Energy savings in greenhouses by transition from high-pressure sodium to LED lighting

• A novel analysis of the effect of LEDs on the greenhouse energy budget is


H I G H L I G H T S G R A P H I C A L A B S T R A C T
• A novel analysis of the effect of LEDs on the greenhouse energy budget is presented. • The total energy system, including heating and lighting demand, was examined. • Energy savings in multiple climate scenarios from around the world were examined. • Energy for light was reduced by 40% while energy for heating increased by 9-49% • A transition to LEDs was predicted to save 10-25% of total greenhouse energy demand.

Introduction
Greenhouse horticulture in high latitudes requires vast amount of energy input. In northern latitudes, heated glasshouses consume energy at a rate of 1100-1900 MJ m − 2 year − 1 [1]. With an estimated 40,000 ha of vegetable glasshouses worldwide [2], and at least that much area for ornamental production [3], greenhouses consume more than 880 petajoule (PJ) of energy every year. In the Netherlands, the greenhouse industry consumes 113 PJ of energy per year [4], resulting in 5.7 megatons of CO 2 emissions, nearly 25% above the targets set for the sector by the government [5].
Worldwide, greenhouse energy use is increasing, due to wider and more intense use of heating and supplemental lighting [1]. Heating is used to maintain indoor temperatures that are favorable for the crop year-round, while supplemental lighting is used either for daylength control, in order to regulate plant developmental processes such as flowering, or as assimilation lighting to increase crop growth [6].
Assimilation lighting requires substantial electrical input. An illustrative example comes from the Netherlands, where production of tomatoes is possible with or without supplemental light [7]. It is estimated that a greenhouse with lamps requires more than double the energy input, and has more than triple the carbon footprint, compared to a greenhouse without lamps. At the same time, the illuminated greenhouse is estimated to provide only 27% higher yields [7]. Another illustrative example comes from Ontario, Canada, where illuminated vegetable greenhouses consume 10 times more electricity than unilluminated greenhouses [8].
These examples illustrate how supplemental lighting enhances greenhouse production but carries high financial and environmental costs. High-pressure sodium (HPS) lamps are currently the most prevalent lamp type used for supplemental lighting in greenhouses, but light emitting diodes (LEDs) are progressively being adopted as an alternative for, or in addition to, HPS lamps [9].
The main advantage of LEDs over traditional lamps is that they are better at converting electrical power to photosynthetic light [1]. For horticultural lamps, the conversion rate from electrical input to photosynthetically active radiation (PAR) is termed as the lamp's photosynthetic photon efficacy (PPE), or efficacy in short. This value is expressed in mols of photons of PAR emitted per energy input, measured in µmol J − 1 [10]. While the efficacy of HPS lamps is between 1.7 [10] to 1.85 [7] µmol J − 1 , commercial LEDs currently provide 3 µmol J − 1 [11]. In other words, LEDs have a 60% higher efficacy than traditional horticultural lamps, and this value is expected to rise in the future [12].
The increase in efficacy of LEDs is promising to bring about meaningful energy savings in greenhouses [1]. At the same time, uncertainty regarding the financial consequences of transitioning to LEDs is a critical factor preventing adoption of this technology by growers [13]. One factor contributing to this uncertainty is the fact that LEDs provide less heat than the lamps that are currently used. HPS lamps emit considerable amounts of radiative heat [14], which reduces the load on the greenhouse heating system [15]. When HPS lamps are replaced by LEDs, some of the heat that is no longer provided by the lamps must be provided in other ways, and the total energy saving in the greenhouse could be less than expected.
Despite this, many studies that examined the energy saving potential of LEDs in greenhouses focused on the savings of the lighting system only [10,12,16,17], and did not quantify the influence of the lighting system on heat demand. Only a few limited studies reported on how LEDs influence the total energy demand of the greenhouse [18], and these suggest that effects on total energy saving might be disappointing. For example, a hybrid system combining LEDs and HPS lamps was compared to a full-LED system. The savings in electricity by using full LEDs were 37%, but the total energy saving was only 11% [19]. Another trial found a 60% savings on lighting by the use of LEDs, but only a 6.5% savings on total energy use, due to a higher heat demand under LEDs [20]. These studies show that the total energy savings achieved by a transition to LEDs are considerably lower than the savings on lighting alone. However, it is not clear what the actual savings may be in a commercial greenhouse, or which factors influence these savings.
A thorough analysis quantifying how much energy can be saved by using LEDs is currently missing in the literature. Such an analysis should also include an assessment of the separate lighting and heating demands of the greenhouse, since these two uses often originate from different energy sources. Heating energy can originate from boilers fueled by natural gas or other fuels, geothermal heat, heat pumps, heat buffers, combined heat and power (CHP) generators, and others [3]. Similarly, electricity for lighting can come from fossil-fueled power plants, photovoltaic cells, wind turbines, CHPs, and more. Each of these sources and systems carries different financial and environmental costs. Since a transition from HPS lamps to LEDs generally involves lower lighting demands but higher heating demands, it is important to quantify these changes, so that they can be evaluated against the available energy sources for each respective greenhouse or region. In this way, greenhouse growers at an individual level, and policymakers at a regional level, can ensure that the lighting and heating systems in greenhouses are compatible with the available energy resources, such that energy is being used in an optimal manner.
Experimental trials comparing HPS and LED lamps in greenhouses are costly and limited in scope. Ideally, such trials should be performed in neighboring, but not bordering greenhouses, so that the weather influence is equivalent but border effects are prevented. Even so, such trials can only provide information for a single greenhouse setting in a single weather scenario, and cannot shed light on general phenomena.
Process-based mathematical modelling of the greenhouse climate is a widely recognized discipline used for the evaluation of greenhouse energy demands. This approach has been used since the 1960 ′ s [21], developed throughout the following decades [22], and used to analyze energy-saving scenarios since at least the 1990 ′ s [23]. Optimal control techniques have been combined with process-based modelling to create methods for reducing the applied heating and cooling [24] and minimizing the energy use [25] of advanced greenhouses. More recently, models have been proposed for predicting the heating demands specifically for greenhouses in cold regions [26], and for analyzing energy saving techniques in those greenhouses [15]. Other recent studies incorporated detailed crop models to more accurately predict both crop yield and energy demand [27]. Nevertheless, process-based models that include both HPS and LED lighting are rare. Two recent exceptions are a model by Righini et al. [28], and the GreenLight model [29].
GreenLight is a novel, open-source model for illuminated greenhouses with a tomato crop. It has several advantages which make it suitable for the analysis of energy demands in illuminated greenhouses. First, it has been evaluated against a dataset of an experimental trial where HPS and LED lighting was compared. The error in predicting energy use was in the range of 1-12% [29]. Second, GreenLight is available as open source MATLAB code at https://github. com/davkat1/GreenLight. The availability of the source code makes the model transparent and extendable. Researchers and practitioners who wish to evaluate, extend, or adjust the model to their own scenarios can freely do so. The framework provided in this study may thus be further applied to investigate in detail the energy balance and the consequences of transitioning to LEDs in any local scenario.
The objective of this study was to systematically quantify how much energy can be saved by transitioning from HPS to LED lighting in greenhouses, under various control and design settings and in a wide range of climates. To the best of our knowledge, this is the first thorough analysis on the influence of LEDs on the greenhouse energy budget. In particular, we quantified the changes in the separate energy demands for heating and lighting. Such an analysis is crucial for growers and policymakers to choose the best form of lighting technology with respect to their local climate and energy market, in a way that makes use of the available energy resources in the most efficient and environmentally sustainable manner. For this purpose, simulations were performed using the GreenLight model, allowing us to closely examine the greenhouse energy system, and to test how various components of the system influence the energy budget, providing a comprehensive view which is impossible to attain in isolated empirical trials. We tested how much energy can be saved by transitioning from HPS lamps to LEDs in terms of lighting and heating, and which energy fluxes in the greenhouse system are responsible for these changes. We examined how the local climate influences the potential energy savings by simulating greenhouses in various locations around the world, ranging from subtropical China to arctic Sweden. Several greenhouse settings were considered, including various indoor temperatures, light intensities, lighting durations, and structure insulations. We further examined which energy fluxes are most influenced by the change from HPS to LED lighting, and how these fluxes change throughout the year. Lastly, we discuss what these results could mean in practice for growers and policy makers, and how they may be further extended.

