Leaf removal effects on light absorption in virtual Riesling canopies ( Vitis vinifera )

Leaf removal is a standard vineyard management technique to influence grape composition or to reduce disease pressure; however, the timing and intensity of leaf removal is a widely discussed issue. The interplay of different goals and effects over time does not make experimental studies any easier. To gain insight into positive and negative consequences of leaf removal on grapevine development, a first step can be to study how leaf removal affects the canopy’s light absorption using a dynamic model approach. Functional–structural plant models combine canopy architecture with physiological processes and allow analysing canopy interaction with the environment with great topological detail. The functional–structural plant model Virtual Riesling simulates Riesling vines in a vineyard set-up depending on temperature and plant management. We implemented leaf removal and applied this method in or above the bunch zone to compare the light absorption in canopies. Leaf removal in the bunch zone led to greater loss of absorbed light, but canopies of both scenarios could compensate for most of the loss during the simulation time frame. Compensation was mainly driven by lateral leaves closing the gaps induced by leaf removal and by leaves in the proximity of the leaf removal zones, re-exposed to light. Results showed similar effects as observed in in vivo studies; hence, we suggest extending these simulations to investigate other effects linked to light distribution such as berry sunburn. Simple modifications of implemented leaf removal techniques also allow for test-ing different application scopes and their impact on canopy light absorption.


IN TROD UCTION
Grapevines (Vitis vinifera) are subject to different management practices over the course of a year to control yield and fruit quality (e.g. Diago et al. 2010;Palliotti et al. 2011;Poni et al. 2018). Management decisions influence the growth of vines strongly and can even have an effect across seasons (Diago et al. 2010;Tóth 2020;Wang et al. 2020). Many practices, such as shoot positioning, pruning or leaf removal, affect the canopy architecture, whilst the canopy plays an important role in the interaction of grapevines with the environment and affects the microclimate Friedel et al. 2015;Poni et al. 2018;Verdenal et al. 2019). Leaf size, shape, position and orientation influence sunlight and wind penetration as well as temperature and humidity inside the canopy (Smart 1985;Zoecklein et al. 1998;Reynolds and Heuvel 2009;Deloire 2012;Schmidt and Kahlen 2018). These conditions affect the risks of diseases and physiological disorders, such as fungus or berry sunburn. If management practices alter the canopy architecture, conditions within the canopy and thus risks of diseases can change abruptly (Gambetta et al. 2021). One such practice with a sudden impact is leaf removal.
Leaf removal is a technique in viticulture to influence and protect grape development. It is traditionally applied to improve grape composition and to reduce disease pressure by changing the microclimate around the bunches (Poni et al. 2018;Verdenal et al. 2019). Leaf removal can also be applied to control yield and to delay berry ripening (Palliotti et al. 2011;Poni et al. 2013;Stoll et al. 2013;Verdenal et al. 2019;VanderWeide et al. 2020). Details of leaf removal techniques, like timing, severity (number of removed leaves) and location in the canopy, differ substantially between studies. Leaf removal 2 • Bahr et al. is applied between fruit set and véraison (onset of berry ripening) to improve light penetration and air circulation in the bunch zone (Diago et al. 2010;Poni et al. 2018) and at prebloom to control yield as well as bunch rot disease (e.g. Diago et al. 2010;Komm and Moyer 2015;Poni et al. 2018;VanderWeide et al. 2021). For details on developmental stages of grapevine, we refer to, e.g., Dry and Coombe (2004). In most of the above-mentioned studies, basal leaves in the bunch zone were removed, reducing the main sources for assimilates to decrease fruit set. In contrast, leaf removal above the bunch zone after flowering aims at delaying maturation (Poni et al. 2013;Stoll et al. 2013) and leaf removal in apical zone or bunch zone after véraison may improve grape composition (Poni et al. 2018). The number of removed leaves ranges from three basal leaves (e.g. Verdenal et al. 2019;VanderWeide et al. 2020) up to all leaves around the bunches (e.g. Gambetta et al. 2019). In certain cases, only east side or sun-exposed leaves were removed (Kurtural et al. 2013;Schüttler et al. 2015;Yu et al. 2016;Gambetta et al. 2019;Wang et al. 2020). Some authors suggested to reduce leaf area proportional to canopy size depending on the respective management (Stoll et al. 2013;Poni et al. 2018). Although excessive leaf removal may pose a risk to grapevine quality, practical recommendations regarding the scope of leaf removal are sparse.
