论文信息

Ali Ahmad / Francisco Javier Diaz Blasco / Sandra Sendra 等 5 人

Precision Agriculture

论文解读

AI 辅助整理
方法
低成本RGB传感器+Arduino/Wi-Fi采集叶片反射率,结合PCA、典型相
发现
传感器指数可在可见褪绿前检测到光谱变化,实现盐胁迫早期无损诊断。
机会点
可将该低成本RGB+无线方案扩展至田间多作物盐胁迫实时监测,并验证跨品种与跨环境泛化性。

论文摘要

摘要 目的 盐胁迫在症状显现之前就已悄然侵蚀作物产量,因此早期、非侵入式检测对精准农业至关重要。方法 我们评估了一种低成本RGB传感器平台(并探索了集成无线网络的可能性),用于对水培芝麻菜(Eruca sativa (Mill.) Thell.)在0、40和80 mM NaCl处理下的实时表型分析。结果 该传感器捕获了七个光谱通道的反射率,原位生成叶片颜色指标。主成分分析、典型相关分析和聚类分析证实了RGB通道与离子比值之间的紧密耦合。生理、CIELAB和光学数据集通过DIABLO(利用潜在成分进行生物标志物发现的数据整合分析)进行融合,取得了优异性能(成分1的平均错误率=0.083;成分2为零),且每个数据块仅使用三个特征。高稳定性变量(Na⁺、Na⁺/K⁺、Na⁺/Mg²⁺、b*、Chroma、White、Green、Red)追踪到清晰的剂量依赖性胁迫梯度,将离子失衡与可能的色素降解及反射率变化联系起来。特别是,传感器衍生的指数在可见褪绿出现之前就检测到了光谱变化,从而实现了早期诊断。结论 这些发现表明,紧凑型RGB传感结合整合分析能够提供快速、无损且具有生物学依据的胁迫诊断。该方法可随时部署于受控或开放田间系统,为在产量损失发生之前实现精准胁迫监测提供了可扩展的途径。图形摘要 可选描述:光学传感器模块与Arduino mega 2560及Wi-Fi模块连接,正在捕获盐处理下芝麻菜叶片的反射率。图中展示了该传感器装置在深水培养系统中测量0、40和80 mM NaCl条件下生长的叶片。生理、CIELAB和传感器衍生数据通过多变量方法和DIABLO框架进行整合与分析,以验证该传感器在盐胁迫早期无损检测方面的潜力。

全文

Precision Agriculture

Precision Agriculture (2026) 27:179
https://doi.org/10.1007/s11119-026-10476-w
Evaluation of RGB sensor system for nondestructive
detection of salinity stress in arugula
Ali Ahmad1,2  · Francisco Javier Diaz Blasco2  · Sandra Sendra2  · Jaime Lloret3  ·
Francesco Di Gioia1
Received: 17 October 2025 / Accepted: 21 October 2026
© The Author(s) 2026
Extended author information available on the last page of the article
Abstract
Purpose Salinity stress silently erodes crop yields long before symptoms appear, making
early, non-invasive detection essential for precision agriculture.
Methods We evaluated a low-cost RGB sensor platform, with possible integartion of wire-
less networking, for real-time phenotyping of hydroponically grown arugula ( Eruca sativa
(Mill.) Thell.) under 0, 40, and 80 mM NaCl.
Results The sensor captured reflectance across seven spectral channels, generating leaf
color metrics in situ. Principal component analysis, canonical correlation and clustering
analysis confirmed tight coupling between RGB channels and ion ratios. Physiological,
CIELAB, and optical datasets were fused using DIABLO, (Data Integration Analysis for
Biomarker discovery using Latent components), achieving exceptional performance (mean
error rate = 0.083 for component 1; zero for component 2) with only three features per block.
High-stability variables (Na⁺, Na⁺/K⁺, Na⁺/Mg2⁺, b*, Chroma, White, Green, Red) tracked
a clear dose-dependent stress gradient, linking ionic imbalance to possible pigment deg -
radation and reflectance shifts. Particularly, sensor-derived indices detected spectral shifts
preceding visible chlorosis, enabling early diagnosis.
Conclusion These findings demonstrate that compact RGB-based sensing, empowered by
integrative analytics, can deliver rapid, non-destructive, and biologically grounded stress
diagnostics. The approach is readily deployable in controlled or open-field systems, provid-
ing a scalable pathway for precision stress monitoring before yield losses occur.
Graphical abstract
Optional description: Optical sensor module, connected to Arduino mega 2560 along with
a Wi-Fi module, is capturing reflectance of arugula leaves under salinity treatments. The
sensor setup is shown measuring leaves grown at 0, 40, and 80 mM NaCl in a deep water
culture system. Physiological, CIELAB, and sensor-derived data were integrated and ana -
lyzed using multivariate approaches and the DIABLO framework to validate the sensor’s
potential for early, non-destructive detection of salinity stress.,

1 3
179 Page 2 of 40
Precision Agriculture (2026) 27:179
Keywords Precision agriculture · Plant diagnostics · Smart sensors · Soilless
agriculture · Plant phenotyping
Introduction
Precision agriculture integrates sensors, data analytics, Artificial Intelligence (AI), remote
sensing, Global Positioning Systems (GPS), Geographic Information Systems (GIS), and
variable-rate technologies to improve crop management and input efficiency (Mgendi,
2024; Padhiary et al., 2024; S. Sharma, 2023). Nevertheless, implementation of these tools
remains constrained by cost, data security, and evolving regulations (Mgendi, 2024; Padhi-
ary et al., 2024; Xu et al., 2024).
Salinity, commonly specified with electrical conductivity above 4 dS m⁻1 (approximately
40 mM NaCl), constrains crop production (Ibrahimova et al., 2021), through osmotic and
ion-specific stress (Rastogi et al., 2020). Additionally, reactive oxygen species (ROS) subse-
quently damage membranes, enzymes, and ionic homeostasis, reducing photosynthesis and
growth (Ibrahimova et al., 2021; C. Sharma et al., 2023). Such cumulative biochemical dis-
ruptions severely constrain photosynthetic capacity, hinder biomass accrual, and culminate
in substantial yield penalties. Eruca sativa L. (Mill.) Thell. (commonly known as arugula
or cultivated rocket) is a nutrient-rich leafy crop (Di Gioia et al., 2018a; Hutchinson et al.,
2025; Stanojković-Sebić et al., 2024), but its texture, turgor, and biochemical quality are
salt-sensitive (Torres et al., 2024).

1 3
Page 3 of 40 179
Precision Agriculture (2026) 27:179
Controlled Environment Agriculture (CEA) and hydroponics support resource-efficient,
year-round production, and increasingly use smart technologies to regulate the growing
environment (Rajaseger et al., 2023; Shahab et al., 2024). In this regard, Internet of Things
(IoT) networks enable continuous data acquisition (Ahmad et al., 2024a; Mansoor et al.,
2025; Suresh et al., 2024), while machine learning (ML) and AI support growth monitoring
(Ahmad et al., 2024b; Kong et al., 2024) and early stress diagnosis based on data retrieved
through sensors (Cardenas-Gallegos et al., 2025; Nagarajan et al., 2025). For example, low-
cost, noninvasive RGB sensors could detect stress-related reflectance shifts (Diaz et al.,
2024; Stevens et al., 2024), and may support continuous hydroponic monitoring within IoT
systems (Ogbolumani & Mabaso, 2023; Stevens et al., 2024).
Despite advancements in optical sensing for plant monitoring, the application of low-
cost RGB sensors for non-invasive detection of salinity stress in hydroponically cultivated
leafy vegetables, particularly arugula, remains largely unexplored. Existing studies have
demonstrated the potential of RGB sensors in related domains but reveal critical limita -
tions in scalability, environmental applicability, and system simplicity that restrict their
broader agricultural adoption. For instance, Brambilla et al. (2021) showcased the utility of
a TCS3200 RGB sensor in detecting nitrogen deficiency in basil, with performance com -
parable to commercial leaf meters under controlled conditions. However, sensor sensitiv -
ity diminished at higher nitrogen levels, and its inability to discriminate between NO 3-N
and NH4-N forms, coupled with limited dataset size, underlines the need for more robust,
generalizable models and broader physiological applications beyond nitrogen diagnostics.
Similarly, Huichun Zhang et al., (2022a, 2022b) highlighted the value of integrating RGB,
hyperspectral, and fluorescence imaging to estimate leaf chlorophyll in sorghum, achiev -
ing R2 values of up to 0.90 through sensor fusion. Nevertheless, the reliance on complex,
high-cost imaging systems and model calibration requirements restricts practical scalability,
especially for resource-constrained environments like decentralized hydroponic farms. In
addition, H. Yang et al. (2025) advanced the field by applying deep learning to classify salt
stress in soybean using RGB leaf imagery. While the approach achieved high accuracy with
limited data, the study was confined to RGB image processing, limiting its easy deployment
to diverse crops or real-world production systems.
Furthermore, physiological studies on arugula, such as the work on salt stress mitigation
using calcium nitrate in soilless systems (Costa et al., 2024), confirmed arugula’s sensitivity
to salinity and emphasizes the crop’s relevance for the CEA industry. However, such stud -
ies focused on traditional physiological assessments and lacked integration of real-time,
sensor-based diagnostic approaches, limiting early stress detection and automated monitor-
ing potential. Finally, de Carvalho Oliveira et al. ( 2022) demonstrated that low-cost RGB
sensors could perform quantitative colorimetric analyses comparable to commercial spec -
trophotometers for liquid samples. Nonetheless, the sensor’s application was restricted to
controlled laboratory conditions and homogeneous liquid matrices, leaving its efficacy for
complex plant tissues under dynamic cultivation environments untested.
Although previous research has demonstrated notable advancements in plant optical sens-
ing, a critical gap remains regarding the application of low-cost, non-invasive RGB sensor
systems for real-time detection of salinity stress in hydroponically cultivated crops under
CEA conditions, particularly for economically relevant leafy vegetables such as arugula.
While salinity is often not a major concern in CEA soilless systems, it can become problem-
atic in closed-cycle configurations where nutrient solutions are continuously recirculated,

