Four independent and identical systems residing in four Percival (Perry, IA) growth chambers were used to facilitate the air temperature and relative humidity monitoring and control.

To manage water delivery in detail, a complex water management system is designed and optimized to fit chamber and shelf setup, as illustrated in Figure 2A . Each incubation chamber contains two units, presented in Figure 2A . The unit contains a three-layer shelf, a 120-L plastic container for water storage, a water upstream pipeline, six incubating pots, and a water downstream pipeline. Water upstream and downstream pipelines and incubating pots are mainly constructed with PVC parts.

The incubating pot was designed and optimized for hydroponically growing Romaine and Oak Leaf lettuce with flowing aqueous nutrient solution, considering easy assembly, water flowability, compatibility with weight balance measurement system, reliability, and flexibility, as illustrated in Figure 2B .

For accessibility of the incubating pots, all PVC was purchased directly from local plumbing vendors. Three-quarter-inch pipes were cut down to the designed length for assembly, whereas other parts were used directly as purchased. The three supporting legs can be rotated to fit different sizes of weighing scale trays. A water exit was designed at the bottom of the 6-inch cap to eliminate dead water. Leg tops were adhered to and fastened onto the cap with PVC cement, stainless bolts, and nuts. Such a design took thermal geometry variation, high humidity, and high salty conditions into consideration.

System air conditions were primarily regulated by a Percival growth chamber mentioned previously via prior defined programs and an on-board interface, as shown in Figure 2C . Programs could define desired temperature and relative humidity with customized time within a 24-h cycle. The chamber operation conditions were logged and could be downloaded from an on-chip micro-SD card. A separate air condition data logger, GZAIR CO₂ Data Logger Monitor (Engerwitzdorf, Austria), was used to log air temperature, humidity, and CO₂ concentration with an acquisition frequency of once per 5 min. This data logger was connected to an on-site PC via a USB cable.

Water conditioning included two parts: scheduling and nutrient conditions. Kasa WiFi-controlled switches (Shenzhen, China) were used to control the Smartpond 80-gallon-per-hour submersible fountain pump purchased from Lowes.com to customize water scheduling. The Flora Nova Grow 7-4-10 one-part nutrient solution, manufactured by General Hydroponics (Berkeley, CA), was diluted 1,600 and 800 times with tap water as two nutrient solutions during the seedling and early growing stages. Complete composition of this formula can be found at https://generalhydroponics.com/resources/floranova-grow-product-label/.

Watering events alter the measured weight significantly. By analyzing measured weight data, watering events were clearly logged and carefully evaluated. Water conditions, including water temperature, pH, and electric conductivity, were monitored by Bluelab (New Zealand) Wi-Fi-enabled guardian monitor.

Lighting was provided by WiFi-enabled K5 series XL750 KindLED (Santa Rosa, CA) grow light. This lighting system only enabled individual control of white, red, and blue lights, although other spectrums of light could also affect plant growth. Photosynthetic photon flux density (PPFD), also known as light intensity, could be modified as variations of power percentage, ranging from 0% to 100%. Lighting conditions could be tuned with minute resolution in the time domain.

Since PPFD was controlled digitally (not analog signals), the delivered PPFD measurement was carried out as photosynthetic active radiation (PAR) meter calibration at the incubating pot level. Each light channel was calibrated individually. Light data were based on calibrated PPFD readings and populated with a frequency of minutes on the time domain.

The weight of each growing lettuce was continuously recorded using a weight measurement system, which consisted of Ruishan (Shanghai, China) weighing balances and Serial RS232-to-Ethernet converters. The capacity and accuracy of the weighing balance were 2,000 g and 0.01 g, respectively. Since the data port of the weight balance was the RS232 serial port, model number USR-TCP232-302 converters, manufactured by PUSR (Jinan, China), were used to convert serial signals to transmission control protocol/internet protocol (TCP/IP) signals.

Four different interfaces were used to communicate sensing and controlling hardware and computation hardware to ensure the stability and reliability of the data communications. With Wi-Fi connections, managerial commands and system status, such as feedback from Kind LED lights and Kasa switches, could communicate back and forth over the Wi-Fi interface. In most cases, chambers were maintained at high humidity levels. A wireless communication approach improved system safety and data communication reliability. All Bluelab monitors were required to upload data via a proprietary Bluelab data link, and a Bluelab connect stick was used to receive wireless data from Bluelab monitors and send it to the on-site PC for data logging via the USB interface.