The more leaves are removed on the sun-exposed side of the bunch zone, the more berries might get sun-burned, since berry sunburn is directly related to sudden sun-exposure (Gambetta et al. 2021). Leaf removal can also lead to adverse source limitation, an effect which can have consequences across seasons (Verdenal et al. 2019;Wang et al. 2020). Such threats are expected to increase due to climate change and adjustments of leaf removal techniques might become more important in the future (e.g. Santos et al. 2020). One such adjustment relates to the timing of leaf removal. Early leaf removal in the bunch zone can allow berries to better adapt to sunlight and, thereby, reduce their susceptibility to sunburn (Gambetta et al. 2021). An adjustment can also be achieved by changing the location of leaf removal. For example, Hayman et al. (2012) explicitly suggest removing only basal leaves instead of leaves in the bunch zone to protect berries from exposure to sunlight and sunburn. Overall, these findings indicate that characteristics of leaf removal applications, such as timing, number of removed leaves and location in the canopy, may have negative effects on grapevine performance as well. To better understand the positive and negative consequences of leaf removal over time, a first step can be to systematically analyse the effect of leaf removal on light distribution within the canopy by the help of a dynamic computer model. Since canopy architecture and its topology is important, functional-structural plant models seem useful to investigate such a complex system (Guo 2006).
Functional-structural plant models combine the three-dimensional architecture of a plant with physiological processes on the basis of plant organs and their topology (Vos et al. 2009). They allow addressing each single simulated plant organ during run time, e.g. for modulating organs depending on their physiological or topological (i.e. location) state. Schmidt et al. (2019) developed the dynamic functional-structural plant model Virtual Riesling, which simulates the growth of grapevines of cultivar Riesling based on descriptive growth functions, management techniques and temperature conditions. The model can simulate daily canopy growth under environmental conditions as they exist at the experimental site (Geisenheim University, Germany), taking into account i.a. growth variability and management such as vertical shoot positioning (VSP trellis system). Recently, Virtual Riesling has been coupled with a light model (cf. Evers and Bastiaans (2016) and Evers et al. (2010)) to simulate diffuse and direct light interacting (absorption, reflection, transmission) by the virtual grapevine canopy (Bahr et al. 2020), but not affecting growth. The current version of the model Virtual Riesling can now simulate a vineyard, measure light interception and manipulate plant organs on a daily basis, such as primary and lateral leaves, i.e. leaves grown at primary and lateral shoots, on a daily basis. Therefore, the model is suitable to be extended for leaf removal applications in virtual vineyards.
The aim of this study was to examine the effect of leaf removal on light distribution using the model Virtual Riesling. First, the model was extended by a method for leaf removal that resembles practical application techniques in vineyards. Then, we simulated grapevine growth within a virtual vineyard applying leaf removal scenarios that differed in application height. The effect of leaf removal on light distribution was evaluated by the temporal development of absorbed photosynthetic radiation over a period of 3 weeks after leaf removal by considering different canopy zones and differentiating between primary and lateral leaves.

Virtual Riesling
The functional-structural plant model Virtual Riesling simulates the dynamic growth of Riesling grapevines from bud burst on the cane to end of flowering in daily time steps (Schmidt et al. 2019). Currently considered plant organs include the cane, the primary and lateral shoots and the leaves at both shoot levels. The architecture is influenced by the growth behaviour of the vine, environmental conditions and management practices. The model includes various stochastic components, i.e. bud burst or shoot orientation in space, to mimic natural variability in the model. It is based on digitized data of vines grown at the experiment site of the VineyardFACE project at Geisenheim University, Germany (Wohlfahrt et al. 2018;Schmidt et al. 2019). The code is written in the programming language XL and the interactive modelling platform GroIMP (v.1.5) was used for model development (www.grogra.de/software/groimp) (Kniemeyer and Kurth 2008).
The following features of Virtual Riesling are based on grapevine and trellis set-up in the VineyardFACE project. The cane-supporting wires are set between posts 0.70 m above-ground. Eight winter buds are placed on the cane in slightly varying positions to mimic an alternate distichous phyllotaxis including natural variability. Shoot positioning is applied four times with two wires, each being attached on one side of the posts. The distance between a pair of wires is 10 cm, i.e. the diameter of posts. Thus, during simulations, virtual shoots are realigned four times within a 5-cm distance from the row's centre to simulate vertical shoot positioning practice at heights of 0.90, 1.10, 1.40 and 1.70 m above-ground.