Potential energy savings considering only the lighting system
As a first analysis, the expected energy savings were calculated in cases where a transition to LEDs does not influence the greenhouse heating requirement. This means that the energy savings are a direct result of the higher efficacy of LEDs. The efficacy of current HPS lamps was assumed to be 1.8 µmol J − 1 [10, including comments to the online article]. For LEDs, two efficacies were considered: an efficacy of 3.0 µmol J − 1 , which is the highest efficacy found in independent tests, and an efficacy of 4.1 µmol J − 1 , representing the limit of current technology [12]. LEDs with these efficacies would require 40% and 56% less energy per photon, respectively, than an HPS lamp. The resulting energy saving by a transition to LEDs was expressed as a fraction of the energy requirement of the HPS greenhouse. For this, we denote Q HPS Light (MJ m − 2 year − 1 ) as the amount of purchased energy used for lighting in the HPS greenhouse. Similarly, Q HPS Heat (MJ m − 2 year − 1 ) is the amount of purchased energy used for heating the HPS greenhouse. The total energy use of the HPS greenhouse is thus: Assuming the heating requirements remain equal, the energy requirements of a greenhouse with LEDs is: where E LED (MJ m − 2 year − 1 ) is the total energy input of the LED greenhouse; ∊ HPS =1.8 µmol J − 1 is the efficacy of the HPS lamp and ∊ LED = 3 µmol J − 1 or 4.1 µmol J − 1 is the efficacy of the LED lamp. Compared to the original energy input, this results in a total relative energy saving of: where F Light = Q Light Q Light +QHeat (-) is the fraction of energy input used for lighting in the HPS greenhouse.
In addition, we calculated the energy savings that can be achieved if a hypothetical infinitely efficient lamp was used, i.e., if the energy input for lighting the greenhouse could be reduced to zero. Following a similar calculation to the one above, in this case the savings are equal to F Light .

