Research on Thermal Environment of Container Farms: Key Factor Identification and Priority Analysis

Abstract

Container farms (CFs), a controlled environment agricultural technology designed to solve food insecurity, are receiving increasing attention from researchers. However, the complex geometric structures and artificial lighting used in CFs present challenges in effectively controlling the thermal environment. This study aims to identify the primary factors that impact the thermal environment of CFs while conducting factor ranking and significance analysis, providing a theoretical basis for future thermal environment optimization. The research method of theoretical analysis, CFD simulation, and an orthogonal experimental design were adopted to achieve the above objectives. Theoretical analysis revealed that factors influencing the thermal environment are the HVAC system’s supply air temperature, humidity, flow rate, and the light source used. Four evaluation indices, including the mean value and range between layers of temperature and moisture content, were used. The results revealed that supply air temperature and light source are significant for mean temperature, while supply air temperature and humidity are significant for mean moisture content. In the case of range between layers, supply air flow rate and light source display a significant correlation. These findings suggest that future optimization should prioritize the regulation of the HVAC system’s supply air and light source.

1. Introduction

1.1. Background

Global warming is a pervasive environmental issue that significantly threatens food security [1]. Affected by environmental problems, such as land desertification and water shortages [2], per capita cultivated land is projected to be less than one-third of the 1970 levels by 2050 [3]. Controlled environment agriculture (CEA) has emerged as a potential solution [4,5]. It comprises two main types: greenhouses and vertical farming [6]. Vertical farming, specifically plant factories, represents a viable approach [7] as it minimizes land requirements [8,9] and reduces water waste [10] while enabling year-round production [6]. Consequently, plant factories have gained significant popularity on the market.

Container farms (CFs) are a type of plant factory constructed by modifying standard-sized metal containers, such as 20 or 40-foot ones. Their unique origin and structure afford CFs advantages such as ease of transportation, rapid construction, and adaptability to various scenarios, setting them apart from traditional vertical farming methods [11]. In areas where conventional planting methods are not feasible, such as deserts, islands, and highlands, CFs can offer stable crop planting environments to farmers. In terms of commercial applications, CFs offer the prospect of high annual productivity and efficient space utilization [12,13].

1.2. Overviews and Limitations

Several studies have emphasized the significant influence of the thermal environment on the growth of various plant species, such as tomatoes [14,15], cucumbers [16,17], and lettuce [18,19]. As a result, controlling the temperature and humidity, which are key elements of the thermal environment, has become a primary concern in CFs. CFs are a specialized form of CEA, sharing many of the same challenges in terms of maintaining optimal conditions for plant growth. However, the uniqueness of CFs lies in their planting systems, including complex geometric structures of planting racks and artificial lighting required for photosynthesis, which poses challenges in achieving precise temperature and humidity control, as well as maintaining energy efficiency within the enclosed systems. Ventilation has emerged as a prominent method for controlling the thermal environment in current CEA practices [20]. Consequently, researchers have focused on optimizing internal airflow to improve the thermal environment [9,21]. For instance, researchers [21] have employed CFD simulations to analyze the impact of various ventilation duct configurations on airflow uniformity and the thermal environment within indoor vertical farms. Nevertheless, there exist notable differences between this expansive indoor vertical farm setting and CFs. Specifically, in a CFD simulation study focused on CFs, the researchers primarily scrutinized airflow uniformity, overlooking the consideration of the thermal environment [9]. Although some studies have explored the impact of ventilation on the thermal environment, comprehensive investigations on this specific issue are limited.

In terms of control facilities, some researchers examined a mechanical ventilation system to enhance temperature control in CEA during the summer season [22]. A new temperature controller and a novel multivariable control strategy were also proposed [23,24]. Although these researchers improved temperature regulation in CFs, they may not have thoroughly addressed the factors influencing the thermal environment. Regarding specific factors affecting the thermal environment, some researchers found a negative correlation between crop height and indoor temperature differences, observing that taller crops could lead to uneven relative humidity [25]. Vent locations were demonstrated to impact temperature distribution significantly [26]. While these studies have identified some location factors influencing the thermal environment, there is still a need to uncover many more, such as the parameters of air conditioning and light sources. Moreover, a more quantitative evaluation of these specific factors is necessary to understand their precise effects on temperature and humidity within CFs.