1 3
179 Page 4 of 40
Precision Agriculture (2026) 27:179
leading to the gradual accumulation of certain ions. Although advanced imaging platforms
such as hyperspectral, multispectral, chlorophyll fluorescence, magnetic resonance imag -
ing, X-ray computed tomography, and automated phenotyping systems have shown strong
potential for in situ crop monitoring, their wider application may be limited by equipment
cost, technical complexity, and data-processing requirements (Faehn et al., 2024; Senesi et
al., 2022). These considerations motivate the evaluation of simpler and lower-cost optical
sensing approaches, such as RGB sensors, for non-destructive assessment of crop stress.
To address this gap, the present study systematically investigates the potential of an
accessible, low-cost RGB optical sensor system for detecting physiological alterations in
arugula induced by varying salinity levels under controlled hydroponic conditions. Our
approach emphasizes the utility of such sensors as practical, scalable diagnostic tools to sup-
port intelligent, data-driven crop management in resource-efficient, soilless production sys-
tems. Although this study was conducted under controlled conditions in a soilless system,
the underlying approach and findings are likely applicable to soil-based crops. Conducting
the work in a soilless setup provided the most straightforward and controlled means to vali-
date the concept and demonstrate its effectiveness. This approach enabled precise manip -
ulation of salinity levels—an otherwise challenging task in soil-based systems—thereby
ensuring consistent stress application. By correlating sensor-derived RGB optical signals
with established physiological markers of salinity stress, we demonstrate the efficacy of
this non-invasive method for early stress detection. Our findings underscore the potential of
RGB sensing as a rapid, scalable, and non-destructive proxy for monitoring salinity-induced
stress, a critical advancement for precision agriculture and enhancing the sustainability of
nutrient-rich food production systems.
The structure of this paper is organized as follows: Sect. "Methodology" presents the
detailed methodology, including experimental design, proposed sensor system, physiologi-
cal assessments, and data analysis procedures. Section "Results and discussion" provides a
comprehensive presentation of results, integrating sensor-based findings with conventional
physiological parameters. Section "Results and discussion" also provides a critical discus -
sion, contextualizing our findings within the broader framework of plant stress monitor -
ing and sensor technology development. Finally, Sect. "Conclusion" concludes the study,
outlining key insights and future research directions for advancing intelligent hydroponic
agriculture.
Methodology
Experimental site and conditions
The experiment was conducted under controlled greenhouse conditions at The Pennsylvania
State University, USA (40°48′11.5"N, 77°51′47.4"W). Seeds of Eruca sativa Mill. (cultivar
‘Astro’) were sourced from Johnny’s Selected Seeds (Winslow, Maine, USA). Seed germi-
nation was performed using rockwool starter cubes placed in standard 25.4 by 50.8 cm Quad
Thick trays with drainage holes. Five to six seeds per starter cube were seeded to ensure
adequate plant establishment. During the germination and early growth phase, seedlings
were irrigated with tap water.

1 3
Page 5 of 40 179
Precision Agriculture (2026) 27:179
At 11 days post-germination, uniform seedlings were transplanted into deep-water cul -
ture (DWC) hydroponic systems to facilitate homogeneous nutrient availability and water
supply. The DWC system consisted of large black trays (135 × 135 cm, 17 cm depth; Botani-
care, Vancouver, WA, USA) mounted on metallic frames. Each tray was fitted with a cus -
tom-cut Styrofoam raft, in which planting holes were precisely spaced. Arugula seedlings
were established in small net pots inserted into these holes, allowing root systems to grow
out of the rockwool starter cubes into the aerated nutrient solution. Aeration was continu -
ously maintained by an air pump (60 L min⁻1; Vivosun, Ontario, CA, USA) delivering air
through four evenly distributed air stones per module, thereby sustaining optimal dissolved
oxygen levels in accordance with recommendations for high-performance hydroponic pro -
duction (Poudel & Di Gioia, 2023). Each DWC module measured 1.44 m 2 (1.2 m × 1.2 m)
and contained 36 planting holes, with four plants established per hole, resulting in a final
plant density of 100 plants per square meter.
A modified Hoagland nutrient solution was utilized, following the formulations described
by Di Gioia et al. ( 2018a) and Urlić et al. ( 2017). Two separate stock solutions containing
macro- and micro-nutrients were prepared using deionized water. The final working solution
supplied the following concentrations of macronutrients: 4.59 mmol · L⁻1 nitrate (NO 3⁻),
0.50 mmol · L⁻1 phosphate (H2PO4⁻), 2.74 mmol · L⁻1 potassium (K⁺), 0.94 mmol · L⁻1 sul-
fate (SO42⁻), 1.35 mmol · L⁻1 calcium (Ca2⁺), 0.51 mmol · L⁻1 magnesium (Mg2⁺), and 0.50
mmol · L⁻1 ammonium (NH4⁺). Sodium and chloride concentrations were negligible (0.0004
mmol · L⁻1 Na⁺ and 0.0 mmol · L⁻1 Cl⁻, respectively), reflecting the intended low-salinity
baseline conditions of the nutrient solution. Micronutrient concentrations were adjusted
to 10.35 µmol · L⁻1 iron (Fe 3⁺), 5.71 µmol · L⁻1 manganese (Mn 2⁺), 0.20 µmol · L⁻1 copper
(Cu2⁺), 0.70 µmol · L⁻1 zinc (Zn2⁺), 7.68 µmol · L⁻1 boron (B3⁺), and 0.21 µmol · L⁻1 molyb-
denum (Mo6⁺).
The nutrient solution was administered to all DWC modules using a proportioning pump
(Dosatron International, LLC., Clearwater, FL, USA) proportional dosing system, equipped
with a liter counter, ensuring precise and uniform composition of the nutrient solution across
treatments. by adding either a dilute acid (phosphoric acid) or a base (potassium hydroxide),
following standard pH adjustment protocols.
Salinity treatments were established 9 days after transplanting to induce controlled abi -
otic stress. Based on previously validated experimental designs for vegetable crops (Corti
et al., 2023; Di Gioia et al., 2013; Urlić et al., 2017), three salinity levels were applied
by supplementing the nutrient solution with laboratory-grade sodium chloride (NaCl) to
achieve final concentrations of 0 mM (control), 40 mM (moderate salinity), and 80 mM
(high salinity).
The greenhouse environmental conditions were continuously monitored throughout the
experimental period. A photoperiod of 10 h of daylight and 14 h of darkness was maintained.
Temperature within the greenhouse ranged from a minimum of 17.18 °C to a maximum
of 39.76 °C, with an average daily temperature of 23.50 °C. Relative humidity fluctuated
between 11.62% and 75.41%, with a mean value of 35.96%.
Plants were harvested eight days after the imposition of salinity treatments. Prior to
harvest, leaf reflectance properties were evaluated using the proposed RGB optical sensor
system to characterize changes in leaf pigmentation associated with salinity stress. Mea -
surements were conducted on a healthy, fully expanded, and blemish-free leaf per plant.
Five plants constituted one biological replicate, and four independent biological replicates

