The core of the system was a multi-tier Deep Flow Technique (DFT) setup (see Figure 1 ). The system was designed with a flexible multi-tier configuration. For the present study, experiments were conducted using two tiers (19.2 m² active cultivation area) to ensure optimal environmental control and uniform experimental conditions across all cultivation units. Each tier comprised a 1.2 m × 8 m cultivation bed (9.6 m² per tier). Beds were fabricated from stainless steel square tubing with uniformly NC-machined perforations. Nutrient solution circulation was facilitated by waterproof liners forming channels within each bed. Expanded polystyrene (EPS) boards (1200×1000×150 mm) fitted with planting panels held the plants.

Figure 2 provides an overview of the constructed smart farm system layout, showing the DFT hydroponic cultivation system for Centella asiatica ( Figure 2A ) and the nutrient solution circulation system ( Figure 2B ).

The DFT approach maintained a constant nutrient solution level within the beds. Adjustable drainage outlets controlled the maximum solution depth, while pumps automatically replenished the solution as levels decreased. A compressor delivered air through six one-way valves per bed to ensure adequate dissolved oxygen levels. Operation of the compressor and artificial lights was managed by timer controllers. The nutrient delivery subsystem included three stock solution tanks (A, B, C), a raw water tank, two return tanks, and a nutrient solution chiller. Standard leafy vegetable nutrient formulations (e.g., Mulpurie) were used for solutions A and B, with diluted nitric acid in tank C for pH adjustment. Solution delivery was automated based on target EC and pH setpoints. An external chiller regulated nutrient solution temperature, particularly during warmer months. Complete nutrient solution replacement occurred monthly. Initial root growth was encouraged by providing only pH-adjusted water for approximately two weeks post-transplanting before introducing the full nutrient solution. EC and pH sensors were calibrated annually.

Environmental monitoring relied on various sensors. An EC sensor (0–20 mA output corresponding to 0.0–20.0 dS/m) and a pH sensor (0–20 mA output for 0.0–14.0 pH) monitored the nutrient solution. Substrate water content sensors (0–100% VWC range) and air temperature/humidity sensors (-40°C to +60°C range) monitored the root zone and aerial environments, respectively. A central controller featuring a 16-bit MCU managed data acquisition and system actuation. It featured 8 general-purpose input/output (GPIO) ports and communicated via RS-232 and RS-485 protocols.

The environmental control system, as shown in Figure 3 , included automated management of ceiling curtains for light regulation, humidity control through humidification systems, ventilation fan control for air circulation, and temperature control via heating and cooling systems. These integrated control mechanisms ensured optimal environmental conditions for Centella asiatica cultivation throughout the growing period.

To assess potential effects of different lighting systems in the multi-tier configuration, the first (lower) tier was equipped with 22W bar-type LEDs, while the second (upper) tier was equipped with 32W tri-phosphor fluorescent lamps. It is crucial to acknowledge that this experimental configuration places different lighting treatments on different tiers, which creates an inherent confounding between lighting type and tier position. This design characteristic represents a fundamental limitation that prevents the definitive isolation of lighting effects from other tier-related micro-environmental factors (e.g., temperature, air circulation gradients). Therefore, observed differences between tiers must be interpreted as the combined influence of lighting characteristics and tier-specific conditions, not as the effect of lighting alone. Future studies would benefit from randomized tier assignments or controlled positioning strategies to de-confound these variables. Lights were arranged in 6 rows, with a total of 12 lamp units per tier (i.e., 2 lamps per row), to promote uniform irradiance. Light intensity measurements confirmed spatial uniformity, with deviations within ±4.5% of the mean value across measurement points. Detailed specifications are provided in Table 1 .

The LED lighting system employed broad-spectrum white LEDs (22W Bar-type, commercial LED manufacturer) with the following approximate spectral distribution: blue (400–500 nm): 25%, green (500600 nm): 30%, red (600–700 nm): 35%, far-red (700–800 nm): 10%. Light intensity was maintained at 90–95 µmol·m⁻²·s⁻¹ Photosynthetic Photon Flux Density (PPFD) with a 16:8 hour photoperiod. Standard tri-phosphor fluorescent lamps (32W) provided broader wavelength distribution across the visible range (400–700 nm) with relatively uniform spectral output. The LED system exhibited approximately 30% higher photosynthetic photon efficacy (PPE) compared to the fluorescent lighting, based on power consumption and PPFD measurements.