Canopy development follows a thermal time approach with a base temperature of 10 °C (Schmidt et al. 2019). Bud break at the cane is modelled with a variability of 12 °Cd. Phytomeres then appear approximately every 22 °Cd on both primary and lateral shoots and grow Light absorption in virtual Riesling canopies • 3 with thermal time depending on their rank. Lateral shoot bud break is modelled with a probability, depending on the phytomere rank, with a maximum asymptotic probability of 98 % approximately reached at rank 13. Before a lateral shoot grows, a thermal delay must be overcome, which also depends on the rank. This delay lasts at least 67 °Cd and contains a stochastic component (standard deviation (SD) of 28 °Cd). Using Geisenheim temperature data for the 2018 season lateral bud break in the simulations is peaking at approximately day of the year (doy) 155 to 160 (Schmidt et al. 2019, Fig. 13). A shoot-related alternate dichtchous phyllotaxis controls petiole orientation. Initial petiole angle to the parent shoot is approximately 45° before they are rotated to match observed angles to the horizontal plane of approximately 40° and 36°, respectively. Petioles on primary shoots grow almost perpendicular to the row, with a variation induced by initial shoot orientation, obstacle avoidance and shoot positioning. The leaf 's midrib follows the petiole orientation, but the horizontal angle of the leaf is altered to match observations. For leaves at primary shoots the horizontal angle is related to leaf size and rank with a trend of smaller leaves being more horizontal than larger leaves. For leaves at lateral shoots the horizontal angle is fixed to approximately −25°, i.e. slightly downward facing. For more details we refer to eqs. 3-17 from Schmidt et al. (2019).
During one virtual day different model steps for plant development are applied in a specific order. A simulation starts with the appearance of new plant organs and growth depending on temperature conditions. Subsequently, vertical shoot positioning is applied when shoots are long enough for defined heights.
Virtual Riesling allows setting up vines in multiple rows with defined row and plant distance to simulate a vineyard. An example of a virtual vineyard simulated using Virtual Riesling is given in Fig. 1. In this set-up rows were aligned along a north-south axis and the cane was always bend towards the south of the trunk, resembling the vineyard in the Geisenheim VineyardFACE project.
Recently, Virtual Riesling has been coupled with a light model (cf. Evers and Bastiaans 2016) to simulate diffuse and direct light interacting within the virtual grapevine canopy, but not affecting growth (Bahr et al. 2020). A dome of 72 light sources for diffuse light and 24 light sources representing the daily course of sun are placed around the centre of the virtual crop (see Supporting Information- Fig.   S3; Supporting Information-Tables S1 and S2 for details on an exemplary configuration on doy 153). In correspondence to the VineyardFACE location the latitude is set to 50°. Other parameters describing the incoming light are taken from the studies of Evers and Bastiaans (2016) and Evers et al. (2010). Ray tracing simulations run using 20 million rays as a compromise between computational time and accuracy (see also Henke and Buck-Sorlin 2017). The interaction of rays with leaves of the canopy includes optical properties for grapevine leaves derived from literature (Cabello-Pasini and Macías-Carranza 2011). Leaf surface reflectance and transmittance to photosynthetic active radiation (PAR, 400-700 nm) are set to 0.14 and 0.15, respectively. During a simulation daily absorbed PAR (PAR abs , μmol s −1 ) is estimated at leaf scale (m 2 ), where PAR is derived from simulated incoming global radiation according to the approach from Evers and Bastiaans (2016). The conversion originates at the light sources where emitted daily incoming global radiation (GR, MJ m −2 ) is translated to photosynthetically active photon flux density (PPFD, μmol m −2 s −1 ) following Eq. (1), considering the current daylength (s) (Spitters et al. 1986, eq. 17), conversion factors for the units and the assumed PAR fraction (0.55).
The simulated course of incoming radiation is given in Supporting Information- Fig. S2, covering a range of approximately 14 to 19 MJ m −2 day −1 during the time frame of the simulations (also see Supporting Information- Table S3). At each leaf PAR abs is then calculated from absorbed PPFD abs by multiplication with the respective leaf area (A Leaf , m 2 ; Eq. (2)): For the results we converted PAR abs to the unit MJ day −1 reverting the calculation from Eq. (1) by replacing PPFD with PAR abs (μmol s −1 ) from Eq. (2).