1 3
179 Page 6 of 40
Precision Agriculture (2026) 27:179
were evaluated for each salinity treatment. Accordingly, measurements were first consoli -
dated at the leaf and plant levels and subsequently at the biological-replicate level to avoid
treating repeated measurements from the same plant as independent observations. Whereas,
for physiological assessments, four biological replicates were employed.
Following the non-destructive physiological assessments, plants were harvested, and
biometric parameters, including fresh biomass (FW; g/m 2), leaf area (LA; cm 2/plant), and
relative water content (RWC; %), were determined using standardized laboratory protocols.
A subset of leaf samples was oven-dried to constant weight at 65 °C to determine dry weight
(DW; mg/m2). Oven-dried samples were milled and passed through a 1-mm sieve and used
for mineral profile determinations.
For biochemical analyses, leaf tissues were lyophilized to constant dry weight using a
LabConco Freeze Dryer (12 L capacity, Model 701,211,150; LabConco Corp., Kansas City,
MO, USA). Freeze-dried samples were subsequently ground to a fine powder using a mortar
and pestle under liquid nitrogen and stored at − 80 °C until further biochemical analyses.
Minerals and nitrate determination
Leaf tissue samples previously oven-dried to constant weight were finely ground using a
grinding mill under laboratory conditions. A portion of the powdered material was, submit-
ted to the Agricultural Analytical Services Laboratory at The Pennsylvania State University
(University Park, PA, USA) for mineral element determination. Total nitrogen (N) concen-
trations were quantified via combustion analysis using a CN elemental analyzer, following
the methodological framework described by Vecchia et al. (2020).
For comprehensive elemental profiling, ground samples underwent acid digestion
employing concentrated nitric acid (HNO3, 70%) and perchloric acid (HClO4, 60%), adher-
ing to established protocols for plant mineral extraction (Huang & Schulte, 1985). The
digested material was analyzed for macro- and meso-minerals, including phosphorus (P),
potassium (K), calcium (Ca), magnesium (Mg), sulfur (S), and sodium (Na), as well as
for microminerals such as manganese (Mn), iron (Fe), copper (Cu), boron (B), and zinc
(Zn), using inductively coupled plasma mass spectrometry (ICP-MS). Macro-minerals
were reported in percentage of DW, while microminerals were reported as mg/kg of DW.
Likewise, the Na⁺/K⁺, Na+/Ca+2, and Na+/Mg+2 ratios were calculated for each replicate as
reported previously (Maathuis & Amtmann, 1999).
Nitrate concentrations in leaf tissues were assessed using a colorimetric approach based
on the modified salicylic acid method, initially described by Cataldo et al. ( 1975). Briefly,
10 mg of freeze-dried, ground leaf material was extracted in 1 mL of deionized water. The
suspension was vortexed for 10 s and subsequently incubated in a water bath at 45 °C for
60 min to facilitate the solubilization of nitrates. Following extraction, samples were cen -
trifuged at 5,600 × g for 15 min, and the supernatant was collected for analysis. To quantify
nitrate, 20 µL of each extract, standard, or blank (deionized water) was mixed with salicylic
acid in sulfuric acid solution (5% w/v). Control blanks consisted of 100% sulfuric acid with-
out salicylic acid. After incubation at room temperature for 20 min, 1.9 mL of 2 M sodium
hydroxide (NaOH) was added to neutralize the reaction. After a 2-min stabilization period,
300 µL aliquots of the final mixture were transferred to a 96-well microplate. Absorbance
was measured at 410 nm using a Synergy H1 microplate reader (BioTek Instruments, Win-

1 3
Page 7 of 40 179
Precision Agriculture (2026) 27:179
ooski, VT, USA). Nitrate concentrations were expressed on a fresh weight basis (mg/kg
FW) using potassium nitrate as an external calibration standard.
Quantification of chlorophyll and carotenoid pigments
Total chlorophyll and carotenoid contents were quantified from freeze-dried plant material
using a solvent extraction method based on Porra et al. (1989), with methodological refine-
ments as described by Darlan et al. (2025). Approximately 15 mg of finely ground, lyophi-
lized leaf powder was suspended in 1.5 mL of 80% acetone. The mixture was subjected to
ultrasonic-assisted extraction for 25 min using an ultrasonic processor (Branson Ultrason -
ics, Brookfield, CT, USA) to maximize pigment solubilization. Extracts were subsequently
centrifuged at 4,000 × g for 5 min, and the supernatant was collected for spectral analysis.
Absorbance readings were obtained at 663 nm (A663), 645 nm (A645), and 470 nm (A470)
using a Synergy H1 microplate reader (BioTek Instruments, Winooski, VT, USA). Concen-
trations of chlorophyll a (Chl a) (1), chlorophyll b (Chl b) (2), total chlorophyll (3), and
carotenoids (4) were calculated using the following equations:
𝐶𝐶𝐶𝐶𝐶 𝐶𝐶 𝐶 𝐶𝐶𝐶𝐶𝐶 𝐶 𝐶𝐶 663 − 𝐶𝐶79 𝐶 𝐶𝐶645 (1)
𝐶𝐶𝐶𝐶𝐶 𝑏𝑏 𝐶 𝐶𝐶𝐶𝐶𝐶 𝐶 𝐶𝐶 645 − 𝐶𝐶𝐶𝐶 𝐶 𝐶𝐶663 (2)
𝑇𝑇𝐶𝐶𝑇𝑇𝐶𝐶𝐶𝐶 𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝑇𝑇𝐶𝑇𝑇𝐶𝐶𝐶𝐶 𝐶 𝐶𝐶𝐶𝐶𝐶 𝐶𝐶 + 𝐶𝐶𝐶𝐶𝐶 𝑏𝑏 (3)
𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝑇𝑇𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶 𝐶 𝐶𝐶𝐶𝐶 𝐶 𝐶𝐶470 − 𝐶𝐶8𝐶 𝐶 𝐶𝐶𝐶𝐶𝐶 𝐶𝐶 − 8𝐶𝐶𝐶𝐶 𝐶 𝐶𝐶𝐶𝐶𝐶 𝑏𝑏𝐶98 (4)
All pigment concentrations were expressed on a fresh weight basis (mg/g FW).
CIELAB colorimetry
Leaf surface color was quantitatively evaluated to assess the visual and physiological
responses of salinity stress in arugula. Color measurements were performed on the adaxial
(upper) surface of fully developed leaves using a portable Chroma Meter (Model CR-400,
Konica Minolta, Tokyo, Japan), ensuring consistent measurement geometry and minimiz -
ing variability due to leaf curvature or surface heterogeneity. The instrument was calibrated
with its reference tile prior to use according to the manufacturer’s specifications.
Colorimetric parameters were recorded for leaves exposed to 0, 40, and 80 mM NaCl
treatments using the CIELAB color space framework, which characterizes color based on
three orthogonal coordinates: lightness (L*), the red-green axis (a*), and the yellow-blue
axis (b*). To further characterize leaf coloration, Chroma (C) (5) and hue angle (h°) (6) were
calculated from the a and b coordinates as reported previously (Cruz et al., 2020). Likewise,
the Saturation Index (SI) (Wilms & Oberfeld, 2018) (7), indicative of color vividness rela-
tive to lightness, was also computed.
𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶 𝐶𝐶𝐶𝐶 𝐶
√
𝐶𝐶2 +𝑏𝑏 2 (5)

1 3
179 Page 8 of 40
Precision Agriculture (2026) 27:179

𝐻𝐻𝐻𝐻𝐶𝐶 𝐶𝐶 ∘𝐶 𝐶 𝐶𝐶𝐶𝐶𝑎𝑎𝑇𝑇𝐶𝐶𝐶𝐶𝑎𝑎 𝐶𝐶𝐶𝐶𝑇𝑇 ( 𝑏𝑏𝐶𝐶 ) 𝐶 𝐶8𝐶∘𝜋𝜋 (6)
𝑆𝑆𝐶𝐶𝑇𝑇𝐻𝐻𝐶𝐶𝐶𝐶𝐶𝐶𝑇𝑇𝐶𝐶𝐶𝐶𝐶𝐶 𝑆𝑆𝐶𝐶𝐶𝐶𝐶𝐶𝑆𝑆 𝐶𝑆𝑆𝑆𝑆𝐶 𝐶 𝐶𝐶𝐿𝐿∗ (7)
Chroma represents the color saturation or intensity, while hue angle reflects the dominant
color tone on a circular scale, expressed in degrees.
To quantify perceptible differences in leaf coloration between treatments, three standard-
ized color difference metrics were applied: the CIE 1976 color difference (ΔE 76), the CIE
1994 index (ΔE94), and the CIEDE2000 model (ΔE00), each providing progressive improve-
ments in the correction of perceptual non-uniformities in color space representation.
The CIE 1976 color difference (ΔE76) was computed as the Euclidean distance between
two points in the CIELAB space (Robertson, 1977) (8):
Δ𝐸𝐸76 𝐶 √𝐶𝐿𝐿1 −𝐿𝐿 2𝐶2 + 𝐶𝐶𝐶1 −𝐶𝐶 2𝐶2 + 𝐶𝑏𝑏1 −𝑏𝑏 2𝐶2 (8)
where L1, a1, and b1 represent the lightness and chromaticity coordinates of the first sample,
and L2, a2, and b2 those of the second.
The CIE 1994 index (ΔE94) introduces weighting factors to account for perceptual differ-
ences in lightness, chroma, and hue (Kheng, 2002), as follows (9):
Δ𝐸𝐸94 𝐶 √𝐶Δ𝐿𝐿𝐿 𝐶𝑘𝑘𝐿𝐿 ⋅ 𝑆𝑆𝐿𝐿𝐶𝐶2 + 𝐶Δ𝐶𝐶𝐿 𝐶𝑘𝑘𝐶𝐶 ⋅ 𝑆𝑆𝐶𝐶 𝐶𝐶2 + 𝐶Δ𝐻𝐻𝐿 𝐶𝑘𝑘𝐻𝐻 ⋅ 𝑆𝑆𝐻𝐻 𝐶𝐶2 (9)
Here, ΔL, ΔC, and ΔH denote differences in lightness, chroma, and hue, respectively; kL, kC,
and kH are parametric factors typically set to unity in non-textile applications; and SL, SC,
and SH are scaling functions that adjust for the influence of each color component.
The most advanced perceptual color difference model, CIEDE2000 (ΔE00), incorporates
additional correction terms to account for non-linearities in human color perception, par -
ticularly in regions of low chroma and blue hues. The ΔE 00 is defined as follows (Kheng,
2002) (10):
Δ𝐸𝐸00 𝐶 √𝐶Δ𝐿𝐿𝐿𝐿 𝐶𝑘𝑘𝐿𝐿 ⋅ 𝑆𝑆𝐿𝐿𝐶𝐶2 + 𝐶Δ𝐶𝐶𝐿𝐿 𝐶𝑘𝑘𝐶𝐶 ⋅ 𝑆𝑆𝐶𝐶 𝐶𝐶2 + 𝐶Δ𝐻𝐻𝐿𝐿 𝐶𝑘𝑘𝐻𝐻 ⋅ 𝑆𝑆𝐻𝐻 𝐶𝐶2 +𝑅𝑅 𝑇𝑇 ⋅ 𝐶Δ𝐶𝐶𝐿𝐿 𝐶𝑘𝑘𝐶𝐶 ⋅ 𝑆𝑆𝐶𝐶 𝐶𝐶 ⋅ 𝐶Δ𝐻𝐻𝐿𝐿 𝐶𝑘𝑘𝐻𝐻 ⋅ 𝑆𝑆𝐻𝐻 𝐶𝐶 (10)
In this formulation, ΔL′, ΔC′, and ΔH′ correspond to perceptually adjusted differences in
lightness, chroma, and hue, respectively. The factors kL, kC, and kH are parametric scaling
constants (typically set to 1), SL, SC, and SH are weighting functions that normalize the con-
tributions of each color component, and RT is a rotation term that compensates for interac -
tive effects between chroma and hue, especially relevant in blue color regions.
These colorimetric indices provided a comprehensive and objective quantification of
salinity-induced alterations in leaf pigmentation, facilitating the interpretation of physi -
ological and stress-related visual changes under saline conditions.