Besides water condition data, air condition data were also transmitted via the USB interface. Water data were logged with a frequency of 1 min, and air condition data were logged every 5 min. TCP/IP protocol was utilized to transmit high-frequency data such as weight measurement data communication and remote data exchange. T-568B terminated category 6 cables were used to ensure hardware reliability and transmitting speed.

One on-site computer (PC) was set up per growth chamber to serve as a secure Wi-Fi broadcaster, commanding system, and data acquisition system. Beyond three environmental condition subsystems and a weight measurement system, 16 web cameras are implemented for operational status checks, condition visual verification, visual inspections, and remote system failure diagnoses. With the large numbers of devices connected via Wi-Fi, a secure and independent broadcast is required to ensure command executing security and reliability. Since the Kind LED light on-board chip uses a web page-based control interface, this on-site PC is perfect for providing remote access to the lighting system via internet connection. The water condition system, air conditioning system, and weight management system all use this PC as a temporary data storage and hub for remote communication.

Beyond this on-site PC, remote computational interventions were needed to provide cloud computing and human intervention access portals. Crop modeling could be hosted on this remote computation equipment, providing model training and updating with in situ data input, operational command suggesting and advising, bolting predictions, and breeding simulations. It also provided remote human intervention access portals, where operation managers could access data and modify operation parameters.

Three daily programs were executed through the seedling and cloning stages to differentiate the conditions, as shown in Table 2 , where programs #1 and #2 had a 24-h cycle and program #3 had a 12-h cycle. Since air circulation in the growth chamber was regulated via a fan with inconstant working schedules, no CO₂ generation source was introduced in the system. System air condition data, including air temperature, humidity, and CO₂ concentration, were logged every 5 min per sensor limitation.

For every six plants, 100 L of diluted nutrient solutions were formulated and stored in the 120-L plastic container for hydroponic circulation. To compensate for evaporation, additional tap water was added to maintain the 100-L total volume. In the seedling stage, the flora nutrient was diluted 1,600 times.

In the cloning stage, the flora nutrient was diluted 800 times. At the start, the water’s reading was 450 ppm (in the ppm 500 scale) for electric conductivity, 19°C for water temperature, and 6.8 for pH. Water conditions were logged with 1 min resolution on time domain.

The LED light was positioned 1 m above the incubating pot, which is shown in Figure 2B . The lighting conditions were programmed with power percentage and changing time. In the 24-day project duration, the light programs are prescribed in Table 3 .

For PPFD calibration, all three spectrum lights were calibrated individually. The PAR sensor remained at the same location during calibration and only lighting conditions were changed. The calibration results showed a highly linear correlation with power percentages, as shown in Figure 3 .

Based on this calibrated PPFD reading, the light condition profiles were populated with minute time resolution, noted as PPFD _s (t), where s denotes the standard profile. When determining the PPFD for each plant, secondary calibrations were conducted. The PAR sensor was located at each plant’s incubating pot height level. During the calibration, the Kind LED light was set as 60% in red, white, and blue light. The calibrated PPFD for this setting was 494 µmol/m²/s. For each plant, a new PPFD reading was logged as PPFD X X . c with XX denoting the plant number. Therefore, the light condition profiles for each plant, PPFD X X ( t ) , can be calculated as

Weight data included weight reading and the corresponding time stamp, as shown in Figure 4A . The periodic mass fluctuations were due to the watering event, water evapotranspiration, and root water uptake. To understand the complex effects of multiple processes that could affect the weight data, we make several definitions based on the observed data.

Consider Plant 38 as an example. The mass data of this plant in November 2022 are plotted in Figure 4A . Figure 4B shows the mass data over a typical watering cycle, where we define eight critical time points:

We define the following physiological indices to characterize plants’ physiological properties and operational effectiveness during the watering cycle:

Plant 38’s transpiration rates before and after watering events are shown in Figure 5A . The negative data indicate mass loss. Minor positive data points may be caused by balance kickback. Before November 15 and after November 22, pre-watering transpirations were slower (less negative). After the water outage on November 14, the pre-water transpiration was faster (more negative) until November 22. Both pre- and post-transpiration were changing periodically with a cycle of 24 h. Most of the data points ranged from 0 to –2 mg/s.

Watering saturation speeds R s t . A and R s t . B are shown in Figure 5B . In general, R s t . B is greater than R s t . A . The R s t . B had a range from 0.01 to 0.04 g/s, the R s t . A ranged from 0 to 0.05 g/s. The periodic (daily) fluctuations reduced as time progressed for both R s t . B and R s t . A .

