPP and Biomass 380
Comparison of oil palm GPP and biomass partitioning between simulations and observations is shown in Figure 9. Compared to the default ORCHIDEE-MICT version (grey bars), simulated results from the ORCHIDEE-MICT-OP version (black bars) are closer to the observations (red bars, Figure 9). GPP is partitioned to GR, MR and NPP whereas NPP is further divided into allocation to stem and frond, root and fruit (Figure 9a). The simulated growth and MR fraction in GPP ranges from 17.1-28.8% and 28.1-54.3% respectively, which is comparable with observations (21-31% and 34-44%) from Henson and Dolmat (2003). 385 The simulated fraction of autotrophic respiration in GPP (60.87%) is also consistent with the observed fraction (60-75% (Henson and Harun, 2005)). In the simulation by ORCHIDEE-MICT-OP, stem and leaf (median of 18.9% in GPP) occupies the largest parts of NPP, followed by fruit allocation (17.5%) and root allocation (2.8%). The differences between the simulated NPP fraction for stem and leaf, root and yield by ORCHIDEE-MICT-OP and observed fraction are 10.9%, -1.4% and -2.0%, respectively, indicating a good representation of NPP allocation to different biomass components in the new model. 390 Simulated partitioning of biomass by ORCHIDEE-MICT-OP is closer to observations (Breure, 1988;Henson and Dolmat, 2003;Tan et al., 2014) than the default ORCHIDEE-MICT version (Figure 9b). The simulated leaf and root and other organs (stem, fruit and branch biomass) proportion of total biomass varies between 51.7-75.1%, 14.7-32.4%, and 8.5-16.0%. The simulated fraction to other organs is higher (14.7%) than observations, and correspondingly it is lower for leaf (-6.1%) and root (-5.6%) fractions, the improvements reaches 18.8%, 13.0% and 6.2% compared to the biases in the default ORCHIDEE-395 MICT. Note that the proportion of fruit bunch and branch of a phytomer is not separated but added in the stem proportion because most of the studies presented fruit and branch biomass fraction as a part of stem biomass (Henson and Dolmat, 2003; https://doi.org/10.5194/gmd-2020-252 Preprint. Discussion started: 22 October 2020 c Author(s) 2020. CC BY 4.0 License.
Van Kraalingen et al., 1989). Also, the time and frequency of collecting fruits and measuring biomass are usually not synchronous. There is only one field study showing that the phytomer (fruit and branch) fraction varies between 5.0-14.5% of the total biomass after fruit harvest (Breure, 1988), which is comparable with the simulated median proportion of 14.4% by 400 ORCHIDEE-MICT-OP.

Phytomer development
Growth of phytomers during the life cycle (initiation, fruit development and productive phases) of oil palm is presented in Figure 10. Figure 10a and b show the fruit and branch growth in single phytomer (8 in 40 phytomers were shown for a better visualization), while figure 10c is the total biomass for all the 40 phytomers as a sum of leaf, branch and fruit components. 405 The initiation phase roughly corresponds to the oil palm tree age between 0 to 3 without any fruit production. Subsequently, age 3-10 is the fruit development phase. After 10 years old, oil palm reaches the productive phase with the maximum and steady fruit yields. This phenological characteristics is consistent with the oil palm development observed in previous studies (Sunaryathy et al., 2015). Some study even shows the productive phase can start as early as ~7 year old (Henson and Dolmat, 2003). 410 Biomass of leaf and branch of all the phytomers starts to increase after planting ( Figure 10c) and reaches about 211.3 and 28.6 gC m -2 at age 3. The fruit production and harvest begin after entering the fruit development phase (>3 years old) (Figure 10a), whereas the total fruit biomass increases rapidly to 367.6 gC m -2 yr -1 at age ≈ 10. From age 3 to 10, phytomer biomass increases with a stair-step shape, and fruit and branch biomass slightly decline when moving from one tree age class to the next older class. This is because values for some parameters (e.g., Vcmax and LAImax, Table S2) are different among the CFT 2-4 in the 415 fruit development phase. For example, LAImax increases from 3.5 in CFT3 to 4.5 in CFT4. In the ORCHIDEE framework, biomass will preferentially allocate to leaf to reach LAImax in order to grow more leaves to increase GPP and then allocate to other biomass parts when LAI reaches LAImax (Krinner et al., 2005). Therefore, when oil palms move from CFT3 to CFT4, the increased LAImax drives more biomass going to leaf ( Figure 10c) and less to fruit and branch at the beginning of CFT4, resulting in the small decline in the fruit and branch biomass. We acknowledge that this model behavior may contradict the 420 reality, but the small magnitude and short duration of declining (Figure 10c) may have little impact on the modeling results.