Modelling leaf removal
The goal of this study was to simulate early leaf removal in a vineyard on the east side of canopies, i.e. the morning side, within or above the bunch zone and to evaluate the impact on light absorption. Therefore, we needed to develop a model that allows removing virtual leaves related to the east side of the canopy at different heights. Realizing leaf removal at different heights or zones in the canopy was straightforward. Using the leaf base position as a reference point for leaf location, leaves within a certain height can be selected and removed from the scene. In order to mimic the local expert opinion on removing leaves only from the east side, we wanted our model to consider, whether a leaf might contribute to the east side canopy. It should not only remove leaves located at the east side of the canopy, also leaves facing east and potentially extending to the east side should be removed. We decided to keep leaves which expand from east to the west side and to remove leaves which expand from the west side to the east, but only if origins of such leaves were not located further away than 5 cm from the centre of the row. Due to the dimensions of leaves in the model this local restriction assures that at least parts of such leaves cross the centre of the row. This part of the leaf removal model assigning leaves to either the east or the west side of the canopy can be summed up in a mathematical expression as follows: considering the location (loc) and cardinal direction (dir) of the leaf and the signed distance from the leaf 's location to the row centre plane, where the sign is based on the distance to the opposite side of the leaf removal target side, in this study the distance to the west side ( D west (loc, row.center), m). This function is combined with a defined height range, so that only east side leaves whose base is located within that range are subjected to a leaf removal event. Furthermore, we also used Eq. (3) in the results to assign leaves to the east and west side of the canopy, rather than just using their location in relation to the row centre.

Virtual leaf removal scenarios
Based on information by local experts and their experiences in the field, possible target zones for leaf removal in the canopy were defined. The general aim was to remove leaves either within or above the bunch zone. This aim required an estimation about where bunches are located in the canopy of vines growing on the VineyardFACE site in Geisenheim. The bottom and top height of the bunch zone were estimated to be at 0.7 m (height of cane) and 1.1 m and these values were taken to define the zone for the first leaf removal scenario (S1). To remove leaves above the bunch zone, scenario two (S2), application heights were set to the range between 1.1 and 1.4 m (Fig. 2). The control scenario was defined as a case without applying leaf removal (control).
Depending on plant growth the number of leaves within these heights changed over time. Therefore, the amount of light which could be absorbed in between these heights changed over time as well. In order to make meaningful comparisons between light absorption of the two leaf removal scenarios, the application day was selected based on the amount of absorbed light in these zones in canopies where no leaves were removed (Fig. 3). On Day 153 of the year 2018 the light absorption was nearly the same in both zones and chosen to be the application day. This day was the second of June 2018, 1 day before the estimated growth stage full bloom (BBCH 65) and 5 days before end of blooming in Geisenheim Figure 2. Leaf removal zones between 0.7 and 1.1 m (bunch zone, east) above-ground in scenario one (S1) and between 1.1 and 1.4 m above-ground in scenario two (S2): leaf removal zones are illustrated as grey areas on a digitized plant (without leaves) (A); three exemplary plants show canopies before (B) and after leaf removal for S1 (C) and S2 (D) with a view of the east side and with a view along the row.
In summary, the developed leaf removal method and the described scenarios remove specific leaves in the virtual canopy derived from practical application techniques in vineyards at Geisenheim University.

Virtual leaf removal simulations
Simulations to study the effect of leaf removal on light absorption were limited to the year 2018, using temperature data from a weather station in Geisenheim located at the experimental site. A simulation started with bud burst on the cane and stopped at an average height of 2.3 m, just before first pruning would have been applied (cf. Schmidt et al. 2019), approximately covering a period from doy 103 to doy 172. Leaf removal in the model Virtual Riesling, if activated, is applied after the appearance and growth of plant organs and after shoot positioning, but before simulating light rays for that simulation day. Simulations of the 5 × 5 vineyard (Fig. 1) were performed 112 times for each variant, i.e. the control and the two leaf removal application scenarios, to analyse light absorption of the resulting 1008 inner plants. It is 1008 inner plants per scenario, as each virtual vineyard leads to 3 × 3 = 9 inner plants and this multiplied by 112 (number of repetitions per scenario) is 1008. This number was selected to assure robust estimates of the average light absorption for stochastic simulations (Byrne 2013;Schmidt et al. 2019). Light absorption data in combination with topological leaf data were analysed with R (v.3.6.3) (R Core Team 2020) and visualized using the ggplot2 package (v.3.3.3.9000) (Wickham 2016). To capture the variability within the simulation results we estimated the arithmetic mean and the 50 % highest density interval (HDI) (R package HDInterval, v.0.2.2) in our statistical analysis. For robustness, both estimates are based on 1008 single plants per scenario. We provide the mean as a point estimate for a best guess given the data. The 50 % HDI represents the narrowest interval containing the most probable, i.e. frequent, values. Overlaps of 50 % HDI indicate that there is a likely chance of similar observations between scenarios. The functional-structural plant model output includes data for all individual leaves of a plant (size, location, parent shoot, etc.). These data allowed us to calculate summary statistics on leaf areas (mean and SD) and number of leaves removed.