1 3
Page 9 of 40 179
Precision Agriculture (2026) 27:179
Fig. 1 Overview of the TCS3200 RGB optical sensor. (a) Structural components of the TCS3200 sensor,
including photodiodes and filtering elements. (b) Block diagram illustrating the functional architecture of
the sensor, showing the conversion of light intensity into frequency-based digital output. (c) Experimental
application and operating principle of the sensing module, including sensor positioning over the leaf can-
opy, integrated LEDs shown in illuminated and non-illuminated states, approximate physical dimensions
of the sensor unit, and the incident–reflected light pathway used for nondestructive leaf optical measure-
ments. (d) Circuit schematic of the TCS3200 interfaced with an Arduino Mega 2560. (e) Data acquisition
and processing workflow for vegetation index calculation and plant stress detection

Sensing system
An optical sensor system based on a programmable RGB optical module (TCS3200) was
employed for the non-destructive assessment of salinity-induced optical changes in arugula
leaves. The experimental system consisted of the TCS3200 sensor interfaced with an Ardu-
ino Mega 2560, which controlled photodiode selection and acquired the frequency-based
sensor output. The objective was to determine whether variations in leaf optical response
could provide measurable information associated with increasing salinity stress. Likewise,
the TCS3200 module was interfaced with an Arduino Mega 2560, selected for its memory
and 54 digital Input/Output (I/O) pins, which supported multi-channel data acquisition.
Selected sensor nodes could also be incorporated into Raspberry Pi 4 units for data process-
ing, storage, and transfer to a centralized access point for downstream analysis.
The TCS3200 optical sensor integrates an array of 64 photodiodes, systematically orga-
nized into four filter groups: 16 red-filtered, 16 green-filtered, 16 blue-filtered, and 16 unfil-
tered diodes (Fig. 1a). These photodiodes are spatially interleaved across the silicon dye to
mitigate spatial non-uniformities in incident light, thereby improving spectral fidelity and
measurement robustness. Diodes of the same spectral type are connected in parallel and
evenly distributed, a configuration designed to minimize positional error and ensure homo-
geneous detection across the sensor surface (TAOS, 2025).
The RGB sensor outputs a square wave signal with a frequency proportional to the inten-
sity of the incident light. The output frequency scaling, programmable at 100%, 20%, or 2%,
facilitates compatibility with a wide range of acquisition systems and improves dynamic
range management, particularly in applications with limited signal processing bandwidth
(e.g., microcontroller-based platforms) (Fig. 1b). In this study, the sensor’s output frequency

1 3
179 Page 10 of 40
Precision Agriculture (2026) 27:179
Fig.  2 Proposed hierarchical wireless sensor network (WSN) architecture for greenhouse-based plant
monitoring using RGB optical sensing nodes. Distributed sensing nodes acquire leaf reflectance data and
transmit measurements via Wi-Fi to a cluster head. The cluster head forwards aggregated data through
a LoRaWAN gateway to a cloud platform, enabling remote storage, visualization, and analysis for plant
health assessment and decision support

ranged from 2 Hz to 500 kHz, with the appropriate scale factor selected to optimize signal
clarity while avoiding saturation or loss of low-intensity data.
Measurements were performed directly on intact plants under ambient greenhouse condi-
tions (Fig. 1c). During each measurement, the sensor was positioned approximately 1–2 cm
above the leaf surface without physical contact, while the integrated LEDs of the TCS3200
module provided localized illumination. Readings were taken by manually positioning the
sensing unit and moving between plants. Maintaining a short and consistent sensor-to-leaf
distance was intended to reduce variations associated with measurement geometry while
allowing rapid, non-destructive acquisition.
The sensing configuration evaluated in this study operated as a standalone system; wire-
less communication, cloud connectivity, and continuous real-time monitoring were not
experimentally implemented. Nevertheless, because the TCS3200 provides a digital fre -
quency output compatible with embedded microcontrollers, the sensing approach could
potentially be extended to automated or periodically operated monitoring systems. In prac-
tical applications, several deployment schemes could be considered depending on produc -
tion scale and monitoring objectives. For example, fixed sensing nodes could be installed
at representative locations within greenhouse production beds for repeated measurements
over time. Whereas portable or handheld units could be used for targeted assessment of indi-
vidual plants or treatment zones. Alternatively, multiple sensing units could be distributed
throughout the greenhouse and incorporated into a hierarchical wireless sensor network
(WSN) for spatially resolved monitoring, as conceptually illustrated in Fig. 2.
A hybrid configuration combining fixed nodes for routine surveillance with portable
measurements for confirmatory assessment could also be feasible. Such deployment strat -

1 3
Page 11 of 40 179
Precision Agriculture (2026) 27:179
egies could support periodic mapping of plant optical responses across production areas
and, following further validation, could provide a basis for integrating RGB sensing into
decision-support systems for greenhouse crop management. However, these deployment
configurations remain prospective applications of the sensing approach and were not imple-
mented or experimentally evaluated in the present study. In such a proposed architecture,
individual sensing nodes could acquire leaf optical measurements and transmit the data via
Wi-Fi to a nearby cluster head. The cluster head could then aggregate measurements from
multiple nodes and relay the data through a longer-range communication protocol, such as
Long-Range Wide Area Network (LoRaWAN), to a central gateway. Subsequently, the gate-
way could forward the collected data through an Internet connection to a cloud-based plat -
form for storage, visualization, and further analysis through a user interface (Ahmad et al.,
2026; Haozhe Zhang et al., 2022a, 2022b). This type of architecture could facilitate scalable
monitoring across greenhouse production areas while reducing dependence on direct wired
connections between individual sensing nodes and the central processing system. However,
the WSN configuration shown in Fig. 2 represents a proposed application framework for
extending the present RGB sensing approach and was not implemented or experimentally
validated in this study.
Data acquisition and preprocessing
Sensor measurements were obtained from three healthy, fully expanded, and visually blem-
ish-free leaves on each plant to account for within-plant variability in leaf optical properties.
All measurements were conducted between 09:00 and 11:00 h to minimize diurnal variation
in plant physiological status and reduce variability in ambient sunlight conditions that could
influence optical sensor responses. Measurements were performed by biological replicate,
with all plants within a given replicate assessed within the same sampling period before
proceeding to the next replicate.
The responses recorded from the three leaves were averaged to generate a single repre -
sentative sensor value for each plant. Five plants constituted one biological replicate, and
four independent biological replicates were evaluated for each salinity treatment. Con -
sequently, each treatment comprised 20 plants (5 plants × 4 biological replicates), corre -
sponding to 60 individual leaf measurements (3 leaves × 20 plants). Following within-plant
averaging, 20 plant-level sensor responses were retained per treatment for subsequent data
processing and statistical evaluation, while maintaining their grouping within the four bio -
logical replicates. This hierarchical sampling approach was intended to reduce the influence
of leaf-to-leaf variability and treating multiple measurements from the same plant as inde -
pendent observations.
The sensor was calibrated to sequentially interrogate seven discrete spectral channels,
encompassing both primary (Red, Green, Blue) and composite (Cyan, Magenta, Yellow,
White) visible wavelengths. Subsequently, vegetation indices (VIs), Eqs. (11) to (16),
were derived directly from raw sensor outputs (intensity-based reflectance values) for Red,
Green, Blue, and White channels to quantify stress-induced spectral changes as reported in
previous studies (Huete, 2014; Özlüoymak, 2023; Wang et al., 2023) (Table 1).
The proposed operational workflow demonstrates a fully automated protocol for continu-
ous and real-time spectral monitoring with an RGB optical sensor. The process begins with
system initialization, during which the computational node and RGB sensor are powered,

