Performance Evaluation of a Commercial Greenhouse in Canada Using Dehumidification Technologies and LED Lighting: A Modeling Study

Abstract

In this study, a lumped parameter model, developed and extensively validated by the authors, is used to simulate the impact of three different dehumidification technologies (mechanical refrigeration dehumidifier, liquid desiccant dehumidifier, and a heat recovery ventilation unit), at a commercial greenhouse growing potted roses in southwestern Ontario, Canada. Typical meteorological year (TMY) data from nearby Vineland, Ontario was used to provide the external weather data used in the model. Each greenhouse bay containing a dehumidification unit was simulated for spring, fall, and winter conditions. The potential reductions in energy use (kWh), greenhouse gas emissions (kg CO_2e), and operating cost were estimated for each test case. The potential energy savings from switching from high-pressure sodium (HPS) to light-emitting diode (LED) lights were also examined. The simulation results showed that switching to LED lamps could reduce the electrical energy usage by up to 60% but would increase the space heating requirements. The expected energy-savings from using dehumidification equipment and switching from HPS to LED lighting in Canadian greenhouses is underrepresented in the literature. With the industry growing in the region, this study provides insight into the expected impact that these systems will have on the energy use in commercial greenhouses.

1. Introduction

The province of Ontario has a large and growing greenhouse sector that accounted for 60% of the greenhouse area in Canada and generated 70% of the greenhouse farm gate value in 2017 [1]. Ontario greenhouse production is experiencing rapid growth recently due to the demand for high-quality, locally grown vegetables, flowers, and cannabis [2]. In 2018, the Ontario greenhouse sector consumed 7.5 TWh of energy, 73% in the form of natural gas and 18% as electricity [1]. The heating energy use in Ontario is highest at the beginning and end of the season; increasing the year-round production would make the energy costs an even greater share of the total costs [3]. Ontario growers must also contend with hot and humid summer conditions when solar radiation can generate excess heat in greenhouse environments [4]. The control of excess heat or humidity in conventional greenhouses is usually achieved by venting, however, dehumidification technologies can be considered to reduce the simultaneous loss of sensible heat via ventilation. Ontario greenhouses need to provide supplemental light during the late fall, winter, and early spring, when natural lighting is insufficient for many crops, most commonly using traditional high-pressure sodium (HPS) lamps or more efficient Light Emitting Diodes (LEDs).

Three different dehumidification technologies were considered in this study. The mechanical refrigeration dehumidifier (MRD) removes water vapor from the air using an electrically-driven refrigeration cycle: humid greenhouse air is passed over a surface cooled by a refrigerant that is below the dew point temperature of the air [5]. The condensation occurs that also releases latent heat into the greenhouse air, which can reduce the heating loads.

An MRD unit was found to be more effective at controlling humidity than an air-to-air heat exchanger at a Saskatchewan, Canada greenhouse, particularly when the outdoor air was very humid [6]. The MRD required less maintenance and reduced the plant losses caused by high humidity from 43.3% to 0.9% [6]. As the MRD released heat into the greenhouse from condensation and electrical waste heat, it was found that 47.8% and 30.3% of the electricity costs of running the unit in 2012 and 2013, respectively, were recovered as heat released into the greenhouse [6].

An MRD-equipped greenhouse (2320 m²) in France was monitored during the winter conditions [7]. The analysis showed that using the MRD decreased the heating requirements and prevented the leaf temperatures from reaching the dew point [7]. It was estimated that the unit consumed 6 to 8.5 times less energy than traditional heating-venting [7]. However, the authors noted that the system economic feasibility was highly dependent on natural gas and electricity prices, and a reduction in overall energy use may not necessarily translate into financial savings.

Liquid desiccant dehumidification (LDD) removes water vapor from the air using a desiccant solution, typically calcium chloride, lithium bromide, or lithium chloride [5]. The water vapor in the air is absorbed into the solution due to the difference in the vapor pressure deficit (VPD), which results in the dilution of the desiccant material [5]. The desiccant solution is regenerated using a heat source to desorb and remove the water from the system.

Italian researchers compared two greenhouses over three years that were identical, except one had an AGAM 1020 liquid desiccant unit, and found winter energy savings of 9.6% to 15.6% in the LDD-equipped greenhouse [8]. When the inside-to-outside temperature difference decreased, the unit could supply up to 100% of the energy demand required for space heating [8].

Heat recovery ventilation (HRV) removes water vapor from the greenhouse through air exchange. The outgoing air is used to warm the incoming air through a heat exchanger, reducing the heat loss. The effectiveness of HRV is very dependent on outdoor air conditions. An HRV-equipped greenhouse in Saskatoon, Canada was effective at managing the humidity in the colder seasons but was not as effective in the summer during the early morning and night [9]. It was estimated that the use of the HRV reduced the percentage of time when the RH exceeded 75% in the summer to 25.7%, compared with 68% without the unit [9].

Researchers in Sweden evaluated the impact of an HRV in a naturally ventilated tomato greenhouse [10]. Simulation results with the HRV saw the thermal energy reduced by 15% to 17% during April to September [10]. The results suggested that additional incoming air preheating may be necessary in the cold weather to prevent a drop in the indoor temperature [10]. Some sources suggest that the HRV units may become clogged with ice when used in the cold weather (below −6 °C), which can be mitigated when the units have a defrost cycle [11]. A bylaw in Whitehorse, Yukon, Canada states that the HRV units in the city must have a sensible heat recovery of 64% when the outdoor air temperature is −25 °C [12].