The water absorption rate R w s and R w e are shown in Figure 5C . Before the November 15 water suspension, R w s decreased, and R w e increased (less negative) over time. After November 18, both R w s and R w e stabilized. They also fluctuated periodically with a 24-h cycle throughout the entire process. The watering absorption speed R w s reflects the plant’s water uptake capability. This capability is determined by xylem osmotic potential as Earles discovered (Earles et al., 2015). Therefore, the R w s can be used as a xylem index to evaluate plant physiology. Water exiting speed, R w e , on the other hand indicates the water exit rate. During the growth, plants’ roots were blocking the water exit. The slight increase (less negative, 0.008 g/s/day with R ² = 0.75) can be used as an index evaluating root growth conditions.

Environmental factors, such as lighting and air conditions, significantly impact plants’ mass change rate, including R t r . B , R t r . A , R s t . B , R s t . A , and R w s . As shown in Figure 5 , most of the mass changing speed fluctuates daily. We suspect that the mass change speeds are correlated with air or lighting conditions since air conditions and light conditions are changed with a 24-h cycle as mentioned in Sections 3.1 and 3.3. On the other hand, the R t r . B and R t r . A behaved differently before the water outage and during the recovery period (November 18 to 21). With uninterrupted watering, plants may utilize sufficient water for respiration to synthesize cellulose and for photosynthesis to accumulate carbon from air. When water availability is limited, plants may modify their photosynthesis and respiration activities to adapt to the water stress physiologically and biochemically (Osakabe et al., 2014).

The watering saturation mass change speed reflects the system watering efficiency. The daily fluctuations indicate that the R s t . B is plant physiology related. It signifies the approaching of efficient watering of the plant. Such evaluation indices can be used as an operational trigger to regulate watering time span to optimize the watering event efficiency. The R s t . A had less daily fluctuations, compared to R s t . B . With universal watering length, the approaching zero of R s t . A signifies that watering reached the plant’s dynamic maximum water holding capacity. Since a considerable amount of water will exit the plant after the watering event ends, modifying the watering length to increase R s t . A will improve the watering utilization.

The duration of rinsing and un-rinsing of Plant 38 is shown in Figure 6A . Abnormal data off the chart were due to watering activity suspension. Regardless of the abnormal data points, the average D r was 104 s, and the average D u r was 72 s. Both D r and D u r fluctuated daily.

The duration of watering saturation is shown in Figure 6B . Abnormal data were removed due to watering suspension. Before November 15, the daily fluctuation was larger than after November 18. After November 24, the moving average of the saturation duration decreased significantly.

The duration of rinsing, D r , in Figure 6A signifies the time period of no water was wasted after the entire watering event. Since the power of the watering pump was constant, the rising duration indicates the amount of water stress the plant was under at the start point of the watering event. Reduced fluctuations along the growing process indicate that the plant adapted to this watering fashion (amount and frequency). The duration of unrinsing, D u r , in Figure 6A , indicates the period of water exiting after the watering event ends. During this period, part of the dynamically held water exited from the incubating pot. Suppose the hydroponic production was carried out because the plant may need to leave the water tank after watering. In that case, the D u r indicates the time needed to water until all the unused nutrient liquid exited the plants.

The duration of saturation, D s t , in Figure 6B , signifies the period when the plant did not absorb a portion of the pumped water during the watering event. The larger fluctuations before November 15 indicate that daily changed environmental factors significantly impacted the water uptake. The reduced fluctuation after the water outage suggested that the water had already adapted to the current watering fashion, as D r indicates. It may also suggest that the plant was less susceptible to daily cycled environmental parameters. The moving average decrease after November 27 suggests that plants need more time to begin saturation, which indicates the improvement in watering efficiency.

Mass data at the beginning of the watering event are plotted in Figure 7A . This is the lowest mass reading within the watering cycle, which includes the mass of the plant, incubation pot, and other accessories. Figure 7A shows that the watering suspension lasts 24 h (four watering events). The plant used three watering events to recover. The abnormal data point was on November 18 due to a minor watering suspension. The abnormal data point on November 22 was due to noises in the raw data. Regardless of the abnormal data, the linear regression yields 5.49 g/day growth with R ² = 0.95.

A plant’s water-holding capacity is one of the important plant indices to evaluate plant growth. Plant 38’s dynamic and static water holding capacity are shown in Figure 7B . Both the dynamic and static water holding capacity fluctuate daily.