1 3
179 Page 12 of 40
Precision Agriculture (2026) 27:179
Vegetation index Function Formula
Excess Green
(ExG)
Empha-
sizes green
reflectance
over Red/
Blue
𝐸𝐸𝐸𝐸𝐸𝐸 𝐸 2𝐸𝐸 𝐸 𝐸𝐸 𝐸 𝐸𝐸 (11)
Excess Red
(ExR)
Sensitive to
antho-
cyanin-
related red
increases
𝐸𝐸𝐸𝐸𝐸𝐸 𝐸 1𝐸4𝐸𝐸 𝐸 𝐸𝐸 (12)
Normalized
Green–Red
Difference Index
(NGRDI)
Normalizes
pigment
balance
changes
𝑁𝑁𝐸𝐸𝐸𝐸𝑁𝑁𝑁𝑁 𝐸 𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐺 (13)
Red-to-Green
Ratio (R/G)
Proxy for
chlorophyll
degradation
𝐸𝐸𝐸𝐸𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅 𝐸 𝐺𝐺𝐺𝐺 (14)
White-to-RGB
Ratio (W/RGB)
Broadband
reflectance
proportion
𝑊𝑊 𝐸 𝐸𝐸𝐸𝐸𝐸𝐸𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅 𝐸 𝑊𝑊𝐺𝐺𝐺𝐺𝐺𝐺𝑅𝑅 (15)
White-to-Green
Ratio (W/G)
Evaluates
white-light
contribu-
tion relative
to green
reflectance
𝑊𝑊𝐸𝐸𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅 𝐸 𝑊𝑊𝐺𝐺 (16)
Table 1 Summary of vegetation
index formulas along with their
functional roles in assessing
plant physiological status,
derived from optical sensor mea-
surements. Indices are calculated
using RGB reflectance values
and are applied to quantify traits
such as greenness and pigment
concentration

synchronized, and configured for uninterrupted acquisition. This ensures hardware–soft -
ware alignment before data capture begins. The system then enters a continuous acquisition
loop, sequentially interrogating each predefined spectral channel. For each channel, the sen-
sor records a frequency signal, which is algorithmically converted into a calibrated intensity
value: transforming raw sensor output into a standardized, quantitative metric. These values
are stored in local memory to enable structured handling and inter-channel comparisons.
Following acquisition across all channels, the system executes a threshold evaluation,
comparing each intensity value against channel-specific reference limits for anomaly detec-
tion or condition monitoring (Fig. 3). Exceeding any threshold triggers an immediate user
alert through the designated interface, enabling rapid intervention. If all intensities remain
within acceptable bounds, the complete multispectral dataset is logged for archival and
further analysis. A fixed sampling delay (3 s) regulates temporal resolution, balancing
responsiveness with data redundancy prevention. This loop operates continuously, ensuring
uninterrupted environmental sensing.
Raw sensor outputs, expressed as frequency data, were normalized via min–max scaling
to a standardized (0, 1) range. All preprocessing pipelines were implemented using Python-
based analytical libraries to ensure computational reproducibility, methodological rigor, and
statistical robustness. This standardization effectively neutralized potential confounders,
such as ambient light variability, sensor drift, and measurement noise, thereby enhancing
the reliability and interpretability of spectral reflectance outputs.

1 3
Page 13 of 40 179
Precision Agriculture (2026) 27:179
Fig. 3 Schematic diagram of the automated workflow for real-time multispectral monitoring using an
RGB optical sensor. The diagram illustrates the operational pipeline enabling continuous spectral acquisi-
tion with the TCS3200 optical sensor

1 3
179 Page 14 of 40
Precision Agriculture (2026) 27:179
Fig. 4 Workflow for integrative analysis of physiological, colorimetric, and optical sensor data. The pipe-
line begins with data collection across three domains: physiological traits, CIELAB colorimetry, and
RGB-based spectral responses from arugula grown at three salinity levels (0,40, and 80 mM NaCl).
Following statistical screening using analysis of variance (ANOV A) and Tukey’s honest significant differ-
ence (HSD) test, significant variables proceed to two analytical branches. The first comprises multivari -
ate methods including principal component analysis (PCA), canonical correlation analysis (CCA), and
clustering to uncover patterns and interrelationships. The second utilizes the DIABLO (Data Integration
Analysis for Biomarker discovery using Latent cOmponents) framework to integrate all three data blocks
for multivariate data fusion, and uncovering the most influential features for non-invasive detection of
salinity stress through optical sensor

Data analysis
The data integration and analysis workflow are depicted in Fig. 4. It begins with the acqui-
sition of three primary data types: physiological parameters, CIELAB colorimetry values,
and RGB optical sensor outputs, collected from arugula plants grown under three levels of

1 3
Page 15 of 40 179
Precision Agriculture (2026) 27:179
salinity stress (0, 40, and 80 mM NaCl) in a DWC growing system. Following data col -
lection, univariate statistical analysis is performed using analysis of variance (ANOV A)
and Tukey’s honest significant difference (HSD) test to identify variables with significant
treatment effects. Upon detection of significance, the pipeline bifurcates into two analytical
pathways. The workflow then diverges into two complementary analytical streams. First, a
multivariate analysis path involves principal component analysis (PCA) for pattern recogni-
tion and dimensionality reduction, canonical correlation analysis (CCA) to explore inter-
dataset relationships, and clustering techniques to classify samples based on phenotypic
similarity. Second, a data integration path utilizes the DIABLO (Data Integration Analysis
for Biomarker discovery using Latent cOmponents) framework, which combines the three
data blocks—physiological traits, CIELAB color coordinates, and RGB sensor outputs
i.e., to identify correlated variables and extract latent components for potential biomarker
discovery.
Biochemical parameters
Plant physiological data were analyzed in R (version 4.5.1) using a sequential approach.
A one way ANOV A was conducted to evaluate the effects of salt treatments, followed by
Tukey’s HSD test ( agricolae package) to identify pairwise differences among treatment
means. Statistically distinct means were assigned different letters to indicate significance.
Variables showing significant differences (p < 0.05) were retained for subsequent statistical
analyses.
Multivariate statistical analysis
A comprehensive multivariate statistical and correlation-based network analysis framework,
as reported previously (Kumar & Mukhtar, 2023; Mertler et al., 2021), was implemented
to elucidate the latent structure, covariation patterns, and interdependencies among opti -
cal sensor output and physiological measurements of arugula. This integrative approach,
grounded in principles from systems biology, graph theory, and unsupervised machine
learning, enabled both dimensionality reduction and the detection of trait co-regulation
patterns under variable salt exposure. All statistical computations and network analyses
were performed in Python (version 3.11), utilizing established scientific libraries including
NumPy, Pandas, Scikit-learn, Seaborn, and NetworkX, ensuring computational reproduc -
ibility and methodological consistency.
The PCA was employed to reduce dimensionality and visualize treatment-induced phe -
notypic variance, thereby circumventing confounding effects from categorical descriptors.
The first two principal components (PC1 and PC2) were extracted to construct biplots.
Likewise, CCA (17) was conducted using identical sets of biometric traits for both canoni -
cal variable groups (X = Y). This symmetric design permitted the extraction of maximally
correlated canonical variables, providing a reduced-dimensional representation of trait co-
dependencies. The first canonical pair (U1, V1) was visualized to evaluate spatial separation
among treatment groups, and the corresponding canonical scores were preserved for down-
stream multivariate interpretation. Furthermore, a Pearson correlation matrix was computed
to quantify pairwise linear relationships among all biometric traits, using the following
formula:

1 3
179 Page 16 of 40
Precision Agriculture (2026) 27:179

𝐶𝐶𝑅𝑅𝑖𝑖 𝐶
𝑎𝑎𝐶𝐶𝑐𝑐 (𝑋𝑋𝑅𝑅 ,𝑋𝑋 𝑖𝑖 )
𝜎𝜎𝑅𝑅 ⋅ 𝜎𝜎𝑖𝑖
(17)
where cov(Xi, Xj) is the covariance and σi , σj are the standard deviations of the variables Xi
and Xj, respectively. A dissimilarity matrix, defined as below (18), was derived from the
correlation matrix to highlight higher-order trait organization.
𝐶𝐶𝑅𝑅𝑖𝑖 𝐶𝐶− ||𝐶𝐶𝑅𝑅𝑖𝑖 || (18)
where rij is the Pearson correlation coefficient between traits Xi and Xj . This transformation
ensured that highly correlated traits exhibit minimal pairwise distances, enabling coherent
clustering.
Furthermore, agglomerative hierarchical clustering, using the Ward linkage criterion,
was applied to the distance matrix. Trait clusters were delineated by truncating the resulting
dendrogram at 50% of its maximal height, yielding biologically meaningful groupings of
covarying traits. Beyond matrix-based representations, correlation networks retaining only
strong associations ( ∣r∣ ≥ 0.8) were constructed. The resulting undirected, weighted graphs
were generated using NetworkX, wherein nodes representing individual biometric traits;
edges denoting significant pairwise associations; edge color encoding association polarity
(green: positive; red: negative); and edge width and gradient were scaled proportionally to
∣r∣, emphasizing interaction strength. This network topology enabled the identification of
highly connected hub traits exhibiting high connectivity and diagnostic potential of optical
sensors in the phenotypic response to salinity stress.
DIABLO analysis
Coordinated biomarker signatures underlying salinity stress responses were identified using
DIABLO (Data Integration Analysis for Biomarker discovery using Latent cOmponents),
a supervised multi-block integration framework implemented in the mixOmics R package
(Rohart et al., 2017). Although originally designed for multi-omics datasets, DIABLO is
equally suited for integrating heterogeneous, high-dimensional datasets measured on the
same experimental units (Singh et al., 2019). Its supervised architecture enables simultane-
ous class discrimination and correlation maximization across data types—an advantage for
dissecting complex traits such as salinity response, where physiological, biochemical, and
optical signals are interdependent.
All data were grouped into three distinct data blocks for DIABLO analysis as follows:
(i) Physiological traits: FW, LA, ion concentrations (N, K, Ca, Mg, Fe, Cu, B, Zn, Na) and
derived ratios (Na+/K+, Na+/Ca2+, Na+/Mg2+).
(ii) CIELAB colorimetry: L*, a*, b*, Chroma, Hue, Saturation Index, and ΔE metrics
(ΔE76, ΔE94, ΔE00).
(iii) RGB indices: raw spectral channels (Red, Green, Blue, Cyan, Magenta, Yellow, White)
and vegetation indices such as ExG (Excess Green), ExR (Excess Red), NGRDI (Nor-
malized Green Red Difference Index), R/G (Red/Green), W/RGB (White/RGB), and
W/G (White/Green).