Another possible energy saving technology available for northern greenhouses is switching from HPS lamps to LED lighting. While electrical energy savings typically can be accurately estimated using the lamp lighting efficacy (μmol PAR·J⁻¹), each lamp type also releases different amounts and types of heat into the greenhouse, potentially changing the amount of needed supplemental heating. Most greenhouse energy models in the literature that consider supplemental lighting [13] assume 40 to 50% of the electrical energy used by the lamp is released to the greenhouse as heat for LEDs, and 60 to 70% for HPS lamps [14]. These values include only radiative and convective heat transfer from the lamp and exclude the portion of electrical input that is converted to photosynthetically active radiation (PAR). However, much of the PAR from a lamp that reaches the crop canopy can be assumed to eventually become heat (typically latent). Typically, less than 3% of the PAR reaching the crops is converted to biomass [15], while the rest is either reflected or absorbed by the canopy.

The estimated energy savings associated with converting from HPS to LED lighting in greenhouses was examined by Dutch researchers [16] using a validated greenhouse model (GreenLight) to estimate the heating and lighting energy requirements with either LED or HPS lamps [16]. Switching to LEDs was associated with an estimated 40% average reduction in the lighting electricity use, and an overall annual energy savings of 10% to 25% when accounting for the differences in supplemental heating requirements [16]. The GreenLight model was designed to predict the energy use of greenhouses with LED and HPS lights, and the validation suggested an error range on predicting heating requirements of 1% to 12% [16]. However, the model was only validated at a single greenhouse and the upper error range of the estimated energy usage is quite high for seasonal energy simulations [17].

In the current study, the impact of using three dehumidification units (MRD, LDD, and HRV) in the spring, fall, and winter conditions was simulated using a greenhouse energy model in the southwestern Ontario context. The seasonal energy impact of using LED lights instead of HPS lamps was also examined. The potential energy (kWh), greenhouse gas (kg CO_2e), and cost ($CAD) reductions were estimated for each test case. The expected energy savings that Canadian growers can expect to achieve by using dehumidification equipment and switching from HPS to LED lighting is underrepresented in the literature. This study seeks to address this lack of information using a novel greenhouse energy model that has been extensively validated against the measured data.

2. Methodology

2.1. Site Information

This study examines a potted rose greenhouse with exhaust fans on the south wall (560 W for each fan). Ceiling energy curtains were present, while supplemental lighting was provided by 600 W HPS fixtures. The supplemental heating was delivered via hot water pipes at three different heights. Evaporative cooling pads were present on the north wall but only used in the summer. CO₂ burners were extensively used during the daytime in the winter to increase CO₂ concentration (units released 0.38 kg CO₂·min⁻¹), but infrequently used in the other seasons. Figure 1 shows a schematic of one of the bays (which had the MRD unit installed).

Table 1. Greenhouse properties.

	Section A	Section B	Section C
Length (m)	65.8	65.8	65.8
Width (m)	25.6	25.6	38.4
Area (m2)	1684	1684	2527
Gutter height (m)	4.3	4.3	4.3
Roof height (m)	6.0	6.0	6.0
Volume (m3)	8675	8675	13013
Crop covered fraction	0.8	0.8	0.8
Crop type	Potted rose	Potted rose	Potted rose
Roof material	Double layer PE	Double layer PE	Double layer PE
Number of exhaust fans	8	8	12
Number of HPS fixtures	120	120	150
Dehumidification type	MRD	LDD	HRV

lists the dimensions and other information of each of the three greenhouse bays considered in the study.

2.2. Model Configuration

This study used a lumped parameter model developed and extensively validated by the authors and described in detail in previous work [18,19,20]. The model predicts the time-series temperature of various greenhouse elements, such as the indoor air and soil, based on external weather data and system controls, using a simulation time step of six seconds. It predicts the absolute humidity of the indoor air based on the crop type and greenhouse controls (ventilation, dehumidification technology, CO₂ burners). The model has been validated in each season against the experimental data from the site, with strong correlation for both the indoor air temperature and the RH [18,20]. It has also been validated successfully at six other sites in all seasons. In all studies involving commercial greenhouses, the mean absolute error (MAE) between the measured and simulated air temperatures was less than 2 °C, while the MAE for the RH was less than 8% [20]. For many greenhouses studied with the model, the MAE for air temperature and the RH were less than 1 °C and 4% [20]. One limitation with environmental data collected from commercial greenhouses is that the sensor calibration history is not usually available to the research team.

There are several advantages to the model used in this study compared with others in the literature [13,17,21]. As mentioned earlier, the model has been validated in each climatic season at seven unique greenhouses, ranging greatly in complexity and site properties. Most literature models are validated for a short duration at one site, leading to uncertainty about the model accuracy in other seasons and at different locations. Another advantage is that the current model can account for the environmental control technology common at commercial greenhouses in Ontario, such as evaporative cooling pads, supplemental heating and lighting, plant evapotranspiration, dehumidification technology, and curtains.

Figure 2 shows the heat transfer pathways considered in the model with the energy curtain at the site deployed. Corresponding equations are given in Appendix A (slightly different pathways exist when the curtain is not in use), and each heat transfer term can be seen in the nomenclature table in Appendix C. An energy and mass balance are performed at each time step (six seconds) to find the changes in the temperature and absolute humidity. The equations used to find the temperature change of each layer, and the absolute humidity change of the indoor air, can be seen in the Appendix A.

Table 2. Summary of parameters used in the model.