At the productive (maturity) phase after age 10, the average leaf, fruit and branch biomass are 683.8, 424.0 and 64.8 gC m -2 , which consists of 58.3%, 36.1% and 5.5% of the total phytomer biomass (40 in total), respectively.

Model performance before and after oil palm implementation 425
Based on the default ORCHIDEE-MICT version and the leaf age cohort scheme in the ORCHIDEE-MICT-AP version, the oil palm PFT has a new phytomer organ and a yield harvest pool ( Figure S2), with other model parameters recalibrated. The new ORCHIDEE-MICT-OP version allows for simulating oil palm morphology, phenology, biomass growth and yields. We https://doi.org/10.5194/gmd-2020-252 Preprint. Discussion started: 22 October 2020 c Author(s) 2020. CC BY 4.0 License. evaluated the LAI, GPP, NPP, yields and biomass of oil palm in ORCHIDEE-MICT-OP using available observations from previous field measurement studies (Table S1). 430 In the default ORCHIDEE-MICT version, oil palm is taken as TBE tree PFT, which causes biases in the simulation. For example, it is impossible to realize regular fruit harvest and phytomer dynamics in the default ORCHIDEE-MICT version without the phytomer structure and the fruit harvest pool. The introduction of phytomer structure and the sequential developing processes allows for reproducing variable developmental stages for each phytomer including the initiation, fruit production, harvest and pruning in the model. Besides, the modification of carbon allocation scheme improves allocation of the assimilated 435 carbon and partitioning of biomass pools (Figure 9). Oil palm trees have special physiological characteristics which are different from other tropical forests. The evolution of physiology with age is implemented by new tree age-specific parameterization scheme based on the tree age cohort module of ORCHIDEE MICT. Carbon assimilation is accelerated with increasing oil palm age. Carbon allocation to phytomer shifts more resources to fruit than leaf and branches as fruits mature.
Consistent with observations, the fruit yields also show an increasing from young to old trees. To our best knowledge this age-440 based allocation dynamics for oil palm has not yet been implemented in the previous LSMs aiming to simulating oil palm biophysical variables. The leaf age cohort-based phenology scheme from ORCHIDEE-MICT-AP was also adapted for oil palms to improve the seasonality of leaf and photosynthesis ( Figure S3). This process was not included in any previous oil palm models either. Moreover, the calibration for age-specific parameters is based on the 14 observation sites with variable climate and soil conditions rather than two limited sites used in the previous oil palm model (Fan et al., 2015). Therefore, our 445 parameterizations of oil palm (Table S2) can also be applied in other LSMs.