Model simulations using various climate locations
In order to evaluate the influence of local climate on the energy use of HPS and LED greenhouses, 15 locations around the world were selected and weather data from these locations was used as input to the model. The locations were chosen such that a varied set of climates would be represented, with a focus on locations where illuminated greenhouses are present. Climate data was retrieved from the Ener-gyPlus website [30], which compiles standard weather data from various sources. The chosen locations, as well as their latitude, longitude, elevation, and the originating dataset, are presented in Table 1.
For each of the climate datasets listed in Table 1, the following variables, given in 1-hour intervals, were used: Global solar radiation I Glob (W m − 2 ); Air temperature T Out ( • C); Relative humidity RH Out (%); Wind speed v Wind (m s − 1 ); Horizontal infrared radiation from the sky I Sky (W m − 2 ). These values were interpolated to 5-minute intervals by using a piecewise cubic Hermite interpolating polynomial. Soil temperatures T SoOut ( • C) were given in monthly average values; these were interpolated to 5-minute intervals by fitting the values to a sine function. Outdoor CO 2 concentration value was set at 410 ppm. Fig. 1 and Fig. 2 present an overview of the weather patterns in the locations used in this study. Fig. 1 provides the monthly means of solar radiation and outdoor temperatures, and their progression throughout the year. Fig. 2 presents the yearly means of solar radiation and outdoor temperature.

Model simulations of HPS and LED greenhouses
The GreenLight model [29] was used for all simulations performed in the study. This model simulates an advanced, Venlo type greenhouse with a tomato crop, equipped with various types of supplemental lights. GreenLight was evaluated against measurements in greenhouse compartments with either HPS or LED top lighting throughout an entire winter season and was found to estimate the greenhouse's heat demand by an error of 1-12%. When taking into account the entire energy demand of the greenhouse, including lighting, the prediction error was Table 1 The 15 locations selected for the simulations used in this study and the originating datasets. The various datasets were retrieved from the EnergyPlus website [30].  [34] 0.5-5% [29]. In this study, the case of either HPS or LED top lights was considered. The simulated season was 350 days long, from September 27 until September 11 the next year.

Reference greenhouse settings
A 4-hectare Venlo type glasshouse was considered for this study, measuring 200 m in width and 200 m in length. The gutter height of the greenhouse was 6.5 m, and the ridge height was 7.3 m. The slope of the roof was 22 • . The roof was composed of glass panels measuring 2.16 m by 1.67 m. A ventilation window was installed in 1 out of every 6 panels, measuring 1.40 m by 1.67 m. The maximal opening angle for the window was 60 • . Thermal screens were installed at a height of 6.3 m. Path width was 1.6 m, and a pipe rail system was installed in the paths, with a total pipe length of 1.25 m m − 2 and a pipe diameter of 51 mm. The maximal rate of the CO 2 injection system was 185 kg ha -1 h − 1 . The

Table 2
Greenhouse design parameters used for the simulated greenhouses with reference settings. Parameters not included were taken from the Dutch greenhouse setting described in the electronic appendix of [35]. 5.14 parameters used for simulating the greenhouses with the reference settings are given in Table 2. Parameters not included in the table were taken from the Dutch greenhouse setting described in the electronic appendix of [35]. For the lamp settings, the photon content of PAR emitted by the lamp ζ LampPAR was set at 4.9 µmol (PAR) J − 1 (PAR) for the HPS lamp [36]. For the LEDs, the modelled lamp consisted of 6% blue LEDs and 94% red LEDs, a combination that was found to be optimal for tomatoes [37]. Assuming the blue LEDs emit at a wavelength of 450 nm, the red LEDs at 660 nm [12] and using Planck's equation [12], the resulting PAR photon content is 5.4 µmol (PAR) J − 1 (PAR). The fraction of input energy converted to PAR η LampPAR was set at 0.37 for HPS lamps and 0.55 for LEDs in order to achieve an efficacy of the lamps (η LampPAR ⋅ζ LampPAR ) of 1.8 µmol (PAR) J − 1 (input) for the HPS lamps [10, including comments to the online article] and 3 µmol (PAR) J − 1 (input) for the LEDs [12]. The lamp energy input θ LampMax was set at 111 W m − 2 for the HPS lamps and 66.7 W m − 2 for the LEDs in order to achieve an equivalent photosynthetic photon flux density (PPFD, η LampPAR ⋅ζ LampPAR ⋅θ LampMax ) of 200 µmol m − 2 s − 1 in both greenhouses. The parameters used for the lamps in the HPS and LED greenhouses are given in Table 3. A full description of the lamp model and its parameters is given in [29].