R E SULTS A ND DISCUSSION
First, we take a closer look at the effects of the two different leaf removal scenarios on the canopy itself. The second part focuses on how leaf removal affects the leaf light absorption of a grapevine plant.

Leaf removal intensities
We conducted simulations to study two leaf removal scenarios, where leaves are removed on the selected day of leaf removal (doy 153) from the east side of the canopy only. To compare expected leaf removal intensities, i.e. number of removed leaves and respective leaf area, we used data from the control scenario where no leaves were removed at doy 153. For S1 we estimated an average removal of 3.59 ± 0.94 (mean ± SD) leaves per shoot, while in S2 on average 1.91 ± 0.77 were removed. The average leaf area of removed leaves in S1 and S2 was 97.53 ± 46.59 cm 2 and 116.81 ± 35.9 cm 2 , respectively. Projecting this to the entire canopy of a single plant, from a total leaf area of 1.06 ± 0.03 m 2 S1 takes out 0.28 ± 0.03 m 2 and S2 0.18 ± 0.02 m 2 . This approximately equals 26 % of the total leaf area for S1 and 17 % for S2. Such values are not uncommon for practical leaf removal applications, for instance, Komm and Moyer (2015) compared prebloom, bloom and 4 weeks postbloom leaf removal applications on Riesling shoots with total leaf areas removed in the range of 22.4-59.6 %, while removing leaves from both sides of the canopy. In the study of Diago et al. (2012) leaf areas of Tempranillo were reduced by ca. 30 % at prebloom. A more closer look at the leaf level showed that in both scenarios removed leaves were mainly primary leaves. In S1 the average number of primary leaves removed from a shoot was 3.28 ± 0.68 compared to only 0.31 ± 0.60 lateral leaves, whereas in S2 1.83 ± 0.71 primary leaves and 0.09 ± 0.30 lateral leaves were removed. This is not surprising, as lateral shoot and leaf growth in the model Virtual Riesling set in at approximately doy 145 in 2018 (Schmidt et al. 2019). These data suggest that the leaf removal intensity is higher in the bunch zone than above the bunch zone. This is related to the fact that the leaf removal zone in S1 covers a total height of 0.4 m, while S2 only spans 0.3 of the canopy. However, as already shown in Fig. 3, at the applied leaf removal day, the removed leaves would have contributed equally to the total absorbed PAR of a single plant.

Absorption of PAR
To evaluate the effect of leaf removal on light absorption in Riesling canopies, we now compare data from the control and the two leaf removal scenario simulations over the covered time period. In Fig. 4 light absorption curves of the control and the two scenarios are separated into east and west side of a row. Light absorption in the control increases continuously over the full period of the simulation. This can be explained by steadily increasing canopy height, leaf area and incoming radiation ( Fig. 5; see Supporting Information- Fig. S2). As expected, light absorption in canopies with leaves removed either in or above the bunch zone decreases on the east side on application day. On the west side light absorption slightly increases for both scenarios. Canopies of S1 absorbed less light on both sides of the row compared to S2 over the period of simulation. This can be explained by the decrease of S1 on the east side being higher than in S2, while at the same time the increase on the west side of S1 is lower than in S2. On the day after application a plant of the control simulation absorbed on average PAR abs = 3.12 MJ day −1 with a 50 % HDI of [2.88, 3.20] MJ day −1 on the east side of the canopy and 3.14 [2.97, 3.30] MJ day −1 on the west side. The average drop on the east side of the S1 canopies is 1.21 MJ day −1 , while on the east side of the S2 canopies a plant only absorbs on average approximately 0.85 MJ day −1 less than the control on doy 153. The aforementioned increases in PAR abs on the west sides of S1 and S2 scenarios immediately after leaf removal application were on average at 0.12 MJ day −1 and 0.20 MJ day −1 , respectively (Fig. 4).