Parameter	Value	Variable	Parameter	Value	Variable
Glazing IR transmissivity	0.1	${\tau }_{IR, 1}$	Floor thermal conductivity (W·m−1·K−1)	1.9	$k_{14–21}$
Glazing solar radiation reflectivity	0.21–0.5	${\alpha }_{sol 1}$	Floor heat capacity (J·kg−1·K−1)	880	$c_{14–21}$
Glazing solar radiation transmissivity	0.48–0.77	${\tau }_{sol, 1}$	Floor and soil IR emissivity	1.0	${\epsilon }_{IR, 8,14}$
Glazing heat capacity (J kg−1·K−1)	2500	$c_1$	Air heat capacity (J·kg−1·K−1)	1007	$c_{2, 4,6}$
Glazing density (kg·m−3)	905	${\rho }_1$	Air density (kg·m−3)	1.16	${\rho }_{2, 4,6}$
Glazing thermal conductivity (W·m−1·K−1)	0.45	$k_1$	Air thermal conductivity (W·m−1·K−1)	0.0263	$k_{2,4,6}$
Curtain density (kg·m−3)	300	${\rho }_5$	Thermal conductivity of potting soil (W·m−1·K−1)	1.25	$k_{8–13}$
Soil and floor solar radiation reflectivity	0.5	${\alpha }_{sol 8, 14}$	Plant solar radiation reflectivity	0.16	${\alpha }_{sol 7}$
Floor density (kg·m−3)	2300	${\rho }_{14–21}$	Plant IR emissivity	0.9	${\epsilon }_{IR, 7}$
Plant density (kg·m−3)	1421	${\rho }_7$	Plant heat capacity (J·kg−1·K−1)	4180	$c_7$
Curtain specific heat capacity (J·kg−1·K−1)	1800	$c_5$	Curtain solar radiation reflectivity	0.07	${\alpha }_{sol 5}$
Curtain solar radiation transmissivity	0.88	${\tau }_{sol, 5}$	IR transmissivity of curtain	0.1	${\tau }_{IR, 5}$
Curtain IR emissivity	0.5	${\epsilon }_{IR,5}$	Glazing IR emissivity	0.9	${\epsilon }_{IR, 1}$
Curtain IR reflectivity	0.07	${\alpha }_{IR, 5}$	Glazing IR reflectivity	0.07	${\alpha }_{IR, 1}$
Density potting soil (kg·m−3)	1300	${\rho }_{8–13}$	Potting soil specific heat capacity (J·kg−1·K−1)	1350	$c_{8–13}$

lists the properties used in the model, which were obtained from multiple sources [21,22,23,24,25]. The solar transmissivity and reflectivity vary based on the solar elevation angle (Table 2).

2.3. Model Controller Configuration

A controller was modelled to realistically set the supplemental heating, lighting, ventilation, curtain, and dehumidification use based on the air temperature and RH setpoints. The CO₂ burners were assumed to run during the daytime in the winter. The upper and lower air temperature setpoints were 21 °C and 25 °C, while the RH setpoints were 65% and 85%. Outside weather data was from an hourly typical meteorological year (TMY) file for Vinland, Ontario [26]. If the simulated indoor air temperature was above or below the setpoint range, the amount of heat per unit floor area to be provided or removed was

$$\Delta Q=\{\begin{array}{cc}\frac{\left(T_{air}-25 ℃\right) c_p {\rho }_{air} V_{GH}}{dt A_{GH}}& T>25 ℃\\ \frac{\left(21 ℃-T_{air}\right) c_p {\rho }_{air} V_{GH}}{dt A_{GH}}& T<21 ℃\end{array}$$

(1)

where T_air represents the air temperature, in °C; ρ_air represents the density of the air, in kg·m⁻³; c_p is the specific heat capacity of the air, in J·kg⁻¹·K⁻¹; dt is the time step, in seconds; and V_GH and A_GH are the volume and surface area of the greenhouse, in m³ and m², respectively. Each of the three greenhouse heating pipes was capable of providing up to 80 W·m⁻² of heat. When the air temperature was below the minimum set point, it was ensured that the amount of heat being supplied by the heating pipes was at least the value calculated in Equation (1).

When the air temperature was too high, the exhaust fans were utilized, assuming the amount of cooling from a single fan was

$$Q_{fan}=\left(\frac{f}{V_{GH}}\right)dx c_p {\rho }_{air}(T_{air}-T_{outside})$$

(2)

where f is the fan flow rate, in m³·s⁻¹; dx is the vertical thickness of the air layer, in m; and T_outside is the outdoor air temperature, in °C.

A dynamic energy balance (DEB) was performed to help minimize the amount of time the air temperature was outside the setpoint range, by calculating the difference between the heat entering and exiting the greenhouse envelope, in W·m⁻². The heat entering the greenhouse came from solar radiation, supplemental heating and lighting, and dehumidification operation, while the heat exiting the greenhouse consisted of the ventilation and leakage heat loss, the infrared (IR) heat lost between the greenhouse components and the sky, the convective heat transfer from the glazing to the ambient air, and the heat used for crop evapotranspiration. The amount of heating and ventilation required when the temperature was within the setpoints was then based on the magnitude of the net energy entering or leaving the greenhouse.

If the RH was too high, ventilation and/or supplemental heating was provided, while the heating and/or ventilation was decreased when the RH was too low. For test cases involving the dehumidification units, the dehumidifiers were run if the RH was above 64%.