A standard hydroponic nutrient formulation was employed based on consultation with local agricultural technology centers. Target parameters were: solution temperature 20 ± 1°C (maintained as a critical requirement), EC 0.9 ± 0.1 dS/m, and pH 5.6 ± 0.2. Dissolved oxygen was maintained above 20 ppm via continuous aeration using bubble generators. Pathogen control was assisted by a solution sterilizer. Daily monitoring and adjustments ensured stable solution quality.

The aerial environment was precisely controlled with target settings of: daytime temperature 25 ± 2°C, nighttime temperature 15 ± 2°C, relative humidity 60 ± 5%, light intensity 6,700–7,000 Lux (equivalent to 90–95 µmol·m⁻²·s⁻¹ Photosynthetic Photon Flux Density, PPFD) and CO₂ concentration 800 ± 100 ppm. Real-time monitoring and automated adjustments were performed by the control system via Zigbee and WiFi communication. A relational database management system (RDBMS) was implemented for comprehensive data logging and analysis. Sensor data (temperature, humidity, light intensity, CO₂) and visual data from PTZ and fixed cameras were collected. Actuator status data (humidifier, air conditioner, vents, fans, air curtain) were also logged. MariaDB and MySQL were utilized for data storage and management. A REST API facilitated real-time data exchange between the server and monitoring/control hardware, enabling efficient operation (see Figure 4 ). Cloud integration enhanced system stability and data accessibility.

For comparison, soil-grown C. asiatica plants were cultivated concurrently. These plants were grown in 15-cm diameter pots filled with a commercial nursery potting mix (Baroker, Seoul Bio, Eumseong, Korea). The pots were placed within the same greenhouse facility to expose them to similar ambient temperature and humidity conditions, but they received only natural sunlight filtered through the greenhouse plastic. Plants were watered daily to maintain adequate soil moisture.

At 15 weeks post-transplanting, when the plant canopy had fully developed, the aerial parts of the plants were harvested for bioactive compound analysis. Functional compounds (madecassoside, asiaticoside, madecassic acid, asiatic acid) were quantified using High-Performance Liquid Chromatography (HPLC; Agilent 1260 Infinity II, Agilent Technologies, CA, USA) equipped with a ZORBAX Eclipse Plus C18 column (4.6 × 100 mm, 3.5 m). Samples were prepared by drying harvested leaves at 60°C for 72 hours, grinding them into a fine powder, and extracting 0.1 g of powder with 10 mL of methanol via sonication for 60 minutes. The extract was filtered through a 0.45 m syringe filter prior to analysis. The mobile phase consisted of 0.1% phosphoric acid in water (A) and acetonitrile (B). The gradient elution was as follows: 0–10 min, 25-35% B; 10–20 min, 35-65% B; 20–22 min, 65-25% B; 22–25 min, 25% B. The flow rate was 1.0 mL/min, the injection volume was 10 L, the column temperature was maintained at 30°C, and detection was performed at 210 nm. Compounds were identified by comparing retention times with authentic standards (Sigma-Aldrich, St. Louis, MO, USA).

The experiment was laid out using three replicate blocks to account for potential spatial variation within the greenhouse. Within each block, the cultivation system consisted of a multi-tier unit where lighting treatments (LED and fluorescent) were assigned to fixed tier positions. We acknowledge that this fixed assignment does not constitute a fully randomized design regarding the lighting treatment. Transplanting occurred on June 10, 2022. Cuttings were placed into individual cups. For root initiation, stem bases were submerged in pH-adjusted water for two weeks. Subsequently, the full nutrient solution was introduced. Growth assessments were conducted over six months. From each treatment, ten plants were randomly sampled periodically across the three blocks to measure leaf length, leaf width, leaf count, fresh weight, dry weight, and root growth parameters. Statistical analyses were performed using SPSS 26.0 (IBM Corp., Armonk, NY, USA). Normality of data distribution was assessed using the Shapiro-Wilk test. When normality assumptions were met, independent t-tests compared means between two groups. For multi-group comparisons (demonstration farm), one-way ANOVA followed by Duncan’s Multiple Range Test (DMRT) was used. Significance levels were set at p ¡ 0.05. Results are presented as mean ± standard error of the mean (SEM). Sample sizes were determined based on available greenhouse space and resource constraints, with n=10 plants per treatment group. We acknowledge that this sample size provides limited statistical power, and future studies would benefit from larger sample sizes to enhance the robustness of the conclusions.