Reference control settings
In order to make a valid comparison between various locations and lamp settings, a general climate control regime was devised. In the following, "daytime" refers to the period from sunrise to sunset and "nighttime" from sunset to sunrise; "light period" refers to the daytime as well as the nighttime when the lamps are on; "dark period" is the part of the nighttime when the lamps are off.
1. Lamps: lamps were on every day between midnight and 18:00 with the following exceptions: o The lamps were switched off whenever global solar radiation outside the greenhouse was above 400 W m − 2 . o The lamps were off if the predicted global solar radiation sum outside the greenhouse during that day was above 10 MJ m − 2 day − 1 .
2. CO 2 injection: CO 2 was injected during the light period, whenever the indoor CO 2 concentration was below the target setpoint of 1000 ppm. 3. Heating: Heating was applied whenever the indoor temperature was below the target setpoint, which was 19.5 • C during the light period and 18.5 • C during the dark period. 4. Roof ventilation: ventilation was applied in any of the following cases: o Ventilation due to excess heat: the roof openings were opened whenever the indoor temperature was 5 • C above the target setpoint for heating. o Ventilation due to excess humidity: the roof openings were opened whenever the indoor relative humidity was above 87%. o However, the ventilation was forced to close if the indoor temperatures were 1 • C below the heating setpoint. 5. Thermal screens: thermal screens were closed during the day if the outdoor temperature was below 5 • C, and during the night if the outdoor temperature was below 10 • C. However, the screens were opened whenever ventilation was needed: o Screen opening due to excess heat: the screens were opened whenever the indoor temperature was 4 • C above the target setpoint for heating. o Screen opening due to excess humidity: the screens were opened whenever the indoor relative humidity was above 85%.
The reference control settings are summarized in Fig. 3. The heating, ventilation, thermal screens and CO 2 injection were controlled using a smoothed proportional controller, which was defined using a sigmoid function: where x is the controlled variable (e.g., the indoor temperature), setPoint is the desired setpoint for the controlled variable (e.g., 18.5 • C during the dark period), pBand is a band defining the width of the proportional control, and Action defines the controller action. At Action = 1, the controller is at full capacity (e.g., heating is at full power), at Action = 0 the controller is off. As can be seen from Equation (4) and Fig. 4, the controller is close to full action at x = setPoint + pBand, and close to no action at x = setPoint. Note that pBand may also be negative, in which case Fig. 4 would be flipped horizontally.
The pBand values for the proportional controllers were: − 1 • C for heating; − 100 ppm for CO 2 injection; − 1 • C for thermal screen closure due to cold outdoor temperatures; 1 • C for thermal screen opening due to excess indoor heat; 10% relative humidity for thermal screen opening Table 3 Parameters used for the HPS and LED lamps in the reference simulations. See [29] for a full description of the lamp model and its parameters. • C 0 0 due to excess humidity; 4 • C for ventilation opening due to excess heat; − 1 • C for ventilation closure due to low indoor temperature; 50% relative humidity for ventilation opening due to excess humidity [23].

Adjustment of heating setpoint under LEDs
Studies have shown that crops under LED lights are colder than those under HPS lamps [36]. In practice, a greenhouse with LEDs may need to increase the air temperature setting in order to achieve the same crop performance as a greenhouse with HPS lamps [39]. To take this into account, a new parameter, p heatAdjLamp ( • C) was introduced to the model. This parameter influenced the greenhouse control settings such that whenever the lamps were on, the heating setpoint for the greenhouse air T sp was increased by the value of p heatAdjLamp . The value for p heatAdjLamp was 0 • C for the HPS greenhouse in all settings. The value for p heatAdjLamp was 0 • C for the LED greenhouse in the reference setting, but was modified to 1 • C in one of the additional scenario simulations, see Section 2.3.4.

Additional scenario simulations
In addition to the reference settings, several other scenarios were examined to test the model predictions under a wider range of greenhouse structure designs and climate control settings. These additional simulations were (adjustment to the reference settings are in bold): Simulations with heat adjustment (scenario 1 above) were performed for all 15 locations considered in this study. The other scenarios (2-8) were tested for 3 locations: Amsterdam (AMS), Calgary (CAL), and Chengdu (CHE), representing three different climates: low radiation and high temperatures in Chengdu, high radiation and low temperatures in Calgary, and mild temperatures and radiation in Amsterdam (Fig. 2).