Light absorption curves of both scenarios converge continuously to the control, although especially the east sides do not reach the control at the end of the simulations. However, an effect of compensation over time is clearly evident. At doy 172 the average difference per plant of S1 to the control decreases to approximately −0.33 MJ day −1 on the east side and in S2 to approximately −0.16 MJ day −1 , which is half the difference of S1 on the east side of the canopy. Differences on the west side are close to zero on Day 172. Brandt et al. (2019) measured mean daily solar radiation in the range of 3.88 to 5.10 MJ m −2 between the 4th-7th July in 2017 (doy 185-188) using horizontally placed lightsensitive films (20 mm × 35 mm) in bunch zones of Riesling vineyards at a height of 1.1 m. Assuming a leaf absorption of 71 % and an approximate leaf declination angle between 20° and 25° (Schmidt et al. 2019) this approximately equals 2.5 to 3.4 MJ m −2 day −1 . These values are higher, but still in the order of values estimated at the end of our simulation time frame, where the average light absorption at doy 172 for leaves in the bunch zone, i.e. between 0.7 and 1.1 m, of the control simulation was found to be 1.2 ± 1.0 MJ m −2 day −1 . Sources of uncertainty in this comparison include the different year and time points, as well as comparing a point wise measurement in the field with a zonal average from the simulations.
To sum this up, light absorption data over time revealed clear differences between the two leaf removal scenarios, representing bunch zone and above bunch zone leaf removal practice; however, differences to the control were almost compensated in less than 20 days after leaf removal.

East side compensation of light absorption due to lateral growth.
Observations in the 3D views of the model Virtual Riesling indicated why compensation of the loss of light absorption was possible. Gaps in the grapevine canopy caused by leaf removal diminished over time and were negligible at the end of the simulation period. This process was particularly related to lateral leaf growth as illustrated in an example with three vines in Fig. 5, where lateral leaves are marked to highlight their contribution to canopy closure.
These visual observations are supported by light absorption data of primary and lateral leaves from the control (Fig. 6). Light absorption of lateral leaves gained momentum approximately at leaf removal and steadily increased thereafter, whereas light absorption of primary leaves peaked 1 week after leaf removal followed by a slight decrease. The increase for lateral leaves was ca. four times larger than the remaining positive balance for primary leaves. This indicates that lateral leaf growth could be predominantly responsible for light compensation in our simulation scenarios.
To assess the role of lateral leaves in light absorption compensation after leaf removal, especially of leaves newly emerging in the zone of leaf removal, we subtracted the contribution of these lateral leaves growing in the leaf removal zone from the simulated east side . Absorbed PAR (PAR abs ) per vine and the 50 % HDI of 1008 simulated plants for two scenarios and the control, separated into east and west side of canopies (leaf removal applied on east side only). Absorption curves are plotted for the control (no leaf removal); scenario one (S1) with leaves removed in the bunch zone and scenario two (S2) with leaves removed above the bunch zone. The dotted line indicates the day of leaf removal application for S1 and S2. In addition, the difference in PAR abs between the control and both leaf removal scenarios for both sides is given as ΔPAR abs .

Light absorption in virtual Riesling canopies • 7
absorption [see Supporting Information- Fig. S1A and B]. The resulting absorption curves show that newly emerging lateral leaves were able to absorb light in the leaf removal zones for the first time on doy 154. If lateral leaves in the zones of S1 and S2 are excluded from the data, the trend of canopies to compensate for lost light absorption after leaf removal became very weak in S1 or even vanished in S2 when comparing absolute values [see Supporting Information- Fig. S1A and B]. The difference between absorbed light data in S2 without the absorption of lateral leaves in the leaf removal zone (no absorption of lateral leaves; n.A.o.L.L.) compared to the control increased from Day 154 to 172 from on average −0.84 to −0.91 MJ day −1 . In S1, with leaf removal in the bunch zone, compensation on the east side of the canopy was still present, but reduced when the contribution of laterals in the leaf removal zone is excluded in the comparison to the control. The decrease in the difference from Day 154 to 172 ranges from on average −1.18 to −0.78 MJ day −1 . This compensation is less than half as strong as when including lateral leaves (−1.19 to −0.33 MJ day −1 , see above). When estimating these data as percentages (Fig. 7A and B) of deviation to the average of the control we see S1 regaining absorption of up to approximately 95 % of the control by doy 172 and S2 of up to 98 %. When not considering lateral leaves only 88 % and 87 % of the control are reached by S1 and S2, respectively.  This reveals a strong contribution to total light absorption of lateral leaves, growing in the zones of leaf removal after application day. These leaves were responsible for the ability of canopies in regaining light absorption capacity in both scenarios. The increasing difference between S2 (n.A.o.L.L. in zone) and the control shows that light absorption of lateral leaves was even more important than in S1 for the compensation.