The latent heat of condensation (for the MRD and LDD units) and sensible waste heat were released into the greenhouse when dehumidifiers were operating. The total heat input was

$$Q_{DH}\left(\frac{W}{m^2}\right)=\frac{\left(0.9 D{H}_{power}+\lambda 1000 \dot{M}\right)}{A_{GH}}$$

(3)

where DH_power is the electrical consumption of the unit, in W; λ is the latent heat of condensation (2500 J·g⁻¹); and $\dot{M}$ is the mass of water removed by the unit every second, in kg·s⁻¹. It was assumed 90% of the electrical input was released as sensible heat [6]. The HRV did not condense the water vapor, so only its electrical power was considered in Equation (3).

2.4. Model Controller Logic

Table 3. Controller logic with dehumidifier in use.

Temperature	RH	Supplemental Heat	Ventilation	Dehumidification
Above setpoint	Above setpoint	None	Equations (1) and (2)	ON
Within setpoints	Above setpoint	ΔQi−1 d	DEB	ON
Within setpoints	Below setpoint	ΔQi−1 d	DEB	OFF
Below setpoint	Above setpoint	Equation (1)	None	ON
Above setpoint	Below setpoint	None	Equations (1) and (2)	OFF
Above setpoint	Within setpoints	None	Equations (1) and (2)	ON
Below setpoint	Below setpoint	Equation (1)	None	OFF
Below setpoint	Within setpoints	Equation (1)	None	ON
Within setpoints	Within setpoints	ΔQi−1 d	DEB	ON

summarizes the simulated controller logic. The logic for the HRV system varied slightly, with air exchange occurring between the indoor and outdoor air, while the MRD and LDD units condensed the moisture directly in the greenhouse. It was assumed the supplemental lights (HPS or LED) were turned off from 12:00 a.m. to 2:00 a.m. every night, and then ran until the target daily light interval (DLI) of 15 mol PAR·m⁻²·d⁻¹ was reached. In Table 3, ΔQ_i−1 represents the supplemental heat supplied at the previous data time step, which is multiplied by a fraction, d (0–1) depending on the dynamic energy balance (DEB) result at each time step and the simulated RH value. If excessive energy was entering the greenhouse envelope, but temperature was still within the setpoints, an appropriate number of exhaust fans were deployed. This prevented rapid oscillations in the simulated air temperature and RH values. With the dehumidifiers running, less ventilation was required to manage the humidity. The simulated air temperature took precedence over the simulated RH when determining the heating and venting schedules.

2.5. Dehumidification Model

Constant power consumption was assumed for the three dehumidification units based on analyzing the site data, and the moisture removal rates for the MRD and LDD were assumed as constant. The HRV moisture removal rate depends on indoor and outdoor conditions and required calculations at each time step.

Table 4. Summary of dehumidification units (n is number of data points available).

Unit	n	Unit Power (kW)	Uncertainty (+/− kW)	Moisture Removal Rate (kg·hr−1)	Uncertainty (+/− kg·hr−1)
MRD	1934	10.00 (electrical)	1.64	35.80	6.32
LDD	614	2.40 (electrical) 18.9 (hot water)	0.28 3.20	12.96	1.80
HRV	673	0.780 (electrical)	0.036	N/A	N/A

shows a summary of the power consumption and moisture removal rates of the three units from the measured data. Each of the three dehumidification units can be seen in Appendix D, along with schematics showing the components and processes.

The temperature ratio η_HRV of HRV systems (in %) can be defined as [27]

$${\eta }_{HRV}=100\%\times \left(\frac{T_{supply}-T_{outside}}{T_{inside}-T_{outside}}\right)$$

(4)

where T_supply is the temperature of air leaving the unit and entering the greenhouse. The measured temperature ratio typically varied between 30% and 60%. The temperature ratio was modeled as a linear relationship based on the fit to the temperature difference between the indoor and outdoor air taken from measured data shown in Figure 3.

The supply air temperature could then be approximated using Equation (4). The amount of sensible heat transfer from the HRV could then be found using the flow rate, f_HRV, which was assumed to be constant at 2.4 m³·s⁻¹ based on experimental values:

$$Q_{HRV}=\frac{f_{HRV} {\rho }_{air} c_p\left(T_{inside}-T_{supply}\right) }{A_{GH}}$$

(5)

The value of Q_HRV was then subtracted from the air layer energy balance at each time step, shown in Equation (A5) in Appendix A.

The specific humidity (kg moisture/kg dry air) of the inside air and the air entering the greenhouse are required to find the amount of moisture removed by running the HRV. The specific humidity of the indoor air (q_in; mass of moisture over mass of dry air) is found using the atmospheric pressure P [28].

$$q_{in}=\frac{0.622 e}{P-e \left(1-0.622\right)}$$

(6)

The vapor pressure e, in Pa, is [29],

$$e=610.94 \left(\frac{R{H}_{in}}{100\%}\right) e^{\frac{17.625+T_{in}}{243.04+T_{in}}}$$

(7)

where T_in is the indoor temperature, in °C. The moisture removal rate is then:

$$\dot{M}=f_{HRV} \left(q_{in}{\rho }_{in}-q_{supply}{\rho }_{supply}\right)$$

(8)

where q_supply and ρ_supply are the specific humidity and density of the supply air. An empirical relationship was derived to calculate the specific humidity of the air leaving the HRV and entering the greenhouse. Ideally, the specific humidity of the supply air would equal the outside air specific humidity, but the system leakage would mean this was not always the case. An empirical linear fit of the measured outdoor specific humidity versus the supply air specific humidity was used:

$$q_{supply}=0.605 q_{outside}+0.0054$$

(9)