Analysis of energy fluxes
In order to better understand which factors influence the greenhouse energy demands, further analysis of the simulation results was performed for the greenhouse in Amsterdam. This location was chosen since Fig. 3. Control of heating, ventilation and thermal screen in the reference setting. T sp is the temperature setpoint for heating. CO 2 was injected during the light period if the indoor CO 2 concentration was below 1000 ppm. Lamps were on from midnight to 18:00 unless momentary solar radiation was above 400 W m − 2 or daily solar radiation was above 10 MJ m − 2 day − 1 . Figure adapted from [38].  (4)). The controller is close to full action when the process variable × is equal to setPoint + pBand. The controller is close to no action at x = setPoint. If pBand is negative, the curve is flipped horizontally.
it has a relatively mild climate in terms of yearly temperature and radiation, located close to the average of the climates tested in this study, and those that are typical for regions with illuminated greenhouses (Fig. 2). The insights gained from analyzing the greenhouse in Amsterdam can expose general phenomena that apply to greenhouses all over the world.
Once the simulations were performed, the yearly sum of the energy inputs and outputs out of the greenhouse system were calculated, to quantify and explain the differences in energy requirements between greenhouses with HPS lamps or LEDs. These inputs and outputs (all in MJ m − 2 year − 1 ) are given below, together with their notation in the studies where they were first described [29,35] (see also glossary): 1. Energy absorbed from the sun by the greenhouse structure, the canopy, and the floor R Glob SunAir + R PAR SunCan + R NIR SunCan + R PAR SunFlr + R NIR SunFlr + R Glob SunCov,e 2. Energy input to the heating pipes H BoilPipe 3. Energy input to the lamps Q LampIn 4. Energy output to the soil H So5SoOut 5. Thermal radiation output emitted towards the sky from the greenhouse cover, screens, lamps, canopy, pipes, and floor R Cov,eSky + R ThScrSky + R LampSky + R CanSky + R PipeSky + R FlrSky 6. Energy output to the outdoor by convection through the cover H Cov,eOut 7. Energy output through ventilation H AirOut + H TopOut 8. Latent heat output, representing the net loss of sensible heat by conversion to latent heat. This value is composed of loss of sensible heat due to transpiration minus gain of latent heat due to condensation on the screens and cover: L CanAir − L AirThScr − L TopCov,in

Model code and execution
The model simulations were executed in MATLAB (MATLAB R2016b-R2019b, The MathWorks). Code used for running the simulations is available at https://github.com/davkat1/GreenLight. The data resulting from the simulations is available on the 4TU.ResearchData database, https://doi.org/10.4121/13096403 [40].

Potential energy savings considering only the lighting system
In cases where a transition from HPS lamps to LEDs does not alter the heating demand, the expected energy savings realized by transitioning to a more efficient lamp were influenced by two factors: the efficiency of the new lamp and the fraction that lighting takes up out of the total greenhouse energy demands (Fig. 5). Naturally, a greenhouse with very little assimilation lighting (fraction for lighting close to 0%) gained very little energy savings by changing the lighting system. On the other extreme, in a greenhouse where the energy input was only used for lighting (fraction for lighting is 100%), the energy savings of the lighting system was equivalent to the total energy savings. Even with an infinitely efficient lamp, the total energy savings percentage was limited by the fraction of the energy input going to lighting.

Energy balance considering both the lighting and heating system
The energy demands as simulated by the GreenLight model varied greatly in the different world locations under the reference settings. In the HPS greenhouses, the total energy demands varied from less than 1400 MJ m − 2 year − 1 in Shanghai (SHA) to nearly 3400 MJ m − 2 year − 1 in Kiruna (KIR) (Fig. 6). As expected, in all locations the change from HPS to LEDs resulted in a 40% reduction in lighting demand, which ranged from a saving of 315 MJ m − 2 year − 1 in Beijing (BEI) to 662 MJ m − 2 year − 1 in Kiruna (KIR). At the same time, in all locations the LED greenhouse required more heating than the HPS greenhouse. The extra heating needed ranged from 79 MJ m − 2 year − 1 in Beijing (BEI) to 224 MJ m − 2 year − 1 in Anchorage (ANC). In relative terms, the increase in heating demands ranged from 9% in Calgary (CAL) to 49% in Chengdu (CHE). The resulting total energy savings ranged from 13% in Calgary (CAL) to 27% in Chengdu (CHE).
In the simulations performed with the reference settings, the fraction of energy that goes to lighting in the HPS greenhouses ranged from 45% in Calgary (CAL) to 85% in Chengdu (CHE). Increasing the temperature setpoint by 1 • C under LEDs, in line with common practice (see Section 2.3.3), reduced the predicted energy savings by around 1.5% (Fig. 7). There is a positive linear correlation between the predicted relative energy savings when transitioning to LEDs and the fraction of energy inputs that goes to lighting in the HPS greenhouse. A linear regression based on the model outputs predicts that the relative energy savings are 0.37x-5.41 percent, where x is the percent of energy that goes to lighting in the HPS greenhouse. This linear model provides an accurate prediction of the relative energy savings, with a coefficient of determination (R 2 ) of 0.90, and a root mean squared error (RMSE) of 1.90%.

Yearly incoming and outgoing energy fluxes
The energy savings achieved by a transition to LEDs are further elucidated by looking at the ingoing and outgoing yearly energy fluxes in the HPS and LED greenhouses, using Amsterdam as a representative case (Fig. 8). Since the greenhouse energy system was on average in steady state, for each greenhouse the sum of its incoming fluxes is almost equal to the sum of its outgoing fluxes. Thus, the savings in energy inputs correspond to a reduction in energy outputs. The biggest change in outgoing energy fluxes when transitioning to LEDs was a reduced loss to latent heat resulting from less crop transpiration. Smaller decreases are seen in the losses due to ventilation and convection through the cover. The LED greenhouse with temperature adjustment, i.e., where the air temperature setpoint was increased by 1 • C whenever the lamps were on, consumed 61 MJ m − 2 year − 1 more in heating compared to the LED greenhouse in the reference setting. This increase was mainly due to more convection through the cover.