This can be explained with decreasing light absorption in lower parts of the canopies. To analyse the vertical light distribution we split the east side canopy of the control simulation into three zones: Zone 1 (Z1: h < 1.1 m) includes the bunch zone and all leaves below the cane; zone 2 (Z2: 1.1 m ≤ h ≤ 1.4 m) is similar to the S2 leaf removal zone and zone 3 (Z3) includes all leaves above h > 1.4 m. The leaves of the control in the zone equal to the leaf removal zone of scenario two (Z2) absorbed more light than those in the bunch zone and below the cane together (Z1) at the end of the simulations on Day 172 (Fig. 7C). During the simulation period of Virtual Riesling the height of canopies increases due to the implemented upward growth (cf. Fig. 5). Hence, lower leaves were more and more suspected to shading and more light was absorbed in upper parts of the simulated canopies over time, reflecting an observed linear decrease in light in the bunch zone (Kurtural et al. 2013).
Canopies, where leaves were removed either in or above the bunch zone, were able to partially compensate for the loss of light absorption over time. The simulation data revealed that lateral leaves were responsible for this effect that we observed on the east side, where leaf removal was applied. Leaf removal led to gaps in the canopies. In the course of the simulations these gaps where filled with emerging lateral leaves that started to absorb light after application day. These findings agree with observations described Light absorption in virtual Riesling canopies • 9 in literature. Canopies are able to recover and regain photosynthetic capacity, similar to non-defoliated canopies at véraison (Palliotti et al. 2011;Poni et al. 2018). The loss of leaves due to leaf removal can partially be compensated. According to our results and other studies this is most likely due to lateral growth (Palliotti et al. 2011;Hayman et al. 2012). Yet, our simulations also indicate a second source of compensation caused by leaves re-exposed to light.

Compensation of light absorption due to re-exposure.
The analysed data showed that light absorption, even immediately after leaf removal, was different between S1 and S2 on both sides of the canopies, although we aimed to remove leaves with similar absorption potential (cf. Fig. 3). This can be explained by leaves no longer covered in the canopy that got re-exposed to light after leaf removal. On the east side this effect was limited to leaves, which absorbed light below zones of leaf removal. Hence, in S1, where leaf removal started at the height of the cane (0.7 m), only a few leaves (0.02 ± 0.01 m 2 per plant, approximately 2 % of the total leaf area at doy 154) were located below the cane because shoots tended to grow upwards and were stabilized with wires. These leaves were from shoots that started growing from the bottom or the side of the cane. Below the zone of scenario two (h ≤ 1.1 m) there were all the leaves of the bunch zone plus those growing below the cane. When leaf removal is applied, light rays, which would have been intercepted by leaves growing in the leaf removal zones, were now reaching lower parts of the canopy. Hence, in S1 some rays got intercepted by leaves growing below the cane. In S2 many more light rays got intercepted by leaves in the bunch zone and below the cane, simply because there were a lot more leaves in their way. This effect is clearly visible in Fig. 8 showing light absorption of the control and both scenarios, but limited to heights below the zones of leaf removal.
The impact of S2 was expected, because leaves lower than leaf removal zone (S2 ≤ 1.1 m) suddenly absorbed a lot more light than the control (approximately +0.52 MJ day −1 ; +20 %). In contrast, in S1 the difference of light absorption below the leaf removal zone (S1 ≤ 0.7 m) to the control was negligible. This effect applied to both sides of the canopies, that is why light absorption increased on the west side and even more in S2. As already seen in Fig. 4, this effect was not limited to heights below the leaf removal zones, because light absorption on the west side increased after leaf removal also at heights corresponding to the respective zone of leaf removal. Overall, canopies, where leaves were removed in the bunch zone, did not benefit from this compensation-by-re-exposure effect as much as canopies with leaf removal applied above the bunch zone. Similar effects of leaves, re-exposed to light, partially compensating the impact of leaf removal on light absorption has also been observed in previous studies on real vines (Poni et al. 2018).