2.6. Lighting Model

The DLI target was set to 15 mol PAR·m⁻²·d⁻¹ at the canopy level based on information from the grower. The solar radiation entering the greenhouse envelope and reaching the crop canopy was converted from units of W·m⁻² (G) to μmol PAR·m⁻²·s⁻¹ using an empirical equation derived by the researchers, and seen below, where τ_glazing is the solar transmissivity of the glazing, and lux and foot-candle are the units of illuminance [30,31].

$$L_{sol} \left(\frac{\mu mol PAR}{m^2 s^1}\right)=\left(\left(122 {\tau }_{glazing} G+50\right)\left(\frac{lux}{\frac{W}{m^2}}\right)\right) \left(\frac{foot-candle}{10.76lux}\right)\left(\frac{0.20 \frac{\mu mol PAR}{m^2 s^1} }{foot-candle}\right)$$

(10)

The site includes 120 HPS lamps in Section A (which is used for the lighting simulations) that deliver approximately 85 μmol PAR·m⁻²·s⁻¹. For continuity, this value was maintained when simulating the LED test cases by changing the assumed number of light fixtures to maintain this intensity.

Table 5. Properties of HPS and LED lights used in model.

Lighting	Lighting Model	Efficacy (μmol PAR·J−1)	Photosynthetic Photon Flux (μmol PAR·s−1)	Electrical Input Power (W)	Number of Fixtures Required
LED A	GLPI630HU660D11	4.06	2436	600	47
LED B	SPS 640-GL1	2.80	1745	626	66
HPS	HPS (Existing)	1.60	950	600	120

gives details of the two commercially available LED lights included in the analysis [32]. This information was taken from a database reporting grow light experimental test results, which often differ from manufacturer-reported values [32].

The color spectrum intensity of the LED lights considered in the study are known, and are shown in

Table A1. Color spectrum of LED lights used in model [32].

Lighting Model	Photon Flux Blue (400–500 nm)	Photon Flux Green (500–600 nm)	Photon Flux Red (600–700 nm)	Photon Flux Far Red (700–800 nm)
GLPI630HU660D11	0.75 μmol/s	1.62 μmol/s	2433 μmol/s	6.65 μmol/s
SPS 640-GL1	322 μmol/s	734 μmol/s	689 μmol/s	36 μmol/s

of Appendix B. While the exact lighting recipe for the existing HPS lights is not directly known, it is typical for HPS lights to emit light in the red and green spectrum (not much in the blue spectrum). This spectrum can be characterized as having a color temperature of 3500 K to 4500 K and being a warm white in appearance.

It is widely recognized that using HPS lamps reduces the greenhouse heating load compared with LED lights. This is partly because LEDs are more efficient, so the same photon flux is delivered with less electrical power than HPS lamps, and partly because the radiant heat (FIR and NIR) emitted by HPS lamps is reflected downwards towards the crop canopy, resulting in higher crop temperatures than LEDs [17,33]. In the model, the electrical input to LED lights was partitioned into convective (26%), radiative (27%), and PAR light (47%) energy outputs, while for HPS lights the energy output was 11% convective, 55% radiative, and 34% PAR light [33]. These values are similar to those used by [17], with one main difference being that no convective cooling was assumed to occur in LED lights. For HPS lamps, the radiative heat component (FIR and NIR) was added to the solar irradiance reaching the crop canopy and floor layers, while for LEDs, the radiative component was assumed to be released to the air [33]. In both HPS and LED test cases, the convective proportion was released to the air layer, while the bulk of the PAR light flux (88%) was added to the crop layer as a heat transfer term, with 3% converted to biomass, and the remainder being reflected and leaving the greenhouse [15].

3. Results

3.1. Controller Validation

Simulations using the controller were tested against the measured data with and without the dehumidifiers in use. The root mean squared error (RMSE) and the mean absolute error (MAE) between the simulated and measured results were calculated for the air temperature, absolute humidity, and RH. The amount of input energy predicted was compared with the actual energy use recorded in the data. Each test case was a week in duration. The natural gas used by CO₂ burners was included in the analysis.

Figure 4 shows the simulated versus the measured air temperature and absolute humidity with the MRD running in the winter as an example.

Table 6. Results of validation test cases. Percent difference shown in parentheses.

Case	Electrical Consumption (kWh)	Natural Gas Consumption (kWh)	RMSE Air Temperature (°C)	MAE Air Temperature (°C)	RMSE RH (%)	MAE RH (%)	RMSE AH (kg·m−3)	MAE AH (kg·m−3)
MRD on (12/23/2019 to 12/30/2019)	Data = 12.119 Model = 12.777 (5.28%)	Data = 13.136 Model = 12.925 (1.62%)	0.95	0.81	5.28	2.95	0.0013	0.0010
MRD off (12/31/2019 to 1/6/2020)	Data = 11.867 Model = 11.121 (6.49%)	Data = 24.209 Model = 24.955 (3.03%)	0.76	0.55	4.71	3.57	0.0010	0.0007
LDD on (12/23/2019 to 12/30/2019)	Data = 10.802 Model = 11.505 (6.30%)	Data = 16.499 Model = 15.755 (4.61%)	1.03	0.89	4.35	3.05	0.0011	0.0008
LDD off (12/31/2019 to 1/6/2020)	Data = 11.864 Model = 11.120 (6.47%)	Data = 20.506 Model = 20.757 (1.22%)	0.74	0.56	4.19	3.01	0.0011	0.0009
HRV on (12/23/2019 to 12/30/2019)	Data = 10.510 Model = 11.184 (6.21%)	Data = 31.445 Model = 30.433 (3.27%)	1.14	0.83	4.46	3.25	0.0013	0.0010
HRV off (12/31/2019 to 1/6/2020)	Data = 11.852 Model = 11.143 (6.17%)	Data = 33.204 Model = 33.486 (0.85%)	1.02	0.69	4.98	3.91	0.0015	0.0011