Seasonal variation in energy savings
A clear variation in energy inputs can be seen throughout the year. In winter (22-153 days after planting, corresponding to October 19 to February 27), the lamps were on for their maximal duration every day, resulting in a constant input of lighting energy, and a constant daily saving in lighting input of 2.9 MJ m − 2 day − 1 by using LEDs (Fig. 9). At the same time, throughout winter the LED greenhouse in Amsterdam required on average 1 MJ m − 2 day − 1 more heating than the HPS greenhouse. From day 154 lamps were only occasionally used. When the   7. Simulated savings in total energy input by transitioning from HPS to LEDs in relation to the energy input used for lighting in the HPS greenhouse. 15 locations throughout the world (Section 2.2) and 9 control and design scenarios (Section 2.3) are given. The efficacy of the HPS and LED lights was 1.8 and 3 µmol J − 1 , respectively. Potential energy saving represents energy saving if a transition to LEDs does not increase heating needs. Achieved energy saving is a linear regression based on the given data points.
lamps were off, the energy requirements of both greenhouses were similar. When the lamps were on (e.g., days 191-194), more heating was often needed in the LED greenhouse.
An examination of the greenhouses' energy fluxes on a typical winter and summer day when lighting was used provides further detail. As seen earlier (Fig. 9), on days when lamps were on for their maximum duration of 18 h, the HPS lamps consumed 7.2 MJ m − 2 day − 1 and the LEDs consumed 4.3 MJ m − 2 day − 1 , saving 2.9 MJ m − 2 day − 1 in lighting (Fig. 10). On a typical winter day, this reduced energy input was counteracted by an added 1.3 MJ m − 2 day − 1 for heating, resulting in a total saving of 1.6 MJ m − 2 day − 1 (Fig. 10). On a typical summer day, the LED greenhouse used only 0.1 MJ m − 2 day − 1 more heating, resulting in a total saving of 2.8 MJ m − 2 day − 1 . Considering the outgoing energy fluxes, in winter the main difference when using LEDs is less convection through the cover (0.8 MJ m − 2 day − 1 ), followed by lower losses to latent heat (0.4 MJ m − 2 day − 1 ) and less ventilation (0.3 MJ m − 2 day − 1 ). In summer, the main difference is reduced latent heat (1.2 MJ m − 2 day − 1 ), followed by less ventilation (0.8 MJ m − 2 day − 1 ) and convection (0.3 MJ m − 2 day − 1 ).
The difference in effects of transitioning to LEDs between winter and summer is further detailed by looking at the trajectories of the energy fluxes throughout a representative day. In winter, heating was similar Fig. 8. Yearly incoming and outgoing energy fluxes for the HPS, LED, and LED with temperature adjustment (LED TA) greenhouses in Amsterdam. Air temperature setpoints where identical in the HPS and LED greenhouses. In the LED TA greenhouse, air temperature setpoint was raised by 1 • C whenever the lamps were on. between the greenhouses when the lamps were off (18:00-24:00), but higher in the LED greenhouse when the lamps were on. Ventilation in both greenhouses was very small, with slightly more ventilation in the HPS greenhouse (Fig. 11A). In summer, some heating was applied when the lamps were off (18:00-24:00) in both greenhouses, and ventilation was considerably higher in the HPS greenhouse compared to the LED greenhouse (Fig. 11B). Ventilation was applied either when indoor humidity or indoor temperature were too high. In winter, the higher ventilation in the HPS greenhouse was due to excess humidity, as the greenhouse air temperature in both greenhouses was far below the ventilation setpoint due to excess heating (Fig. 11E). Higher humidity in the HPS greenhouse was due to higher transpiration, which resulted in more losses to latent heat (Fig. 10).
The CO 2 concentration and injection were influenced accordingly: in winter, both greenhouses reached the desired setpoint of 1000 ppm during the light period (0:00-18:00), with slightly more CO 2 injected in the HPS greenhouse (Fig. 11C). In summer, considerably more injection was needed in the HPS greenhouse, to the extent that in the middle of the day the desired setpoint could not be reached (Fig. 11D). In summer, ventilation was also needed to remove excess heat (Fig. 11F).