CON CLUSIONS
To study effects of leaf removal on grapevine canopy light absorption in silico, we implemented a new method for simulating leaf removal techniques into the functional-structural plant model Virtual Riesling. We compared two scenarios, where leaf removal was applied in and above the bunch zone of canopies in virtual vineyards using the extended Virtual Riesling model. Applying leaf removal on the same day in different heights in canopies led to evident differences in light absorption between both scenarios. Furthermore, the model allowed us to study the impact of leaf removal on light absorption over the entire simulation period. Thereby, simulation results were consistent with findings of previous studies of vines that were able to regain a large proportion of light absorption capacity after leaf removal, similar to levels without leaf removal. Investigating the role of leaf growth revealed that in the simulations lateral leaves were mainly responsible for compensating for lost light absorption within 20 days after leaf removal. Such data sets created by the model Virtual Riesling will be used in future studies to model and estimate other impacts on vines, especially the risk of berry sunburn, which is closely related to local light intensities. Implementing a responsiveness of growth on local light conditions might further enhance the models predictive potential of microclimatic effects. Since plant management can be adjusted in simulations, the model is also a promising tool to systematically analyse different management techniques and their impact on plant architecture and light distribution in canopies in silico.

SUPPORTIN G INFOR M ATION
The following additional information is available in the online version of this article- Figure S1. (A) Absorbed PAR (PAR abs ) per vine and the 50 highest density interval (HDI) from 1008 simulated plants at the east Figure 8. Absorbed PAR (PAR abs ) per vine and the 50 % HDI from 1008 simulated plants below leaf removal zones (≤ 0.7 m, ≤ 1.1 m) including both sides of the canopy for: canopies without leaf removal (control) compared to the canopies from the respective leaf removal scenarios (S1 : ≤ 0.7 m, S2 : ≤ 1.1 m).
side for: canopies with no leaf removal (control), scenario one (S1) canopies and S1-n.A.o.L.L. canopies which exclude light absorption of lateral leaves growing within the leaf removal zone (0.7-1.1 m) after leaf removal. (B) Absorbed PAR per vine at the east side for: canopies with no leaf removal (control), scenario two (S2) canopies and S2-n.A.o.L.L. canopies which exclude light absorption of lateral leaves growing in the leaf removal zone (1.1-1.4 m) after leaf removal. (C) Absorbed PAR (PAR abs ) per vine and the 50 % HDI from 1008 simulated control plants in three vertical zones at the east side of the canopy: below 1.1 m (Z1), between 1.1 and 1.4 m (Z2) and above 1.4 m (Z3). The dotted line indicates the day of leaf removal. Figure S2. Simulated absorbed radiation (MJ m −2 day −1 ) using 1-m 2 black body sensors with three orientations horizontal (ground), vertical (east, west). Upper panel: 2018, full year; lower panel: zoom to main simulation time frame. Plus: Global radiation (MJ m −2 day −1 ) data from local weather stations: historic  and from the year 2018. Figure S3. Exemplary distribution of light sources (dome (n = 72) and sun path (n = 24)) and contributed energies for doy 153. Fixed model parameters from Evers and Bastiaans (2016) and Evers et al. (2010) were used estimating a fraction of diffuse to direct radiation of raddif fuse radglobal = 0.8. This value was computed using an empirical function from Spitters et al. (1986): raddif fuse radglobal = 1.33 − 1.46 · ta, using a transmissivity value of ta = 0.3548 (Evers and Bastiaans 2016;Evers et al. 2010). Table S1. Description of light dome (half sphere) consisting of 72 diffuse light sources on six layers (beta angles) and empirical factors f 1 (light intensity allocation) and f 2 (sine angle correction) from Evers and Bastiaans (2016). Exemplary corresponding light source energies contributed for day of the year 153 as presented in Supporting Information- Fig. S3. Table S2. Exemplary path of the sun at day of the year 153 [see Supporting Information- Fig. S3] and corresponding contributed direct light source energies. Table S3. Environmental input variables used in the simulations for 2018 doy 103-172. Temperature data for Geisenheim were available from the Germany's National Meteorological Service (the Deutscher Wetterdienst (DWD)) local weather station as daily mean, minimum and maximum temperatures (2 m above-ground). Daylength (h) is calculated using the local latitude of Geisenheim (50°) following Spitters et al. (1986, eq. 17).