1.3. Aims and Methods

This study aims to comprehensively investigate and quantitatively evaluate the factors that impact the thermal environment of CFs. The objective is to establish a theoretical foundation for future research endeavors on thermal environment optimization and energy efficiency in CFs. Through theoretical analysis of the energy and mass conservation equations within CFs, specific factors that impact the thermal environment are extracted. Furthermore, quantitative prioritization and significance analysis of these specific factors will be performed using orthogonal experimental simulation. By identifying and analyzing these factors, researchers and CF users can gain a deeper understanding of the thermal environment, which, in turn, will facilitate the application and optimization of CFs.

In terms of research methodology, computational fluid dynamics (CFD) simulation has been widely employed as a powerful tool for evaluating the thermal environment in CEA studies [9,21,27,28,29,30,31,32]. This is supported by numerous studies focusing on crop heat and mass transfer modeling [2,33,34]. Consequently, this research adopts a CFD numerical simulation approach. A numerical model of the plant canopy within the CF is established (Section 2.1) and incorporated as source terms in the CFD simulation. Simulations are then conducted based on different cases (Section 2.2). Experimental data are used for model validation to ensure the accuracy of the simulations (Section 3.1).

2. Materials and Methods

2.1. Canopy Numerical Model

In order to simplify the calculations, accurate crop models were not developed in this study. Instead, some CFD simulation studies have utilized porous media to represent the plant canopy [21,30,31]. Within CFs, the energy exchange between plants and the environment occurs through sensible heat, latent heat, and radiation [2]. This can be described using Equation (1):

$$R_{\mathrm{LED}}=Q_{\mathrm{sen}}+Q_{\mathrm{lat}}$$

(1)

where R_LED can be calculated using Equation (2) [21]:

$$R_{\mathrm{LED}}={\alpha }_{\mathrm{LED}}\cdot P_{\mathrm{LED}}\cdot {\eta }_{\mathrm{LED}}\cdot CAC$$

(2)

where α_LED is 0.93 [33], P_LED is 12 W, η_LED is 52%, and CAC is 0.9 in this study [34].

For the canopy model, the difficulty lies in calculating sensible heat and latent heat (water evaporation). The sensible heat is driven by the temperature difference, as follows [21]:

$$Q_{\mathrm{sen}}=LAI\cdot {\rho }_{\mathrm{a}}\cdot {cp}_{\mathrm{a}}\cdot \left(T_{\mathrm{leaf}}-T_{\mathrm{a}}\right)\cdot {r_{\mathrm{a}}}^{-1}$$

(3)

According to research [2], LAI is taken as 2.96 m²/m² in this study. The aerodynamic resistance (r_a) is a parameter related to l and u, which is given by Equation (4) [34]:

$$r_{\mathrm{a}}=350\sqrt{l/u}\cdot {LAI}^{-1}$$

(4)

where l is taken as 0.1 m [21].

The water vapor pressure difference between the canopy and the surrounding air drives the latent heat exchange of the canopy. According to the research in vertical farms [34], the latent heat can be given by

$$Q_{\mathrm{lat}}={\lambda }_{\mathrm{w}}\cdot ET={\lambda }_{\mathrm{w}}\cdot LAI\cdot \frac{{\omega }_{\mathrm{leaf}}-{\omega }_{\mathrm{a}}}{r_{\mathrm{a}}+r_{\mathrm{s}}}$$

(5)

where ω_leaf and ω_a are given by Equations (6) and (7), respectively, and r_s is given by Equation (8).

$${\omega }_{\mathrm{leaf}}=\frac{P_{\mathrm{w}-\mathrm{leaf}}\cdot M_{\mathrm{w}}}{R_{\mathrm{mol}}\cdot T_{\mathrm{leaf}}}$$

(6)

$${\omega }_{\mathrm{a}}=R{H}_{\mathrm{a}}\frac{P_{\mathrm{w}-\mathrm{a}}\cdot M_{\mathrm{w}}}{R_{\mathrm{mol}}\cdot T_{\mathrm{a}}}$$

(7)

$$r_{\mathrm{s}}=60\frac{1500+PPFD}{200+PPFD}$$

(8)

P_w-leaf and P_w-a can be calculated from the Clausius–Clapeyron equation, see Equation (9); the PPFD of LED in this study is 250 μmol/m² s.

$$P_{\mathrm{w}}=611\exp \left[\frac{{\lambda }_{\mathrm{w}}\cdot M_{\mathrm{w}}}{R_{\mathrm{mol}}}\left(\frac{1}{273.15}-\frac{1}{T}\right)\right]$$

(9)

Equations (1)–(9) together form the canopy heat and mass transfer model, in which the unknown parameters include T_leaf, T_a, u, and RH_a. T_a, u, and RH_a can be obtained through simulation, and the only unknown parameter in the model is T_leaf. Then, the leaf temperature can be calculated from Equation (1). Write Equations (1)–(9) as user-defined functions (UDF) and load them into CFD.