Energy savings in the greenhouse by transitioning to LEDs
This study showed that only a comprehensive approach to the greenhouse energy system, where the various components of the greenhouse energy balance are considered, can provide an insightful quantification of the energy saving by a transition to LEDs in greenhouse horticulture. While using more efficient lamps resulted in direct savings on energy inputs for lighting (Fig. 5), a transition from HPS lamps to LEDs was associated with an increase in the greenhouse's heating demands, resulting in lower total energy savings (Fig. 7). Using a mechanistic and dynamic model such as GreenLight allowed us to quantify these demands, and understand how they depend on the greenhouse settings and the evolving outdoor climate. Especially during cold weather, the heat emitted by the HPS lamp contributed to heating the greenhouse, and this loss of heat had to be compensated by the heating system when LEDs replaced HPS lamps.
The simulations performed in this study, using several greenhouse design and control settings and various climates around the world, predicted that in most cases transitioning to LEDs will save 40% on electricity input for lighting, but will increase the heating demand by 9-49%. The result is a saving of 10-25% of the total energy inputs in nearly all cases (Fig. 7). The relative energy savings correspond to the inputs of the HPS greenhouse: denoting x as the percent of energy inputs that is used for lighting in the HPS greenhouse, the percent of energy savings predicted by a transition to LEDs is estimated as 0.37x − 5.41 (Fig. 7). These savings are lower than 0.4x, which is the predicted saving when only lighting was considered (Fig. 5).

Changes in balance between energy sources and uses
A switch to LEDs influenced not only the total energy inputs to the greenhouse, but also the ratio between the heating and lighting needs (Fig. 6, Fig. 8). In general, converting to LEDs resulted in a lower energy input for lighting and a higher input for heating. A decision on whether this transition is favorable or not, for instance in terms of costs for the grower or environmental footprint, depends on the energy sources used. For example, consider a greenhouse where heating is supplied by a boiler burning natural gas, and lighting is supplied by wind or solar power. In this case, a switch to LEDs will reduce the use of clean energy and increase the use of fossil fuels. On the other hand, if a cheap and environmentally clean source of heating such as geothermal heat is available, the benefits of lower electrical inputs for the lamps might outweigh the costs of extra heating in the LED greenhouse. The model simulations presented in this study quantify the benefits of a transition to LEDs to support practitioners in choosing the best lighting system for their local situation.
Furthermore, many greenhouses use a combined heat and power (CHP) generator to produce both electricity, heating, and CO 2 by burning natural gas [1]. The CHP in the greenhouse may produce electricity in a cleaner manner than the public power plant, since the CHP also makes use of the heat and CO 2 that is generated when burning gas. This electricity, if not used in the greenhouse itself, may be sold back to the public grid, resulting in a reduction of emissions on a national level [5] which can be counted as a negative carbon footprint for the greenhouse, as well as an extra source of income for the grower [7]. In a greenhouse where both a boiler and a CHP are available, choosing which of the two to use at what time is a complicated problem [41]. If a grower wishes to use the CHP for electricity, the heat that is generated must be used simultaneously in the greenhouse or stored in a buffer. If the heat buffer is full and the greenhouse does not require heating, the CHP cannot be used without wasting heat, for instance by simultaneously heating the greenhouse and cooling it through ventilation.
A typical CHP generates around 1.25 MJ of heat for every 1 MJ of electricity [41]. This ratio is very close to the energy demands found for a Dutch LED greenhouse in winter (Fig. 10) and over a full year (Fig. 8).
In contrast, the heat to light ratio in an HPS greenhouse is 0.5-0.6. This means that in principle, an LED greenhouse can supply all its lighting needs using a CHP without wasting heat. In contrast, an HPS greenhouse that would use a CHP for even half its lighting demand would still need to waste some of the generated heat. It follows that a greenhouse with a CHP will likely see larger energy savings when transitioning to LEDs, compared to those found in the current study. This potential for energy savings by combining LEDs and a CHP should be further explored in future studies.

Daily and seasonal patterns of transition to LEDs
The main influence of a transition to LEDs on the greenhouse energy demand was seen in winter, when light from the sun was limited and lamps were used extensively (Fig. 9). Later in the year the lamps were used less frequently and thus the choice of lamp had little influence on the energy demand. In winter, HPS lamps provided heat which was used in the greenhouse. Replacing HPS lamps with LEDs resulted in a higher heat demand when the lamps were on (Fig. 10), but not when the lamps were off (Fig. 11). In contrast, in summer when the lamps were on there was a need to ventilate the extra heat, and the demands from the heating system were minimal under both lamp types. Since the greenhouses consumed considerably more energy in winter compared to summer, the energy savings in winter had a meaningful impact on the total yearly energy demand.
As noted previously [36], transpiration is higher under HPS lamps compared to LEDs. This effect was seen both in winter and summer (Fig. 10) and resulted in a higher ventilation rate in the HPS greenhouse, which increased the heating demand in winter. In summer, the main driver for ventilation when the lamps were on was excess heat, but during those times the choice of lamp did not influence the heating demand of the greenhouse, which was minimal for both HPS and LED lighting. However, there is an advantage for LEDs during warm days when supplementary lighting is needed, as these lamps reduce the problem of over-heating the greenhouse.