Throughout the cultivation period, light intensity measurements confirmed stable conditions within the target range of 6,700–7,000 Lux (90–95 µmol·m⁻²·s⁻¹) for both LED and fluorescent treatments. Spatial uniformity analysis across six points per tier indicated deviations within ±4.5% of the mean, ensuring a consistent light environment.

The smart farm control system successfully maintained air temperatures at 25 ± 2°C (day) and 15 ± 2°C (night), and relative humidity around 60 ± 5%. This stable control effectively mitigated the extreme temperature fluctuations often encountered in traditional soil cultivation during summer and winter.

The DFT system provided stable control over key nutrient solution parameters. Temperature was maintained at 20 ± 1°C, EC at 0.9 ± 0.1 dS/m, and pH at 5.6 ± 0.2. Continuous aeration ensured dissolved oxygen levels remained above 20 ppm, fostering a favorable root-zone environment.

Comparing cultivation tiers, madecassoside content did not differ significantly between the LED-equipped and fluorescent-equipped tiers (2.63 ± 1.18 mg/g vs. 2.12 ± 0.98 mg/g; t(18) = 0.576,p = 0.572). However, asiaticoside content was significantly higher in the LED-equipped tier (52.08 ± 1.18 mg/g) compared to the fluorescent-equipped tier (40.90 ± 1.49 mg/g; t(18) = 10.196,p < 0.001). Similarly, madecassic acid content was 2.3-fold higher in the LED-equipped tier (1.25 ± 0.04 mg/g) than in the fluorescentequipped tier (0.54 ± 0.03 mg/g; t(18) = 24.857,p < 0.001). Asiatic acid content was also 1.6-fold higher in the LED-equipped tier (0.81 ± 0.03 mg/g) versus the fluorescent-equipped tier (0.51 ± 0.02 mg/g; t(18) = 12.786,p < 0.001). Again, we emphasize that these differences reflect the combined effects of tier-specific cultivation conditions rather than isolated lighting effects.

At the demonstration farm, the position of cultivation tiers influenced bioactive compound levels. Madecassoside content did not vary significantly across the three tiers (1st floor: 1.84 ± 0.48 mg/g; 2nd floor: 2.11 ± 0.72 mg/g; 3rd floor: 2.10 ± 0.54 mg/g; DMRT, p¿0.05). However, for asiaticoside, the 2nd (43.48 ± 1.12 mg/g) and 3rd (41.79 ± 1.49 mg/g) tiers showed significantly higher content (DMRT, p¡0.05) than the 1st (bottom) tier (31.76 ± 0.32 mg/g). Madecassic acid content increased with tier height, being lowest on the 1st floor (0.86 ± 0.04 mg/g), intermediate on the 2nd (1.33 ± 0.04 mg/g), and highest on the 3rd (top) tier (1.51 ± 0.01 mg/g; DMRT, p¡0.05). A similar trend was observed for asiatic acid, with the 2nd (1.10 ± 0.03 mg/g) and 3rd (1.19 ± 0.04 mg/g) tiers showing higher content (DMRT, p¡0.05) than the 1st tier (0.89 ± 0.08 mg/g).

When comparing compound profiles between cultivation methods, plants grown in the fluorescentequipped hydroponic tier showed significant differences compared to soil-grown plants. For example, asiaticoside content was significantly lower in fluorescent hydroponics (40.90 ± 1.49 mg/g) compared to soil cultivation (67.64 ± 1.82 mg/g; t(18) = −19.692,p < 0.001). Conversely, madecassic acid content was significantly higher in fluorescent hydroponics (0.54 ± 0.03 mg/g) than in soil cultivation (0.46 ± 0.04 mg/g; t(18) = 2.807,p = 0.012). Asiatic acid content was lower in fluorescent hydroponics (0.51 ± 0.02 mg/g) compared to soil cultivation (0.60 ± 0.01 mg/g; t(18) = −5.592,p < 0.001).