2.2. Case Design

2.2.1. Variable Extraction

This study aims to examine the influencing variables of the thermal environment inside the CF and their degree of influence; therefore, the variables need to be extracted first. To facilitate the theoretical analysis, the following assumptions have been applied:

(1)

The airtightness of the CF under study is excellent, meaning that the penetration of external wind is negligible.

(2)

The auxiliary facilities in the CF, such as water pumps and fans, maintain stability and a constant calorific value.

(3)

The research focuses on the summer season, and the impact of the external environment changes at different moments is not considered.

The energy balance equation of the CF is as follows [2]:

$$Q_{\mathrm{canopy}}+Q_{\mathrm{LED}}+Q_{\mathrm{equipment}}+Q_{\mathrm{envelope}}+Q_{\mathrm{HVAC}}+Q_{\mathrm{infiltration}}=0$$

(10)

According to the assumptions, Q_infiltration can be considered 0 and Q_equipment constant. Q_canopy, Q_LED, Q_envelope, and Q_HVAC are given by Equations (11)–(14), respectively:

$$Q_{\mathrm{canopy}}=Q_{\mathrm{sen}}+Q_{\mathrm{lat}}$$

(11)

$$Q_{\mathrm{LED}}=N\cdot {\eta }_{\mathrm{LED}}\cdot P_{\mathrm{LED}}$$

(12)

$$Q_{\mathrm{envelope}}=K\cdot A_{\mathrm{envelope}}\cdot \left(T_{\mathrm{o}}-T_{\mathrm{a}}\right)$$

(13)

$$Q_{\mathrm{HVAC}}={\rho }_a\cdot c{p}_a\cdot q_{\mathrm{s}}\cdot \left(T_{\mathrm{s}}-T_{\mathrm{a}}\right)$$

(14)

According to Equation (10), the unknown variables that affect T_a are P_LED, q_s, T_s, T_leaf, u, and RH_a. T_leaf is a variable related to T_a (Section 2.1), and u is a variable related to q_s (the supply air mode is fixed), so there is the following equation:

$$T_{\mathrm{a}}=f\left(P_{\mathrm{LED}},q_{\mathrm{s}},T_{\mathrm{s}},R{H}_{\mathrm{a}}\right)$$

(15)

Similar to the energy equation, the humidity in the CF is analyzed through the following mass balance:

$$m_{\mathrm{canopy}}+m_{\mathrm{HVAC}}+m_{\mathrm{media}}=0$$

(16)

$$m_{\mathrm{canopy}}=V_{\mathrm{canopy}}\cdot ET\cdot LAD$$

(17)

$$m_{\mathrm{HVAC}}=q_{\mathrm{s}}\cdot \left({\omega }_{\mathrm{s}}-{\omega }_{\mathrm{a}}\right)$$

(18)

In Equation (16), m_media is 0 because the media where lettuce grows is covered [35]. Similarly, the function of humidity can be obtained:

$$R{H}_{\mathrm{a}}=f\left(T_{\mathrm{a}},q_{\mathrm{s}},T_{\mathrm{s}},R{H}_{\mathrm{s}}\right)$$

(19)

Equations (15) and (19) show that the CF’s thermal environment (T_a, RH_a) is influenced by P_LED, q_s, T_s, and RH_s. In other words, the light source and HVAC system supply air parameters affect the thermal environment inside the CF.

2.2.2. Factor Level

The orthogonal experimental design, particularly suitable for investigating multifactor and multilevel problems, offers a cost-effective alternative to traditional enumeration methods by leveraging orthogonality to select representative factors and levels from a comprehensive test [36]. It is an efficient and scientifically rigorous approach that helps elucidate the significance of various factors influencing test indicators. Many researchers have utilized the orthogonal experimental method to effectively reduce the number of simulations [36,37,38]. In line with this, we conducted simulations using the orthogonal experimental method. Four factors were extracted in Section 2.2.1 and considered in the orthogonal experimental design.

Table 1. Factors and levels.

Factor		Level
		1	2	3	4
A	Temperature of supply air, Ts (°C)	16	18	20	22
B	Flow rate of supply air, qs (m3/h)	2000	2250	2500	2750
C	Relative humidity of supply air, RHs (%)	50	60	70	80
D	LED duty ratio, (-)	0.4	0.6	0.8	1

shows each influencing factor and level value, and

Table 2. Orthogonal design table.