1 3
Page 17 of 40 179
Precision Agriculture (2026) 27:179
All variables were numeric, block-wise z-scored to remove scale effects, and integrated
using a full design matrix (all off-diagonals = 1). The number of retained variables per com-
ponent (keepX) was optimized via leave-one-out cross-validation (LOOCV) to minimize
balanced error rate and avoid overfitting. Bootstrap resampling (n = 100) was used to assess
feature stability, with selection frequency on the first component identifying robust bio -
markers (Guerrero-Sánchez et al., 2022; Singh et al., 2019).
Results and discussion
The effects of short-term salinity exposure on hydroponically grown arugula were evaluated
across a range of physiological, biochemical, and fluorescence parameters. We hypothe -
sized that increasing NaCl concentrations would elicit measurable responses; however, for a
few physiological traits measured no statistical differences were observed across treatments.
Arugula subjected to 0, 40, and 80 mM NaCl exhibited only subtle directional patterns.
For instance, while a significant treatment effect was observed for the dry matter content,
parameters such as nitrate concentration total flavonoids, pigments (total chlorophyll) and
specific leaf area were not affected by the salinity levels tested (Table S1). Similarly, appar-
ent transpiration rate, stomatal conductance, total conductance to water vapor, electron
transport rate, and chlorophyll fluorescence remained largely unchanged, except for a sig -
nificant reduction in Phase 2 slope at 80 mM (Table S2).
The absence of statistical differences observed in response to salinity stress for several
physiological parameters measured can be attributed, at least in part, to the relatively short
duration of exposure (eight days), which may not have been sufficient to trigger marked
physiological responses in arugula, a species generally regarded as moderately tolerant to
salinity (Cavalcante et al., 2025). In addition, the sensitivity of the applied measurements
may not have captured subtle or early-stage alterations to salt stress. These considerations
indicate that arugula may buffer short-term saline conditions without exhibiting pronounced
shifts in a few of the studied parameters.
Taken together, these findings provide a contextual framework for interpreting the sub -
sequent analyses of growth performance, mineral profile, and stress-related spectral shifts,
which further elucidate how arugula responds to increasing NaCl concentrations.
Effects on plant growth
Exposure to increasing NaCl concentrations resulted in a concentration-dependent decline
in arugula growth, consistent with classical salt stress responses. Specifically, plants exposed
to 80 mM NaCl showed significant reductions in shoot FW (16.5%) and leaf area (17.5%)
relative to the control, whereas shoot DW and RWC were unaffected across treatments
(Fig. 5a, b). This pronounced reduction likely reflects osmotic and ionic stress limiting
water absorption and impairing cellular expansion, as previously reported for salt-sensitive
leafy vegetables. These observations are consistent with the dual impacts of salt stress e.g.,
osmotic limitation followed by ionic toxicity (Torres et al., 2024). Initially, elevated external
NaCl reduces water potential and thereby restricts cell expansion; thereafter, progressive
Na⁺ accumulation perturbs membrane potential and enzyme systems, which further con -
strain photosynthetic carbon gain and expansion growth (Munns & Tester, 2008). Moreover,

1 3
179 Page 18 of 40
Precision Agriculture (2026) 27:179
Fig. 5 Biomass accumulation and water content in arugula. ( a) Shoot fresh and dry weight; ( b) Rela-
tive water content and leaf are. Data are presented as means (n = 4). The error bars represent standard
error (SE). Statistical differences among treatments were assessed using one-way analysis of variance
(ANOV A), followed by Tukey’s Honestly Significant Difference (HSD) post-hoc test at a significance
level of α = 0.05. Distinct lowercase letters indicate statistically significant differences between means.
The notation “ns” denotes non-significant differences

the preservation of RWC despite reduced FW suggests active osmotic adjustment or redis -
tribution of water within tissues, so that relative hydration was maintained while absolute
turgor-driven expansion was limited. In turn, such a pattern implies that energy or carbon
was diverted from growth to protective processes (for example, compatible solute synthesis,
ion compartmentalization, or antioxidant defenses) (Shahid et al., 2020; Shariatinia et al.,
2021; Tesfaye & Bayih, 2024), which would explain the decoupling between RWC and FW.
Importantly, the absence of detectable growth penalties at 40 mM NaCl indicates a threshold
response: arugula tolerates moderate salinity without substantial morphological costs, yet
the transition to 80 mM marks a physiological tipping point beyond which compensatory
mechanisms fail to fully protect expansion processes and biomass accrual. Similar results,
where salinity stress reduced cell elongation and production, have been reported previously
on broccoli and cauliflower (Di Gioia et al., 2018b; Giuffrida et al., 2017).
Effect on plant mineral profile
Arugula exhibited marked, concentration-dependent disruptions in macro- and micromin -
erals homeostasis under salinity, exhibiting both direct ionic interference and downstream
physiological adaptations to salinity stress.
The concentrations of essential cations declined markedly, with K⁺ decreasing by 18.7%
at 40 mM and 31.4% at 80 mM, and Ca2⁺ diminishing by 11% and 25.3% under the respec-
tive treatments (Fig. 6a). Collectively, these patterns point to two complementary mecha -
nisms. First, competitive inhibition at uptake sites and depolarization-driven efflux likely
explain K⁺ depletion. In saline solutions, high external Na⁺ competes with K⁺ for trans -
port via both high-affinity and low-affinity pathways, whereas membrane depolarization
increases passive K⁺ loss; consequently, cytosolic K⁺ homeostasis becomes compromised,
which disrupts osmotic adjustment and enzyme cofactor functions (Chakraborty et al., 2022;
Shabala & Cuin, 2008). Second, reduced Ca2⁺ availability will exacerbate membrane insta-
bility and impair Ca 2⁺-dependent signaling pathways that regulate stomatal behavior, ion

1 3
Page 19 of 40 179
Precision Agriculture (2026) 27:179
Fig. 6 Summary of ( a) macro- and ( b) micro-minerals profile in arugula under varying salinity stress.
Data are presented as means (n = 4). The error bars represent standard error (SE). Statistical differences
among treatments were assessed using one-way analysis of variance (ANOV A), followed by Tukey’s
Honestly Significant Difference (HSD) post-hoc test at a significance level of α = 0.05. Distinct lower -
case letters indicate statistically significant differences between means. The notation “ns” denotes non-
significant differences

Table 2 Sodium (Na⁺) accumulation and derived ion ratios for corresponding NaCl treatment
NaCl (mM) Na+ Na+/K+ Na+/Ca2+ Na+/Mg2+
0 104.3 c 0.0015 c 0.004 c 0.031 c
40 14,229.6 b 0.246 b 0.61 b 4.55 b
80 22,401.2 a 0.459 a 1.15 a 7.24 a
Data are presented as means (n = 4). Statistical differences among treatments were assessed using one-way
analysis of variance (ANOVA), followed by Tukey’s Honestly Significant Difference (HSD) post-hoc test
at a significance level of α = 0.05. Distinct lowercase letters indicate statistically significant differences
between means.
channels and stress responses, thereby creating a feedback loop that amplifies ion leakage
and dysfunction (V olkov & Beilby, 2017; White & Broadley, 2003). Hence, the measured
declines in K⁺ and Ca2⁺ plausibly account for impaired leaf expansion and reduced biomass
accumulation at 80 mM.
Nitrogen (N) concentration in arugula leaves decreased significantly under both 40 mM
and 80 mM NaCl treatments compared with the control, declining from 7.45% in the con -
trol to 7.18% and 7.22% on average, respectively (Fig. 6a). Although the magnitude of
reduction was modest, the statistical significance suggests that even moderate salinity can
adversely affect N assimilation or retention. Such a decrease may be attributed to impaired
root uptake or disruptions in N metabolism under ionic stress, potentially contributing to the
reduced growth and pigment synthesis as reported previously (Fallahi et al., 2015; Hussain
et al., 2021). In contrast, sulfur (S) and phosphorus (P) concentrations remained statistically
unchanged across treatments, indicating a relatively stable homeostasis under the imposed
salinity levels.
A drastic increase in sodium (Na⁺) accumulation was observed, with a 13.5-fold and
21.4-fold increase at 40 mM and 80 mM relative to the control, respectively ( Table 2).
This excessive Na⁺ influx was paralleled by sharp elevations in ionic ratios, with Na⁺/K⁺
increasing 16.7-fold at 40 mM and 31.2-fold at 80 mM, and comparable upward shifts
were recorded for Na⁺/Ca2⁺ and Na⁺/Mg2⁺. The extreme escalation of Na⁺ ratios provides a
quantitative explanation for the loss of physiological function. High Na⁺/K⁺ impairs protein