Root water-holding capacity reflects the amount of roots and morphology. Plant 38’s root dynamic water holding capacity data are shown in Figure 7C . The daily fluctuations before November 15 were significantly greater than those after November 18. The dynamic water loss (negative mass change) represents the amount of water that exits after the watering event stops. The dynamic water loss slightly decreases (less negative) at 0.24 g/day with R ² = 0.78 with abnormal data exclusion.

One of the most important functions of this hydroponic system was to monitor plant growth by evaluating plant weight. As shown in Figure 7A , Plant 38’s mass increased linearly as time progressed. Water outages on November 15 and 18 reduced the growth speed. The water outage and recovery also demonstrated the plant’s tolerance, resistance, and response to water stress. The reduced daily fluctuation after the water outage and recovery indicated that the plant was growing conservatively in response to the environmental condition change. Such insights may be useful for commercial hydroponic growers to evaluate emergency water outages and calculate production loss.

As one of the important physiology indices, water holding capacity, shown in Figure 7B , signifies the water availability for plants between watering events. The static water holding capacity, M s t . h , evaluates the net water retention after water exits. It is the true water amount that plants can use. The dynamic water holding capacity, M d y . h , evaluates the water holding capacity with water running. Beyond static water holding, dynamic water holding also includes the water retention from surface rinsing due to surface tension. There were many factors that can change the surface tension and water retention capacity between the xylem cell wall and watering nutrition liquid, such as temperature, salt concentration (here as electric conductivity), water running speed, xylem cell wall surface roughness, and, most importantly, xylem surface area (Kumar et al., 2022). The daily fluctuations for both M s t . h and M d y . h indicate that the water holding capacities are highly correlated with environmental parameters that are cycled every 24 h.

As side evaluation indices, dynamic water loss, M d y . l , and dynamic root holding capacity, M d y . r , shown in Figure 7C , provide useful insights. Dynamic water loss is the direct evaluation of the amount wasted after the watering event ends. Reducing this evaluation index would reduce energy usage and improve system efficiency. In this experiment, reduced water loss fluctuations signify that the plant is less susceptible to environmental changes. The dynamic root holding capacity fluctuated less than other evaluation indices from the beginning of the growth progress. M d y . r signifies the power of root growth to prevent water loss. Similar to M d y . h , root surface area significantly impacts M d y . r . Therefore, M d y . r can be treated as the root evaluation index.

Plant physiology status is susceptible to environmental condition changes. In this subsection, we use the physiological indices to provide insights into the physiological properties of Plant 38 by computing and plotting the correlation coefficient matrix between the physiological indices and environmental factors and the secondary correlation coefficients matrix between the physiological indices and their temporal variability, which is measured as the moving standard deviation with a step size of four watering events. The full name of the environmental factors and evaluation indices can be found in Table 1 and Section 3.4

The correlation matrix heat map of Plant 38 between physiological indices and environmental factors is shown in Figure 8A . Several correlation coefficients are worth noting, such as the correlation between pH and M d y . h ( – 0.86 ) , M d y . l ( 0.78 ) , M c ( 0.87 ) , R w s ( – 0.81 ) , and R w e ( 0.84 ) . As an important water condition descriptor, pH affects plants’ physiology and water uptake from many perspectives. With lower pH, the higher availability of Ca and Mg ions in the liquid requires plants to uptake higher amounts of water to balance the osmotic pressure between cell membranes. Therefore, the negative correlations between pH and M d y . h and R w s signify a higher water retention at lower pH. The correlation between N o w t and M d y . h ( – 0.85 ) , M d y . l ( 0.83 ) , M c ( 0.86 ) , R w s ( – 0.83 ) , and R w e ( 0.85 ) .

The correlation matrix between the temporal variability of the physiological index and environmental factors is shown in Figure 8B . The correlations between the temporal variability of R t r . B and RH (0.57) and N o w t ( – 0.64 ) are also worth noting, meaning that the temporal variability of plant transpiration before watering responds positively to relative humidity and negatively to the number of watering events, N o w t . Similar observations have been made in Negrão et al. (2016).

Figures 9A, B are the counterparts of Figure 8A for Romaine and Oak Leaf varieties, respectively. The correlation coefficients for M c with EC and N o w t for Romaine lettuce in Figure 9A are considerably higher than the values in Figure 9B for Oak Leaf. The higher correlation coefficient averages in Figure 9B , especially for RH, EC, pH, and N o . w t suggested that Oak Leaf lettuce is more susceptible to changes of these environmental factors.