summarizes the validation results for all six test cases, including the predicted and measured energy consumption values (in kWh) for natural gas and electricity. The percentage difference between the simulated and measured electrical energy consumption ranged from 5.28% to 6.49%, while the differences for natural gas ranged from 0.85% to 4.61%. The simulated RH never exceeded 95%, and rarely exceeded 90%. Figure 4 shows that the simulated absolute humidity values drop each night at midnight, which is due to the HPS lights being turned off for two hours. In the measured data, the lights were on all night, and were occasionally turned off during the day, suggesting a reason for the discrepancy between the absolute humidity measurements and simulations.

3.2. MRD Results

The TMY data was used for the spring, fall, and winter to run test cases with and without the MRD in use at Section A. The summer was neglected due to the high ventilation requirement for temperature control. As mentioned above, it was assumed that the MRD removed 35.80 L·hr⁻¹ of the moisture from the air and consumed 10 kW of electricity.

Table 7. Summary of MRD results.

Season	MRD Status	Electrical Energy Consumption (kWh)	Natural Gas Energy Consumption (kWh)	kg CO2e	Energy Saved with MRD in Use (kWh)	Range of Savings ($CAD)
WINTER	On	132.389	238.234	50.137	67.405 (15.4%)	−990 to 3272
	off	112.733	325.295	64.853
FALL	on	135.177	126.931	30.139	78.654 (23.1%)	−818 to 3833
	off	115.314	225.448	46.918
SPRING	on	95.080	64.345	16.686	49.646 (23.8%)	−1002 to 2397
	OFF	78.114	130.957	27.843

summarizes the energy results, along with a range of possible savings ($CAD), and an estimate of CO₂ emissions (kg CO_2e). Two electrical rate constants were used (0.055 $·kWh⁻¹ and 0.127 $·kWh⁻¹), which were obtained from discussions with growers. A range of natural gas prices was used based on the typical rates, with $4.80·GJ⁻¹, $10.00·GJ⁻¹, and $14.00·GJ⁻¹ considered [34]. The carbon intensity in g CO_2e·kWh⁻¹ for the natural gas and electricity in Ontario was assumed to be 181 and 53, respectively [35].

Running the MRD saved energy (the sum of electrical and natural gas) in each season tested compared with the non-use cases, due to the reduction in the sensible heat loss and the capture of latent heat. Cost savings depend heavily on natural gas and electricity rates, with the cheaper electricity and expensive natural gas resulting in larger savings. Figure 5 shows the cost of removing 1 L of moisture with the MRD or with one exhaust fan during one week in winter using two sets of electricity and natural gas rates. It was assumed that the heat lost from running the exhaust fan was entirely from the supplemental heating (i.e., natural gas), when in reality, some of it would be from solar radiation or waste heat from the HPS lamps. Thus, when calculating the cost to remove one liter of water vapor from the greenhouse air using an exhaust fan, natural gas rates were used instead of electrical. It costs more to remove 1 L of moisture with the MRD compared to an exhaust fan when electricity rates are high, but this reverses when the lower rate is used. This means that the feasibility of implementing the MRD at a particular greenhouse is highly dependent on the energy rates paid by the grower.

A comparison was made to experimental data collected from the site during the early spring conditions [36] (

Table 8. Comparison between experimental data and simulation results for MRD spring test case.

	MRD Status	Average RH (%)	Unit Power (kWh)	Heat (kWh)
12–26 March 2019 [36]	ON	78.2	1450	15.024
	OFF	81.3	N/A	20.267
	Savings (%)	N/A	N/A	25.9
20–27 March TMY Current Study	ON	75.9	1464	11.102
	OFF	77.6	N/A	17.852
	Savings (%)	N/A	N/A	37.8

). In the comparative study [36], the authors conducted ON/OFF trials for two weeks with the dehumidification units (MRD, LDD, HRV). However, a larger seasonal analysis was not performed. It is impossible to experimentally measure the energy savings from running the MRD versus not running the MRD at a particular greenhouse bay at the same time, with the same weather conditions. This adds to the value of accurate, validated simulations, which can simulate the same external weather and site conditions with and without the unit running.

Table 8 shows that the MRD use can reduce the average indoor RH by 1–3% and reduce heating energy requirements due to the decreased ventilation. The simulation results showed a higher reduction in heating energy with the unit running than the experimental results. This can be partly attributed to the outdoor air temperature being warmer with the unit off (2.8 °C) than with the unit on (1.9 °C).

When installing an MRD, it must be properly sized for the greenhouse, and should be placed in the middle of the greenhouse to promote uniform temperature and humidity profiles [1]. As these units directly condense water in the greenhouse, they contribute both sensible and latent heat to the greenhouse environment. In the warmer weather, when high rates of ventilation are required for temperature control, the units are typically not run, as the treated air would be quickly removed through ventilation. As such, the spring, fall, and winter are the optimal seasons to operate the units in the southern Ontario context.