Uncertainty in the model
Although the simulation of oil palm shows a significant improvement in the new model, there are some limitations in this version. The growth of oil palm is simplified to be incorporated into the model structure. For example, we assumed a constant maximum phytomer number of 40 for each oil palm through its whole life cycle. However, the expanded phytomer number 450 may decrease with age according to some studies, and the maximum number is lower than the actual value in some areas (e.g., 32) (Corley and Tinker, 2015). The maximum number of phytomer is externalized as an input parameter in the model, making it flexible to be changed by users' choice. Some factors related to oil palm yields such as the gender of inflorescence and the rate of inflorescence abortion are not considered because of the limited understanding of underlying mechanisms (Breure and Menendez, 1990;Henson and Mohd, 2004). Instead, a simplified structure of one phytomer carrying one fruit bunch is used. 455 Also, considering the oil palm is a highly manged plantation unlike natural forest, some rigid parameterization is adopted such as fruit harvest interval, phytomer pruning interval and leaf longevity. For example, the phytomer initiation in sequence is determined by a fixed time interval (16 days), which is based on the number of initiations per year (12-24), the regular pruning practices (about twice a month) and the fixed number of total phytomers. The assumption of these fixed input parameters (also can be changed as external variables) in our model is thus a compromise between the plant growth and human management 460 https://doi.org/10.5194/gmd-2020-252 Preprint. Discussion started: 22 October 2020 c Author(s) 2020. CC BY 4.0 License. practices. A previous study also used the period of thermal time (Fan et al., 2015) to identify the starting point of two subsequent phytomers.
The accessibility and data sources of observations also vary from site to site, which influence the calibration of parameters and the evaluation of model performance. Without direct annual observations for parameters related to LAI and autotrophic respiration, some age-specific parameters are empirically calibrated based on multiple observations like GPP, NPP and 465 biomass. The observations used for calibration and evaluation such as yields, biomass and GPP also vary from genotypes, management practices, and measurement methods. For example, the annual fruit yield data in Site 3 (red line in Figure 7d) is a fitted curve using fruit yields from a nearby research station (Tan et al., 2014) while some others are measured fruit weight after fruit harvest every time (Henson 2003). Destructive and non-destructive based methods were used to obtain the AGB for different sites, and different allometric equations applied in the non-destructive based method may cause up to 10% biases 470 (Corley and Tinker, 2015). In site 6, the simulated GPP by ORCHIDEE-MICT-OP is 50% higher than the observed value. The mismatch between model and observation may also be caused by the uncertainty in observations or non-resolved soil fertility effects. Specifically, since the model can generally capture NPP (simulated NPP = 1700 gC m -2 yr -1 at Site 1; only Site 1 has both GPP and NPP observations), and the proportion of autotrophic respiration in GPP is 60-75% (Henson and Harun, 2005), the estimated GPP at site 1 should be 4256.6-6810.5 gC m -2 yr -1 , much higher than the observed value of ≈ 3360 gC m -2 yr -1 . 475 Moreover, yield of oil palm usually ranges from 587 to 996 gC m -2 yr -1 , so the low observed GPP at site 1 may not be consistent with this yield range. Factors such as genotypes, management practices (excepted fruit harvest and phytomer pruing) and plantation scales that influence oil palm biomass and fruit yield are not fully included in the model, and thus it is impossible to perfectly reproduce the all site-level observations using our model. The reported fruit yields of different genotypes vary from 114.4-112.2 kg plant -1 yr -1 to 81.7-98.5 kg plant -1 yr -1 in Kandista and Batu Mulia (Lewis et al., 2020), and leading 480 plantation companies in Indonesia and Malaysia have achieved average fruit yields of 173.7 kg plant -1 yr -1 (Donough et al., 2009). The amount and types of fertilizers used in oil palm plantation also vary from site to site. In some area, applied fertilizer amount is according to the leaflet nutrient contents while regular fertilization was applied in some other places Legros et al., 2009). In the current ORCHIDEE-MICT version, however, nitrogen and phosphorus cycles are not explicitly included, limiting the implementation of fertilization effects on plant growth in the model. The scales of plantation 485 also impact oil palm biomass and yields due to the differences in managements (e.g. dedicated managements in the large industrial plantation and extensive practices in smallholders). Similarly, oil palm planting density ranges from 120-200 would also cause biases in observed biomass and yield when converting the unit from ton per tree to ton per hectare.