No	Factor
	A (°C)	B (m3/h)	C (%)	D (-)	Error
1	16	2000	50	0.4	1
2	16	2250	60	0.6	2
3	16	2500	70	0.8	3
4	16	2750	80	1	4
5	18	2000	60	0.8	4
6	18	2250	50	1	3
7	18	2500	80	0.4	2
8	18	2750	70	0.6	1
9	20	2000	70	1	2
10	20	2250	80	0.8	1
11	20	2500	50	0.6	4
12	20	2750	60	0.4	3
13	22	2000	80	0.6	3
14	22	2250	70	0.4	4
15	22	2500	60	1	1
16	22	2750	50	0.8	2

is the corresponding orthogonal design table. The four levels in Table 1 were selected based on the air conditioning supply parameters adopted during the experiment presented in Section 2.5.

In Table 1, the LED duty cycle is a current light source modification solution. Several scholars have discovered that altering the duty cycle of the LED to create pulsed light rather than continuous light (duty cycle of 1) [39,40,41] does not adversely affect crop growth. By appropriately setting a reasonable duty cycle and frequency, the photosynthetic efficiency can approach that of continuous light. The duty cycle determines the time the LED remains powered on within a given unit time. When applied to CFD simulation, this can be regarded as a multiplication of P_LED by the duty cycle, thus modifying the parameters of the P_LED.

2.3. Physical Models and Grids

The physical model employed in this study is based on an actual CF (located in Shanghai, China). As illustrated in Figure 1, the CF consists of five planting layers on both the left and right sides, with each layer equipped with 40 LED light strips (SANANBIO, Quanzhou, China). The physical model is created using the DesignModeler component in Workbench 2020 R2. Fluent Meshing (Ansys, Canonsburg, PA, USA) is employed for grid division. Additionally, certain boundary conditions are encrypted, and boundary layers are generated to ensure accurate simulations.

To ensure accuracy, five sets of grids were established with different levels of precision. Figure 2 presents the independence analysis of these grids. The average relative error between the standard (2.14 million) and the precision grid (3.42 million) is 0.46%, while between the fine grid (2.95 million) and the precision grid (3.42 million), it is 0.68%. However, both the coarse grid (1.10 million) and general grid (1.51 million) exhibit average relative errors above 1.5%. The max relative errors are 15.63%, 6.21%, 3.19%, and 4.46% for the coarse, general, standard, and fine grids, respectively. Based on the results of the independence analysis, the standard grid (2.14 million) was selected for simulations.

2.4. Solution Setting

The simulation is conducted with Fluent 2020 R2. The primary investigation is the canopy plane’s thermal environment (temperature, humidity) in the CF. The simulation is conducted under the steady state [30].

2.4.1. Solution Equations

The energy, turbulent viscosity, radiation, and species models were turned on. The equations involved, which include continuity, momentum, energy, and species, are as follows:

$$\frac{\partial }{\partial x_{\mathrm{i}}}\left({\rho }_{\mathrm{i}}{u}_{\mathrm{i}}\right)=0$$

(20)

$$\frac{\partial }{\partial x_{\mathrm{i}}}\left({{\rho }_{\mathrm{i}}u}_{\mathrm{i}}{u}_{\mathrm{j}}\right)=-\frac{\partial }{\partial x_{\mathrm{i}}}P{\delta }_{\mathrm{ij}}+\frac{\partial }{\partial x_{\mathrm{i}}}{\mu }_{\mathrm{t}}\left(\frac{\partial u_{\mathrm{i}}}{\partial x_{\mathrm{j}}}+\frac{\partial u_{\mathrm{j}}}{\partial x_{\mathrm{i}}}\right)+{\rho }_{\mathrm{i}}{g}_{\mathrm{i}}+S_{\mathrm{m}}$$

(21)

$$cp\frac{\partial }{\partial x_{\mathrm{i}}}\left({\rho }_{\mathrm{i}}{T}_{u\mathrm{i}}\right)=\frac{\partial }{\partial x_{\mathrm{i}}}\left({\lambda }_{\mathrm{f}}+\frac{{\mu }_{\mathrm{t}}cp}{{Pr}_{\mathrm{t}}}\right)\frac{\partial \mathrm{T}}{\partial x_{\mathrm{i}}}+S_{\mathrm{s}}$$