3.3. LDD Results

Table 9. Summary of LDD results.

Season	LDD Status	Electrical Energy Consumption (kWh)	Natural Gas Energy Consumption (kWh)	kg CO2e	Energy Saved with LDD in Use (kWh)	Range of Savings ($CAD)
WINTER	ON	117.393	246.110	50.768	43.179 (10.6%)	234 to 2136
	OFF	112.720	293.962	59.181
FALL	ON	119.929	139.170	31.546	55.068 (17.5%)	444 to 2730
	OFF	115.293	198.873	42.107
SPRING	ON	81.875	76.629	18.209	31.971 (16.8%)	121 to 1579
	OFF	77.932	112.543	24.501

shows the LDD results. While smaller energy savings were predicted with the LDD compared with the MRD, the lower electrical consumption of the LDD unit resulted in a decreased likelihood that negative cost savings would occur due to the high electricity rates. In all three seasons, it was found that running the LDD saved both energy and cost. The natural gas energy usage includes the heat energy required to regenerate the desiccant solution.

To illustrate the seasonal effect on the ventilation efficiency, a comparison between the early winter and early fall was performed (Figure 6). The amount of moisture that can be removed by the exhaust fans is higher in the winter when humidity differences between the indoor and outdoor air are greater.

Table 10. Comparison between experimental data and simulation results for LDD spring test case.

	MRD Status	Average RH (%)	Unit Electrical Energy (kWh)	Unit Regeneration Energy (kWh)	Heat (kWh)
12–26 March 2019 [36]	ON	84.3	368	3070	14.801
	OFF	87.7	N/A	N/A	20.136
	Savings (%)	N/A	N/A	N/A	26.5
20–27 March TMY Current Study	ON	77.7	357	2818	11.201
	OFF	78.9	N/A	N/A	15.609
	Savings (%)	N/A	N/A	N/A	28.2

shows a comparison of the measured data [36] and simulation results for the LDD. An excellent correlation is apparent between the simulations and measured data. The magnitude of the heat energy reduction using the LDD matched well with the measured data, as did the overall LDD power consumption.

Similar to the MRD, the LDD should be placed in the middle of the greenhouse for optimal performance. However, the LDD unit requires heat (typically delivered via hot water pipes) to regenerate the desiccant solution, which could complicate its placement in the greenhouse. The solar-driven regeneration of the desiccant solution has been examined in recent years, but the intermittent nature of solar radiation is a limitation in the widespread application [37]. As with the MRD, the unit is typically not operated in summer conditions, due to the high rate of ventilation for the temperature control.

3.4. HRV Results

Table 11. Summary of HRV results.

Season	HRV Status	Electrical Energy Consumption (kWh)	Natural Gas Energy Consumption (kWh)	kg CO2e	Energy Saved with HRV in Use (kWh)	Range of Savings ($CAD)
WINTER	On	102.158	430.274	83.294	21.446 (3.87%)	344 to 1071
	off	101.908	451.970	87.208
FALL	on	104.867	279.102	56.075	4048 (1.04%)	42 to 201
	off	104.612	283.406	56.841
SPRING	on	73.521	115.576	24.816	4463 (2.31%)	46 to 222
	OFF	73.240	120.320	25.660

shows simulation results for the HRV test cases. As the HRV consumed the least electricity of the three units (0.78 kW), the difference in the electrical consumption between the HRV on and off was the lowest of the three technologies. This resulted in cost savings in every scenario. Since the temperature efficiency ratio is less than 1.0, colder air was introduced into the greenhouse when the HRV was operating, which increased heating loads compared to the MRD and LDD units and resulted in lower CO_2e reductions. In addition, it is notable that the greenhouse bay with the HRV unit (Section C) had a surface area that was 33% greater than the bays with the MRD and LDD, resulting in higher total heating requirements. Note that the winter, fall, and spring in Table 7, Table 9 and Table 11 correspond to 21 December to 19 March, 22 September to 20 December, and 20 March to 20 June. In Table 7, Table 9 and Table 11, the natural gas energy refers to the energy required for space heating (and desiccant regeneration for the LDD), while the electrical energy is the amount of energy required to operate the supplemental lights, exhaust fans, and dehumidification units.

Table 12. Comparison between experimental data and simulation results for HRV spring test case.

	MRD Status	Average RH (%)	Unit Power (kWh)	Heat (kWh)
12–26 March 2019 [36]	ON	76.1	131	28.002
	OFF	70.8	N/A	28.465
	Savings (%)	N/A	N/A	1.6
20–27 March TMY Current Study	ON	70.3	45	20.614
	OFF	70.2	N/A	21.183
	Savings (%)	N/A	N/A	2.7

shows controller simulations compared with the corresponding measured data [36]. Both the simulation results and measured data showed a relatively small decrease in the heating energy when using the HRV unit, while the measured data showed that the average RH was higher when the unit was on than when it was off.

The effectiveness of an HRV is highly dependent on the outdoor weather conditions. When the ambient outdoor air is hot and humid, the system cannot adequately dehumidify the greenhouse [9]. During the wintertime, colder air may be introduced to the greenhouse, as the sensible heat recovery efficiency is not 100%. While the electrical energy required to operate the HRV was the lowest of the three units considered, it also had the smallest impact on reducing the energy consumption. Altering the controller logic could possibly improve the results and is worth exploring in greater detail in future work. In general, the optimal seasons for the use of HRV systems are the shoulder months in the spring and fall [9]. Installing an HRV unit typically requires additional duct work for incoming and outgoing air pathways. As the unit does not condense water vapor directly inside the greenhouse, the latent heat is lost from the greenhouse through ventilation. As with the MRD and LDD systems, the unit should be properly sized for the greenhouse, with a suggested moisture removal rate of 0.018 L·hr⁻¹·m⁻² for cool periods in the Canadian Prairies [9].