Implication and application of ORCHIDEE-MICT-OP
The newly developed ORCHIDEE-MICT-OP can be a useful tool to predict future oil palm yields, simulate LUC carbon 490 emissions and estimate impact on ecosystem services. Malaysia and Indonesia experienced the highest oil palm expansion (3.8 and 9.7 million ha) over the world from 2001 to 2016 (Xu et al., 2020).The drainage and replacement of peatland (3.1×10 6 ha, https://doi.org/10.5194/gmd-2020-252 Preprint. Discussion started: 22 October 2020 c Author(s) 2020. CC BY 4.0 License. 27%) in Malaysia and Indonesia by oil palm expansion turned this carbon-rich region to a carbon source (Miettinen et al., 2016). It is thus important to simulate the carbon budget and calculate the carbon changes after oil palm expansion. Previous studies calculated the potential carbon emissions from forest conversion by oil palm using a uniform carbon density value 495 without considering spatial heterogeneity and temporal variations (Carlson et al., 2013;Cooper et al., 2020). In reality, the biomass loss from deforestation is fast but soil carbon change may take a long time. Based on the framework of gross land use changes, the grid-based ORCHIDEE-MICT-OP could thus contribute to the quantification of spatial and temporal dynamics of LUC carbon emissions from oil palm expansion. Given the high rate of oil palm expansion in Malaysia and Indonesia, it is also in urgency to evaluate the potential impacts on the water and energy cycles in tropics (Fan et al., 2019). Further 500 modifications of oil palm-specific canopy structure can help to understand the biophysical changes after oil palm conversion.
Moreover, although the expansion of oil palm cultivation is seen as a severe threat for the conservation of rainforest and swamp areas and their associated ecosystem services (Koh et al., 2011;Koh and Wilcove, 2008), oil palm is admittedly the most productive oil crops with 3-5 times yields of other oil crops. To replace oil palm, much more lands will thus be needed for other oil crops to produce the same amount of oil production. This is also in dispute among policy-makers. The model with 505 explicit representation of oil palm and calibration using site-level data can provide spatial oil palm biomass density, yield and water consumption in future land use scenarios and would help to identify the most suitable areas for growing oil palms as well as to contribute to the policy formulation for the sustainability of oil palm plantation.

Conclusion
In this study, oil palm was incorporated in the ORCHIDEE-MICT LSM as a new PFT by introducing the phytomer structure 510 and a fruit harvest pool, modifying carbon allocation and implementing a systematic parameterization scheme. The leaf seasonality represented by different leaf age cohorts was also merged into this model. The developed MICT-OP version performs reasonably well in simulating photosynthesis, carbon allocation, biomass stock and fruit yields at multiple observation sites. Compared with the default ORCHIDEE-MICT version, ORCHIDEE-MICT-OP shows improved performance of GPP partitioning, NPP allocation and biomass components. The new oil palm version, parameterized with age-515 specific parameters, generally captures temporal dynamics of oil palm biomass and yields. Implementation of more management practices (e.g., fertilization and irrigation) and parameterization of biophysical variables are further needed. Generally, our model improves the representation of oil palm in LSMs and further applications of ORCHIDEE-MICT-OP include but are not limited to regional carbon budget and water demand estimation, yield prediction and the sustainable development of oil palm industry. 520 https://doi.org/10.5194/gmd-2020-252 Preprint. Discussion started: 22 October 2020 c Author(s) 2020. CC BY 4.0 License.

Code availability
The source code for ORCHIDEE-MICT-OP revision 6850 is available via https://forge.ipsl.jussieu.fr/orchidee/wiki/GroupActivities/CodeAvalaibilityPublication/ORCHIDEE-MICT-OP-r6850 (last access: 23 July 2020; Xu, 2020, the doi will be updated later). This software is governed by the CeCILL licence under French law and abiding by the rules of distribution of free software. You can use, modify, and/or redistribute the software under the 525 terms of the CeCILL licence as circulated by CEA, CNRS, and INRIA at the following URL: http://www.cecill.info.
Competing interests. The authors declare that they have no conflict of interest.