(22)

$$\frac{\partial }{\partial x_{\mathrm{i}}}\left({\rho }_{\mathrm{i}}{C}_{u\mathrm{i}}\right)=\frac{\partial {\phi }_{\mathrm{ij}}}{\partial x_{\mathrm{i}}}+S_{\mathrm{w}}$$

(23)

where i and j represent the three directions of the three-dimensional Cartesian coordinate system; ρ_i adopts the Boussinesq assumption in the simulation; φ_ij results from temperature and concentration gradients. There are three source terms in the equation. S_m represents the momentum source term generated due to porous media, S_s is the heat source term generated by sensible heat exchange of lettuce, and S_w is the species source term of lettuce water vapor, as follows:

$$S_{\mathrm{m}}=-\left(\frac{\mu }{\alpha }{u}_{\mathrm{i}}+0.5C_2\cdot \rho \left|u\right|u_{\mathrm{i}}\right)$$

(24)

$$S_{\mathrm{s}}=Q_{\mathrm{sen}}\cdot LAD$$

(25)

$$S_{\mathrm{w}}=ET\cdot LAD$$

(26)

In Equations (25) and (26), LAD is considered 2.467 [2].

2.4.2. Model Selection and Solver Settings

The Realizable k-ε model has been utilized in the research of turbulence flow in greenhouse and vertical farms [27,28,29,32]. Compared to the standard k-ε model, it has demonstrated improved accuracy in predicting jet diffusion [32], especially in the context of the high-speed fan utilized in our simulations. This improved accuracy is particularly advantageous for capturing the dynamics of the jet generated by the high-speed fan, ensuring more reliable predictions and insights into the flow behavior. For the radiation model, the discrete ordinates (DO) model was chosen due to its suitability for complex structural scenes. DO models have been successfully employed in various agricultural environment simulations [21,30].

The Pressure-Based algorithm was used in the simulation, with the SIMPLE algorithm employed to solve the pressure–velocity coupling due to the presence of porous media and porous–jump boundary conditions. Except for the pressure term, which utilized PRESTO! [21], second-order upwind discretization schemes were applied for the remaining terms. To save computational costs, adjustments were made to the residuals used as convergence criteria. Turbulence residuals were set to less than 1 × 10⁻³, while continuity and momentum residuals were set to less than 1 × 10⁻⁴. The DO radiation residual was set to less than 1 × 10⁻⁵, and both energy and species residuals were set to less than 1 × 10⁻⁶.

2.4.3. Boundary Conditions Settings

The settings of all boundary conditions (BCs) of the model are shown in

Table 3. Settings of boundary conditions.

BCs	Type	Setting
Inlet	Velocity–inlet	From case
Bag Pipe	Porous–jump	Face Permeability: 2.81 × 10−12 Pressure–Jump Coefficient: 8.58 × 106
Fan	Fan	Pressure Jump: polynomial velocity: from 8.27 m/s to 10.63 m/s
Frame	Wall–Heat Flux	None
LED	Wall–Heat Flux	From case
Plant	Fluid–Porous Zone and Source Terms	UDF
Outlet	Pressure–outlet	/
Wall (except Floor)	Wall–Convection	Heat Transfer Coefficient: 0.7 Free Stream Temperature: 296.01 (Chinese Standard Weather Data) Wall Thickness: 0.05
Wall-Floor	Wall–Heat Flux	None

.

2.5. Experiment Setup

In December 2022, a planting trial was conducted to validate the accuracy of the model. In all layers except the fifth planting layer, 100 heads of Italian lettuce (Lactuca sativa L. var. ramosa Hort) were planted on both frames, resulting in a total of 800 heads. Figure 3 illustrates the positioning of nineteen measurement locations within each layer, which were used to collect temperature and humidity data near the crop canopy (0.05 m away from the plants). Within the planting area on the first to fourth planting layers, four measurement points were placed, with a spacing of 0.5 m from the front and rear of the planting frame. On the fifth planting layer, two measuring points were situated at the center of the planting frame. The last one was at the center of the outlet. The Temperature and Humidity Data Logger (Testo 175H1, Testo, Titisee-Neustadt, Germany) was used at these 19 measuring points. To collect input parameters for the simulation, a hot wire anemometer (VelociCalc 9565, TSI, Shoreview, MN, USA) was employed to measure the supply air velocity at the inlet. The Temperature and Humidity Data Logger (COS03, Renke, Jinan, China) recorded the temperature and humidity of the supply and outdoor air. A detailed description of all instruments can be found in Appendix A. The Temperature and Humidity Data Logger was placed at each measuring point for seven days, while only data from the light period were selected. Simultaneously, the hot wire anemometer was conducted during the same time frame. Since the supply air flow rate remained stable throughout the experiment, the average value was used.