1 3
179 Page 20 of 40
Precision Agriculture (2026) 27:179
synthesis, enzyme activity, and stomatal regulation, and therefore directly constrains carbon
assimilation and growth. Likewise, elevated Na⁺/Ca2⁺ destabilizes membranes and signaling
cascades, which in turn promotes ion leakage and cellular damage. Consequently, these ion-
ome shifts form a coherent mechanistic chain: excessive Na⁺ accumulation leads to cationic
imbalance, followed by disrupted membrane and enzyme function, which further results in
reduced photosynthetic efficiency and cell expansion, ultimately leading to decreased fresh
mass and leaf area. Similar findings have been reported by (Di Gioia et al., 2018b; Hussain
et al., 2021) in their studies on broccoli and Chinese tamarisk production under saline stress
where they described the increase in these ratios, specifically in plant leaves and stems.
Microminerals content revealed element-specific reconfiguration, indicating altered
transport, membrane permeability, and metabolic demand (Fig. 6b). Iron (Fe) and zinc (Zn)
concentrations increased markedly under high salinity, with Fe rising by 9.5% and Zn by
48.4% at 80 mM NaCl, whereas copper (Cu) remained stable and boron (B) showed a mod-
est increase. Two non-exclusive mechanisms may explain these patterns. First, elevated
Fe and Zn could result from stress-induced upregulation of metal transporter expression
or increased root membrane permeability, both of which can support redox homeostasis
and stabilize macromolecules under oxidative stress (Kesawat et al., 2023). Second, the
apparent enrichment may partly reflect a concentration effect: when biomass accumulation
slows but ion uptake persists at baseline rates, foliar concentrations can increase despite
unchanged or even reduced whole-plant acquisition. Without whole-plant mineral budgets
or root-to-shoot flux measurements, it remains unclear whether the observed increases are
driven primarily by enhanced uptake or by this concentration artifact.
These results illustrated the hierarchical disruption of mineral homeostasis in arugula
under salinity stress, with severe K⁺ and Ca 2⁺ depletion driving ionic imbalance, accompa-
nied by selective micromineral reallocation and altered N metabolism. While partial com -
pensatory mechanisms, such as Fe and Zn accumulation, might suggest attempts to preserve
metabolic function. Though the overwhelming Na⁺ influx and Na⁺/K⁺ disequilibrium might
indicate the plant’s limited capacity to maintain ionic homeostasis under increasing salin -
ity. Collectively, the findings delineate a two-tiered response strategy: at low to moderate
salinity (40 mM), physiological adjustments, such as osmotic regulation and selective ion
transport, largely sustain growth and hydration, whereas at high salinity (80 mM), excessive
Na⁺ accumulation surpasses the plant’s capacity for selective exclusion and compartmental-
ization, leading to systemic homeostatic failure. Consequently, several testable hypotheses
emerge: (i) vacuolar compartmentalization via tonoplast Na⁺/H⁺ antiporters (e.g., NHX fam-
ily) and selective root Na⁺ exclusion (e.g., HKT transporters) are engaged but insufficient
at 80 mM; (ii) impaired K⁺ retention through channel regulation and increased K⁺ efflux
contribute to cytosolic K⁺ depletion; and (iii) antioxidant pathways are likely upregulated
but fail to fully prevent pigment loss and membrane destabilization.
CIELAB colorimetry and quantitative leaf color shifts
Quantitative colorimetric analyses using the CIELAB system revealed perceptible, dose-
dependent alterations in leaf optical properties of arugula subjected to salinity stress, thereby
providing sensitive, non-invasive markers of physiological deterioration (Figs. 7, 8).
Results showed that under severe stress (80 mM NaCl), chroma decreased by 18.76%,
while the SI declined by 13.5%, compared to the control, reflecting reduced pigment con -

1 3
Page 21 of 40 179
Precision Agriculture (2026) 27:179
Fig. 7 Three-dimensional representation of CIELAB color space under different salinity treatments. Each
colored circle represents the mean CIELAB coordinates (L*, a*, b*) for a given salinity level (n = 4). Dif-
ferent colors correspond to distinct salinity treatments

centration and diminished visual vividness (Table 2). Although the hue angle increased
moderately (2.26%), the statistical significance of this change indicates a subtle rebalancing
of pigment composition, possibly driven by preferential chlorophyll degradation relative to
carotenoids.
On the other hand, the computed ΔE values, a widely accepted metric for perceivable
color differences, exceeded the visual detection threshold of 2.3 in all salinity treatments,
with particularly large deviations in ΔE76, ΔE94, and ΔE00, which increased by 341- 220-,
and 232-fold, respectively, compared with the control (Table 3). These substantial shifts not
only surpass the human eye’s perceptibility limit but also suggest severe disruption of chlo-
roplast integrity. Such effects are consistent with reports that NaCl stress promotes reactive
oxygen species (ROS) overproduction, activating chlorophyllase and destabilizing thyla -
koid membranes (Huichun Zhang et al., 2022a, 2022b). The most prominent single optical
signature was a 9.6% reduction in L* (lightness), indicating leaf darkening, potentially due
to reduced water content, solute concentration effects, or cellular compaction under osmotic
stress (Merzlyak et al., 1999). Together, these changes in chroma, SI, and hue are consis -
tent with biochemical pigment loss and compositional shifts, while the darkening suggests
structural alterations at the cellular level, possibly accompanied by increased carotenoid or
anthocyanin-like pigment contributions, as previously observed in salt-stressed leafy crops
(Melo et al., 2017).

1 3
179 Page 22 of 40
Precision Agriculture (2026) 27:179
Fig. 8 Bivariate relationships among CIELAB color parameters in arugula leaves under salinity treat -
ments. Scatter plots show pairwise relationships between CIELAB components for leaves subjected to
0 mM, 40 mM, and 80 mM NaCl. Each colored point represents the mean value (n = 4) for a given
treatment

Table 3 Summary of color parameters derived from CIELAB data across salinity treatments
NaCl (mM) Chroma Hue Saturation Index ΔE76 ΔE94 ΔE00
0 33.16 a 122.78 b 0.74 a 1.62 b 1.28 b 1.21 b
40 28.86 b 125.19 a 0.69 ab 5.39 ab 3.5 ab 3.36 ab
80 26.94 b 125.55 a 0.64 b 7.15 a 4.09 a 4.02 a
Values include chroma, hue angle, saturation index, and color difference (ΔE) metrics calculated for each
salinity level.
Data are presented as means (n = 4). Statistical differences among treatments were assessed using one-way
analysis of variance (ANOVA), followed by Tukey’s Honestly Significant Difference (HSD) post-hoc test
at a significance level of α = 0.05. Distinct lowercase letters indicate statistically significant differences
between means
The results from CIELAB colorimetry demonstrated that salinity imposes rapid, quantifi-
able alterations in the leaf optical profile of arugula, characterized by darkening, reduced
chromatic intensity, and perceptible color divergence. The convergence of declining L*,
a*, b*, chroma values and elevated ΔE metrics provided robust, non-destructive indicators
of stress progression, which is tightly coupled to underlying physiological and biochemi -
cal deterioration. These findings reinforce the possibility of the use of low-cost, portable
colorimetric sensing platforms as effective tools for early stress detection, real-time crop
monitoring, and precision agriculture applications in hydroponic and controlled-environ -
ment systems.
Three-dimensional CIELAB projections (L*, a*, b*) further illustrate these optical shifts
(Fig. 7). Under control conditions, leaves exhibited higher L* values (46.02), denoting
brighter coloration, with more negative a* values (–18.06) and elevated b* values (28.30),
corresponding to intense green and yellow hues, respectively. Salinity induced a marked

1 3
Page 23 of 40 179
Precision Agriculture (2026) 27:179
Fig. 9 Spectral frequency responses of the optical sensor to standard reference surfaces. Mean output
frequencies are presented for the Red ( a), Green ( b), Blue ( c), Cyan ( d), Magenta ( e), Yellow (f), and
White (g) channels in response to White and Black calibration tiles. Each value represents the average of
20 independent measurements (n = 20), highlighting the sensor’s sensitivity and spectral discrimination
across color channels

decline in L* (–9.5% at 40 mM NaCl), an increase in a* (less negative, diminished green -
ness), and a reduction in b*, reflecting yellow pigment loss. Similar trends have been asso-
ciated with chlorophyll breakdown, carotenoid depletion, and structural degradation under
salt stress (Ashraf & Harris, 2013; Huichun Zhang et al., 2022a, 2022b). Likewise, arugula
plants grown in floating system have been reported to exhibit diminished color characteris-
tics, specifically L* values, under higher salinity conditions (Vetrano et al., 2020).
Bivariate projections (a* vs. b*, L vs. a*, L* vs. b*) further substantiated these observa-
tions, highlighting concurrent reductions in brightness, greenness, and yellowness under
increasing salinity (Fig. 8). Changes in leaf coloration under salinity stress were further
visualized through a polar plot of hue angle and chroma (Fig. S1). Such optical signatures
are consistent with decreased photosynthetic pigment pools and impaired leaf tissue integ -
rity, corroborating biochemical data from pigment assays (Deinlein et al., 2014; Lino et al.,
2023; Parida & Das, 2005).
RGB-based optical sensing
The application of RGB optical sensor provided compelling evidence for its high sensitivity
in detecting subtle, yet physiologically significant, salinity-induced changes in arugula leaf