3.5. Lighting Results

The TMY data from each season (spring, summer, fall, and winter) were simulated for the LED and HPS light scenarios. Similar to the sections above, the amount of energy required for heating and for electricity (lights, exhaust fans) was tracked and is presented in

Table 13. Summary of lighting simulation results.

Season	Light Type	Electrical Energy Consumption (kWh)	Natural Gas Energy Consumption (kWh)	kg CO2e	Energy Saved Relative to Baseline HPS Use (kWh)	Range of Savings ($CAD)
	HPS	108.378	274.780	55.479	N/A	N/A
WINTER	LED A	43.325	355.728	66.683	−15.895	−469 to 6861
	LED B	61.905	336.522	64 191	−15.269	−531 to 4834
	HPS	111.091	171.864	36.995	N/A	N/A
FALL	LED A	44.905	238.385	45.528	−335	314 to 7255
	LED B	64.054	222.597	43.685	−3696	50 to 5096
	HPS	74.863	102.496	22.520	N/A	N/A
SPRING	LED A	31.634	122.597	23.867	23.127	1372 to 5142
	LED B	44.144	118.149	23.725	15.065	907 to 3630
	HPS	75.614	25.055	8542	N/A	N/A
SUMMER	LED A	34.454	31.977	7614	34.234	1918 to 5107
	LED B	46.368	30.575	7992	23.724	1332 to 3618

. The dehumidification equipment was assumed to be turned off for these simulations, and the evaporative cooling pads were only included in the summer simulations. The amount of heating required is higher for the LED test cases in each season than for the HPS cases, as expected. In both the fall and winter, the LED test cases required more total energy than the corresponding HPS test cases, due to the higher heating demand. However, since electricity is more expensive than natural gas on a unit energy basis, cost savings were possible in every season.

The presented results do not account for any differences in the product yield or quality that may be influenced by switching to LED lights. Researchers have found higher yields and improved quality comparing LED lighting systems with HPS for cucumber crops [38], which offers the potential for additional financial benefits that are not captured in this study. In addition, it is worth noting that the study greenhouse (growing potted roses) has a lower photosynthetic photon flux density (85 μmol PAR·m⁻²·s⁻¹) provided by the assimilation lighting than more intensive vegetable or cannabis operations. This means that the findings from this study could be magnified for greenhouses growing crops with a higher DLI requirement.

While the capital cost of investment is often the limiting factor in switching to LEDs at greenhouses in Ontario, incentives are available through programs such as Save on Energy, which can cover up to 50% of the retrofitting costs [39]. LED fixtures also last longer and require less maintenance than HPS lamps [14]. Growers can also use combinations of LED and HPS lamps, which mitigates the higher space heating requirements of switching to LEDs, particularly in the colder seasons. Furthermore, it is worth noting that due to lower crop temperatures with LED lamps, growers often raise the indoor air temperature by 1 °C, which would add to the space heating loads [17].

4. Conclusions

This study suggests good potential for applying the dehumidification technology in southern Ontario greenhouses, with energy savings possible from all three types of dehumidification studied in the winter, spring, and fall conditions. Switching to LED lights from HPS lamps was found to save energy in the spring and summer, with financial savings possible in every season. Major findings include:

• The results suggest that the best season to run the LDD dehumidification equipment is the fall (17.5% reduction in the overall energy), however, energy savings with the LDD were possible in each season studied and it was profitable in all costing scenarios.
• The greatest energy savings were found with the MRD in the fall (23.1%). Due to the higher electrical consumption, cost savings were highly dependent on the electricity and natural gas rates.
• The HRV, with the lowest electrical consumption, was profitable in all costing scenarios, but saved the lowest overall energy of the units studied. It would likely be less effective in summer conditions, as the cost saving effectiveness of the HRV is highly dependent on the humidity difference between indoor and outdoor air [9].
• Due to the LDD relying on both electricity and natural gas (for regenerating the desiccant solution), the cost effectiveness was less sensitive to the varying natural gas and electricity rates, and it offered strong energy savings potential compared with the baseline of dehumidification by the exhaust fan ventilation.
• Seasonal carbon emission (kg CO_2e) reductions of up to 35.8% were possible with the MRD unit, and significant reductions are also possible with the LDD.
• Non-operating costs, including capital and maintenance costs associated with each technology, were not quantified nor discussed in this study. With carbon taxes being implemented in Ontario, dehumidification technologies offer a potential pathway for growers to reduce CO₂ emissions, due to the low carbon intensity of the Ontario power grid. Seasonal reductions of up to 16,000 kg CO_2e are possible with the MRD, with lower reductions seen with the HRV. With the Canadian federal carbon tax increasing to $50/tonne CO_2e in 2022 [40], a 16,000 kg CO_2e reduction corresponds to savings of $800.
• At the study greenhouse, switching to LED lamps can save up to $7,000 in the fall, and up to $24,300 annually at the studied greenhouse, when higher electricity and lower natural gas rates are used. However, space heating loads increase in each season using LEDs instead of HPSs, and a net increase in the overall energy was found in simulation results for the winter and fall seasons.