2.6. Evaluation Index

The air around the canopy of each layer is divided to present the results and discuss them more intuitively. Taking the first planting layer (L1) as an example, as shown in Figure 4, the plane is divided into 200 areas. Each area’s height is 0.05 m above the crop.

To assess the relative influence of different factors on the thermal environment around the canopy, the arithmetic mean of temperature and moisture content (MT, MMC) for each layer was used as an indicator for data analysis. Furthermore, to evaluate the variations between layers, the range between layers (RBL) was calculated as the range of mean differences. These two indicators, MT and MMC, provide insights into the overall temperature and moisture levels within each layer, respectively, while the RBL takes into consideration the variability between layers.

3. Results and Discussion

3.1. Validation

The comparison between the test and simulated data for temperature and humidity is presented in Figure 5. The relative error between simulation and experimental data is commonly used as a metric to evaluate the acceptability of simulation results and is typically set at around 5%, taking into account factors such as instrument inaccuracies, measurement uncertainties, model simplifications, and parameter variations. In Figure 5a, most of the simulated values fall within the box region of the test values, indicating a good agreement. The majority of the relative errors between the simulation and the test are less than 5%. The average error for temperature is 0.22 °C, and the average relative error is 2.59%, both falling within the measurement uncertainty of the instrument.

In Figure 5b, humidity is expressed as moisture content. Only a few individual values deviate slightly from the uncertainty range, and all the differences in humidity do not exceed 1 g/kg. The average error for humidity is 0.39 g/kg, which, when converted to relative humidity, is equivalent to 2.88%, close to the instrument’s uncertainty. The average relative error is 5.02%. It is important to note that considering all lettuces in the simulation are at the same growth stage (with the same LAI) compared to the actual scenario (with different LAI), these deviations are acceptable. The verification work demonstrates that the established model effectively represents the temperature and humidity distribution within the CF.

However, it is necessary to pay attention to the excessive deviations observed at specific measuring points. Measurement location 11 exhibits the largest temperature discrepancy, with a relative error of 6.43%. Measuring points 6 and 16 display significant humidity gaps (0.989 g/kg and 0.842 g/kg, respectively). For temperature, small deviations in the positioning of measuring points, especially those influenced by high-speed fans (as in the case of measuring point 6), can result in considerable temperature errors [21]. Neglecting geometric details, such as hydroponic system accessories near measuring locations 6 and 16, may contribute to humidity errors.

3.2. Thermal Environment of Canopy Plane

3.2.1. Temperature Distribution

In Section 2.6, the canopy plane is divided into 200 areas, and the mean temperature (MT) and standard deviation (SD) for each layer obtained from the simulation in case 4 are calculated and presented in Figure 6. The results for all cases can be found in Figures S1–S16 in the Supplementary Material.

Figure 6 reveals that the majority of high SD areas are located in Layer 1 and Layer 2 predominantly concentrated on the left side of the canopy plane, indicating a significant temperature gradient in a specific portion of the left canopy area. Across all layers, the SD in the edge area exhibits a substantial increase, which can be attributed to the difference between the temperatures in the non-planted area and the canopy itself. When comparing different layers, the SD gradually decreases with the increasing height of the layer. This can be attributed to the high-speed airflow generated by the fan and the air supply duct, which ensures effective air circulation in the upper layers, enabling the timely discharge of hot air and promoting a uniform temperature distribution.

In most areas, MT on the left side is higher than on the right due to airflow tendencies. Nevertheless, there are instances when the MT is lower on the left side. Figure 6 indicates that in Layer 3, the MT is generally higher on the right side than on the left. The velocity vector diagram of the L3 canopy plane depicted in Figure 7 clarifies this observation, revealing the presence of four vortex areas. Apart from the two prominent vortex areas, a smaller one can be observed on the right side of the canopy (top of Figure 7), along with a tiny vortex on the left side. The air surrounding these vortex areas exhibits low velocity and sparse vector distribution. The existence of these two vortexes corresponds with the high MT area identified in Figure 6.

3.2.2. Moisture Distribution

The MMC and SD of each area in case 4 are calculated and shown in Figure 8. The results of all cases can be viewed in Figures S17–S32 in the Supplementary Material.

As seen in Figure 8, the areas of high SD of humidity are larger than the temperature, revealing that the vortex zone has a more pronounced impact on moisture. No direct relationship between high MMC and high SD is observed. Examining both Figure 6 and Figure 8, it can be noted that the MMC follows a similar trend as the MT.