Influence of model assumptions and settings
The focus of this study was the energy balance of the greenhouse, while the influence on crop growth and yield were not studied in detail. A main assumption during simulations was that the crop behaves similarly under HPS and LED lights. In practice, the different spectral outputs of HPS and LEDs could have an influence on crop behavior (see Section 4.5). Furthermore, the HPS greenhouses required more ventilation, which resulted at times in lower indoor CO 2 concentration (Fig. 11), which could lead to reduced growth under HPS lamps.
While this study focused on energy use alone, further studies may consider the influence of the lamps on crop growth and development. As described in Section 2.3.3, crops under LEDs may require different temperature settings to achieve growth and development comparable to crops under HPS lamps. Here, we considered the effect of increasing the air temperature by 1 • C whenever the LEDs were on and examined the influence on energy demands. Including this temperature adjustment reduced the expected relative energy savings by around 1.5% (Figs. 7,8), indicating that it has a relatively small influence on total energy savings.
Weather inputs used in this study describe standard meteorological years in each of the considered locations. These datasets were compiled between the years 1999 and 2008 and may not take into account recent changes in climate such as rising temperatures. Nevertheless, our findings regarding the energy savings when transitioning to LEDs were consistent throughout all weather scenarios considered (Fig. 7), and thus provide a reliable initial assessment. When using the framework offered here, users may provide their own weather scenarios as input, which may be more recent standard meteorological years, measured data, or predicted scenarios of a future climate.
The study was based on model simulations performed using the GreenLight model, which has been tested against greenhouse data and found to estimate the greenhouse's heating needs by an error of 1-12% [29]. To the best of our knowledge, this is the only process-based model which includes both HPS and LED lighting and has been evaluated against energy use measurements. As such, it provides the best currently available predictions regarding the energy use of illuminated greenhouses. GreenLight is offered in an open-source format and can therefore be used by more researchers to evaluate against measurements. Further studies should continue to evaluate the model in various climates and greenhouse settings, and validate our findings using data from commercial greenhouses.

Further opportunities
The current study focused on the energy savings that can be achieved by switching from HPS lamps to LEDs, while maintaining all other factors (light intensity, light duration, air temperature) equivalent. However, LEDs offer a vast range of new opportunities in the greenhouse. One of the reasons direct comparisons between HPS lamps and LEDs are rare is that in fact, LEDs offer the ability to increase the supplemental light intensity or duration in the greenhouse, in a way that is impossible with HPS lamps alone. Using HPS lamps in warm weather [42], or with a very high lamp intensity [43], could result in too high temperatures inside the greenhouse, especially if a blackout screen is used, which in some regions is required in order to prevent light pollution to the environment. Growers often use LEDs in addition to HPS lamps in order to increase supplemental lighting [44]. The trend is thus towards a higher use of supplemental lighting in greenhouses [5]. Further studies should examine how energy efficiency can be maintained, or even improved, while supplying increasingly more light.
LEDs offer new possibilities on greenhouse and crop manipulation that are impossible with HPS lamps. It has been suggested that dynamic control of the intensity of LEDs may reduce the energy consumption for lighting while achieving similar yields [45] and increasing profits [46]. The ability to control the spectrum of light emitted by LEDs can offer greater control of the crop, realizing more efficient production. In particular, far-red light (701-750 nm) has been shown to have the potential to increase growth through a higher photosynthesis rate [47], promote light capture due to higher leaf area [48], and increase the partitioning of assimilates to fruits [49], although at a possible cost of reduced resistance to disease [50].
Transpiration played an important role in the greenhouse energy balance by converting sensible heat to latent heat (Figs. 8, 10) and by driving ventilation needed for dehumidification (Fig. 11). Further studies on the influence of lamp type and light spectrum on crop transpiration will increase the accuracy of the GreenLight model and its applicability to a wider range of scenarios.
Furthermore, the greenhouses considered in this study used passive ventilation through the windows to control humidity, which is the traditional dehumidification method in northern latitudes [1]. While our analysis considered energy inputs to the heating and lighting systems, in effect some of the energy demand of the heating system was used for dehumidification, when both heating and ventilation were used simultaneously to reduce the indoor relative humidity. It is estimated that 10-20% of the energy demands of the greenhouse are due to humidity control [51]. Future studies may further analyze how the lighting system influences the energy demand for dehumidification, for instance by comparing various methods for humidity control under different lamps.
The advantage of using a model such as GreenLight for assessment of the greenhouse behavior is that it offers a systematic approach, which considers the various greenhouse balances and components. Since GreenLight is a process based, open source model, it can be used to examine scenarios that were not included in the current study, such as any particular greenhouse attribute or local climate. Furthermore, new model components can be integrated into the model to include crop physiological mechanisms, as well as novel control strategies or other developments in greenhouse technology.

Conclusion
Simulations using a process-based greenhouse model were performed in order to evaluate how much energy can be saved in greenhouses by transitioning from high-pressure sodium (HPS) to lightemitting diode (LED) lighting, and how these savings were affected by greenhouse control, design, and outdoor climate. The key findings of the study were as follows: 1. A transition from HPS lamps with an efficacy of 1.8 μmol J − 1 to LEDs with an efficacy of 3 μmol 2. J − 1 resulted in a 40% saving on the greenhouse's lighting demand. However, in all cases, the LED greenhouse required more heating than the HPS greenhouse. Since heating and lighting are often derived from different energy sources, a detailed analysis