3.3. Factor Analysis

Figure 9 displays each case’s average value and RBL using the evaluation index in Section 2.6. It is evident that across all 16 cases, there are minimal differences in temperature and humidity from the first to the fourth planting layers, while a significant decrease is observed in the fifth planting layer. The layer at which the maximum value occurs varies across cases, whereas the minimum value consistently emerges in the fifth planting layer. Case 15 exhibits the highest MT in Layer 4, with a value of 28.70 °C, while case 13 has the highest MMC in Layer 1, with a value of 14.388 g/kg. The smallest MT and MMC values are observed in case 1, with 18.03 °C and 6.658 g/kg, respectively, in Layer 5. Regarding RBL, case 9 demonstrates the maximum TRBL at 2.38 °C, and it also has the highest MCRBL at 0.712 g/kg. On the other hand, case 12 exhibits the minimum TRBL at 0.47 °C, while case 8 has the minimum MCRBL at 0.140 g/kg.

3.3.1. Range Analysis

Orthogonal experiments offer the advantage of providing a comprehensive and comparable understanding of the influence of changing factors within an experiment. This understanding is typically obtained by analyzing the range (R) of the average values for each level of the factors [36,37,42]. The size of R is positively associated with the degree of influence exerted by the factors.

Table 4. Range analysis table.

Evaluation Index	RA *	RB	RC	RD	Order of Importance
MT of L1	5.749	1.319	0.671	4.211	A > D > B > C
MT of L2	5.723	1.315	0.703	4.065	A > D > B > C
MT of L3	5.889	1.562	0.723	3.993	A > D > B > C
MT of L4	5.827	1.501	0.714	4.032	A > D > B > C
MT of L5	5.639	0.561	0.610	3.183	A > D > B > C
TRBL	0.119	0.758	0.236	1.019	D > B > C > A
MMC of L1	3.795	1.015	3.668	0.721	A > C > B > D
MMC of L2	3.800	0.960	3.663	0.695	A > C > B > D
MMC of L3	3.808	1.040	3.636	0.739	A > C > B > D
MMC of L4	3.779	0.984	3.679	0.712	A > C > B > D
MMC of L5	3.671	0.717	3.789	0.518	A > C > B > D
MCRBL	0.119	0.290	0.142	0.289	B > D > C > A

* R_i is the range value of factor i.

lists the range analysis table of 12 indicators. The specific calculation of each indicator can be found in Table S1 in the Supplementary Material.

According to Table 4, the factors influencing MT and MMC in order of significance are as follows: for MT, it is A > D > B > C, while for MMC, it is A > C > B > D. Hence, changes in supply air temperature have the most substantial impact on both MT and MMC. For MT, the LED duty cycle is second, while the flow rate and humidity of the supply air have little impact. In the case of MMC, the humidity is second, whereas the others have minimal effects.

The order of factors influencing TRBL is D > B > C > A, whereas for MCRBL, it is B > D > C > A. For TRBL, the LED duty cycle has the highest impact, followed by the supply air flow rate. Conversely, for MCRBL, the flow rate is the highest. Both temperature and humidity have little effect on the two RBLs.

Figure 10 provides a visual representation depicting how each factor influences MT, MMC, TRBL, and MCRBL. The consistency of the range rules of MT and MMC across each layer allows the use of L1 data as an illustrative example.

When factor A increases, both MT and MMC exhibit significant increases, TRBL shows a relatively mild change, and MCRBL experiences a slight increase. When factor B increases, all indices decrease, but the RBLs decrease significantly. When factor C increases, MMC exhibits a significant increase, while MCRBL shows a downward trend. When factor D increases, apart from MMC, the other three indicators exhibit slight changes initially, followed by a rapid increase.

3.3.2. Analysis of Variance

An analysis of variance (ANOVA) can be used to determine whether each factor has a significant impact on the results [36,37,42]. The significance level is typically selected at 0.01 and 0.05, and it is determined by the p-value derived from the F-distribution. When p < 0.01, it indicates a highly significant impact of the factor on the results, denoted by “**”. When 0.01 ≤ p < 0.05, it indicates a significant effect, represented by “*”. When p ≥ 0.05, the factor is considered to have no significant impact. The ANOVA results of the orthogonal test are presented in

Table 5. ANOVA.

Index	Factors	SS *	Df *	MS *	F	p	Significant
MT of L1	A	73.839	3	24.613	181.805	0.001	**
	B	3.728	3	1.243	9.179	0.051
	C	1.192	3	0.397	2.934	0.200
	D	44.225	3	14.742	108.890	0.001	**
	error	0.406	3	0.135
MMC of L1	A	32.210	3	10.737	129.106	0.001	**
	B	2.328	3	0.776	9.332	0.050
	C	30.105	3	10.035	120.670	0.001	**
	D	1.432	3	0.477	5.739	0.093
	error	0.249	3	0.083
TRBL	A	0.028	3	0.009	0.493	0.712
	B	1.202	3	0.401	20.868	0.016	*
	C	0.114	3	0.038	1.987	0.294
	D	2.664	3	0.888	46.232	0.005	**
	error	0.058	3	0.019
MCRBL	A	0.030	3	0.010	5.772	0.092
	B	0.176	3	0.059	34.346	0.008	**
	C	0.046	3	0.015	8.967	0.052
	D	0.205	3	0.068	39.959	0.006	**
	error	0.005	3	0.002

* SS is sum of squares deviations from the mean, Df is degrees of freedom, MS is mean square. “**” indicates a highly significant impact while “*” indicates a significant effect.

, and the corresponding data can be found in Table S1 in the Supplementary Material.

As shown in Table 5, considering L1 data as an example (since the results of MT and MMC in each layer are consistent), factors A and D have a highly significant impact on MT, while factors A and C have a significant effect on MMC. For both the two RBLs, factors B and D are significant, whereas factors A and C are not. Moreover, the ranking of the F-values for each indicator is consistent with the R-value in Table 4.

3.4. Limitations

In our CFD simulation, we utilized a steady-state approach. However, it is important to acknowledge that there might exist discrepancies between the simulation and the actual situation due to the dynamic changes occurring within the CF. However, considering the applicability of steady-state simulation for this study, this gap may not be significant. Nonetheless, in future research, transient-state simulations can be attempted for validation. Another aspect to consider is that the canopy model employed in our study focused on specific crop species. It is conceivable that various crop species may exhibit distinctions in the computation of the canopy model. Hence, further discussion, such as controlled variable experiments in future research, is needed to determine whether the model applies to other crops.

Lastly, it is worth noting that the canopy model used in our research is static. However, plant growth is a dynamic process that is influenced by date [2]. The dynamic changes in plant growth involve LAI and LAD values in the canopy model, which can affect the characterization of heat transfer in the canopy model and further affect the calculation conclusions of T_leaf. However, the experimental results in reference [2] indicate that there are no significant differences between different stages. Nonetheless, in order to achieve a higher level of accuracy, future studies should consider proposing a more precise model that accounts for this dynamic nature, such as changing the constant LAI and LAD values to variable ones. Meanwhile, the interaction between factors can also serve as the next research object.

4. Conclusions

This research primarily focused on the thermal environment within CFs. Through theoretical analysis, we identified the key factors that affect canopy temperature and humidity. Subsequently, a physical model and grid for the CF were established and validated using experimental data. Our study employed CFD numerical simulation and conducted an orthogonal experiment to investigate the order and significance of these factors. The main conclusions are as follows:

(1)

The distribution patterns of temperature and humidity within the CF are generally consistent. Heat and moisture tend to accumulate in the vicinity of the outlet due to airflow tendencies. The average values in the first to fourth planting layer exhibit similarity, with a noticeable decrease in the fifth planting layer. The vortex, which has a more significant role in humidity, influences the distribution of the thermal environment.

(2)

Range analysis indicates that the order of factors influencing MT is A > D > B > C, and for MMC, it is A > C > B > D. Regarding TRBL, the order is D > B > C > A, while for MCRBL, it is B > D > C > A. Increasing the supply air temperature leads to a rapid rise in MT and MMC, with slower changes in RBLs. Higher supply air flow rates cause a decrease in all four indicators, with RBLs showing a more pronounced effect. An increase in supply air humidity contributes to a significant rise in MMC, but both RBLs exhibit a downward trend. The LED duty cycle initially yields a gentle effect on the indicators, followed by a rapid increase. However, MCRBL experiences a slight decrease initially.

(3)

ANOVA reveals that the significant factors influencing MT are supply air temperature and LED duty cycle, while supply air temperature and humidity are critical for MMC. This suggests that future efforts in optimizing the thermal environment around CFs should prioritize adjusting the supply air temperature, humidity level, and light source. For both RBLs, supply air flow rate and LED duty cycle demonstrate a significant correlation, emphasizing the importance of these two factors in achieving a more uniform thermal environment.