1 3
179 Page 24 of 40
Precision Agriculture (2026) 27:179
Fig. 10 Spectral response of arugula leaves under salinity stress measured by the optical sensor. Normal-
ized output frequencies are shown for the Red ( a), Green ( b), Blue ( c), Cyan ( d), Magenta ( e), Yellow
(f), and White ( g) spectral channels. Data represent the mean values of 20 independent measurements
(n = 20), capturing changes in leaf reflectance associated with varying salinity treatments

reflectance, even prior to the onset of visible stress symptoms. The baseline measurements
of the sensor using Black and White reference tiles (Fig. 9a-g) confirmed the operational
integrity and spectral fidelity of the system, ensuring that subsequent reflectance measure -
ments were free from systematic bias or inter-channel interference; an essential prerequi -
site for quantitative plant phenotyping (Machwitz et al., 2021). The stability of reflectance
readings under Black and White backgrounds provided internal validation of system per -
formance, ensuring that observed spectral changes stemmed from physiological responses
rather than instrumental artifacts.
Optical response to salinity stress
Sensor output revealed clear spectral signatures across all measured channels, corresponding
to NaCl-induced physiological perturbations (Fig. 10a–g). Significant shifts in Red, Green,
Blue, Cyan, Magenta, Yellow, and White channels’ reflectance were observed, despite the
absence of visible leaf discoloration, highlighting the superior diagnostic sensitivity of opti-
cal sensors over conventional visual assessments.
Among all channels, the Green, Blue, and Cyan spectral channels exhibited the most
pronounced and statistically significant changes with increasing salinity. Specifically, reflec-

1 3
Page 25 of 40 179
Precision Agriculture (2026) 27:179
Table 4 Vegetation indices derived from optical sensor data across salinity treatments
NaCl (mM) ExG ExR NGRDI Red/Green White/RGB White/Green
0 12.0 c −10.6 a 0.66 a 0.20 b 0.091 b 0.20 b
40 14.5 b −12.8 b 0.64 b 0.22 a 0.097 a 0.21 a
80 17.4 a −15.9 c 0.64 b 0.22 a 0.094 ab 0.21 ab
Indices were calculated to quantify vegetation responses under varying salinity levels.
Data are presented as means (n = 4). Statistical differences among treatments were assessed using one-way
analysis of variance (ANOVA), followed by Tukey’s Honestly Significant Difference (HSD) post-hoc test
at a significance level of α = 0.05. Distinct lowercase letters indicate statistically significant differences
between means
tance in the Green and Blue channels increased by 30–50% under moderate salinity (40 mM
NaCl) and exceeded 80–100% under severe salinity (80 mM NaCl), compared to control
leaves. Such reflectance enhancement is a well-established optical indicator of declining
chlorophyll content, pigment degradation, and altered internal leaf architecture associated
with salt-induced osmotic and oxidative stress (Cardenas-Gallegos et al., 2025; Machwitz
et al., 2021; Parida & Das, 2005).
In parallel, Cyan and composite White reflectance increased significantly under salinity
stress, corroborating the phenomenon of global disruptions in leaf optical properties and
surface scattering behavior. These findings are consistent with stress-induced modifications
in mesophyll structure, vacuolar anthocyanin accumulation, or changes in water content, all
of which affect leaf reflectance profiles (de Carvalho Oliveira et al., 2022).
Considering the results of vegetation indices ( Table 4), the ExG increased by 45% rela -
tive to the control at 80 mM NaCl, indicating a marked intensification of Green reflectance.
This increase in Green reflectance is consistent with stress-induced optical adjustments,
potentially arising from reduced chlorophyll degradation in younger tissues or structural
changes in the mesophyll that increase green light scattering. Conversely, the ExR declined
sharply by 49.72%, reflecting diminished red reflectance that may be attributable to reduced
anthocyanin content or altered chlorophyll absorption within the red spectral domain. Nar -
rowing of blue and red reflectance ranges has previously been associated with shifts in pig-
ment absorption (Carter, 1993) and chlorophyll content under stress conditions (Gitelson et
al., 2003).
In the present study, the Red/Green ratio increased by 10% and the White/Green ratio by
5%, suggesting subtle yet quantifiable modifications in the balance between chromatic and
achromatic components of reflected light. Such changes likely arise from a combination of
pigment compositional shifts and altered surface optical properties under osmotic stress.
Comparable early stress detection patterns in arugula have been reported using single-image
normalized difference vegetation index (SI-NDVI), which exploits Red–Green spectral con-
trasts alongside NIR (Beisel et al., 2018). In contrast, broadband indices such as NGRDI
and White/RGB ratio exhibited negligible variation, indicating lower sensitivity to pigment
degradation at the tested stress intensity.
Collectively, these results demonstrate the potential of the low-cost TCS3200 RGB
optical sensor for detecting salinity-associated changes in arugula under the experimen -
tal conditions evaluated, supporting its possible application in non-destructive crop stress
assessment. The capacity to resolve specific spectral signatures, characterized here by
enhanced green reflectance and reduced red reflectance, prior to visible symptom develop -
ment, supports its utility in proactive crop stress management, phenotyping, and saline cul-

1 3
179 Page 26 of 40
Precision Agriculture (2026) 27:179
tivation optimization. Such spectral patterns align with evidence that NaCl stress accelerates
chlorophyll–protein complex disassembly while altering carotenoid stability, thereby recon-
figuring leaf optical behavior. The clear discrimination between treatments in these indices
further demonstrates the robustness of RGB-derived metrics as rapid, non-destructive indi-
cators of stress-induced pigment dynamics in leafy vegetables.
Multivariate analysis
Principal Component Analysis (PCA) results
The PCA revealed clear patterns of multivariate differentiation among treatments. Particu -
larly, the PC1 accounted for the majority of variance (69.11%) providing the strongest dis-
crimination between control and 80 mM NaCl plants (Fig. 10). The high positive loadings
of Na, Na⁺/K⁺, Na⁺/Ca2⁺, and RGB channels (particularly White and Red) on PC1 indicate
that ionic accumulation and spectral reflectance shifts are primary drivers of variation. Con-
versely, negative PC1 loadings for FW, LA, K, Ca, Mg, and ExR reveal the physiological
cost of ionic toxicity, reduced biomass and mineral depletion, mirrored by specific reflec -
tance signatures (Fig. S2).
The dispersion of 40 mM NaCl samples along PC1–PC2 suggests heterogeneous
responses at intermediate stress, possibly reflecting genotype-specific tolerance thresholds
or micro-environmental variation in ion uptake rates (Fig. 11). PC2, with 10.29% variance,
was primarily influenced by ΔE metrics and chromaticity indices, representing variation
in pigment degradation independent of biomass loss (Fig. S2). This orthogonal separation
implies that spectral deterioration can occur even before severe growth suppression, a pat -
tern consistent with early chloroplast dysfunction preceding visible wilting in glycophyte
species (Melo et al., 2017).
Canonical correlation analysis (CCA) results
The CCA revealed a tightly coupled multivariate relationship between physiological and
optical traits, with a clear progression from negative to positive canonical scores across the
0, 40, and 80 mM NaCl treatments (Fig. 12). Negative loadings of FW, LA, K, Ca, and Mg
reflect the depletion of essential macrominerals under osmotic and ionic stress, while posi -
tive loadings for Na, their ratios, and microminerals (Fe, Cu, Zn, B) point to competitive ion
displacement and altered mineral partitioning (Fig. S3).
The parallel response of RGB variables, particularly strong positive loadings for Blue,
Magenta, Yellow, and White channels, suggests that specific spectral shifts are mechanisti-
cally linked to ionic toxicity (Fig. S3). Elevated reflectance in the blue and magenta ranges
is likely associated with loss of chlorophyll absorption capacity, while increased yellow
reflectance may arise from carotenoid prominence during chlorophyll decline (Navarro et
al., 2022). The inverse loadings of ExR and b* further indicate that Red-excess signals are
diminished in parallel with pigment breakdown, whereas chromaticity b* tracks the yel -
lowing process. The near-symmetry of X and Y loadings confirms the robustness of the
optical–physiological linkage, supporting the view that RGB-based sensing can function as
a reliable surrogate for destructive physiological measurements.

1 3
Page 27 of 40 179
Precision Agriculture (2026) 27:179
Fig. 11 Principal Component Analysis (PCA) biplot showing sample distribution along the first two prin-
cipal components (PC1 and PC2), derived from physiological and optical sensor-based variables. The
plot illustrates clustering patterns and variance structure among treatments. Colored circular markers
represent samples subjected to different salinity treatments (0, 40, and 80 mM NaCl), while colored
elliptical overlays indicate variable groupings based on their multivariate associations within the PCA
space. This visualization facilitates interpretation of treatment effects and variable interrelationships in
reduced dimensionality

Correlation network and trait connectivity results
Correlation network mapping resolved two sharply antagonistic clusters (Fig. 13). The
first, growth–integrity, comprised FW, LA; K, Ca, and Mg concentrations, all exhibiting
exceptionally strong intercorrelations (r = 0.91–0.99) and a consistent decline with increas-
ing salinity. The second, sodium–spectral signatures, encompassed Na accumulation, Na-
derived ionic ratios, RGB reflectance channels, and the ExG, which were tightly interrelated
but negatively correlated with growth traits. This inverse association underscores the central
trade-off under salt stress: escalating ionic toxicity and pigment degradation coincide with
mineral depletion and biomass suppression.
Conspicuously, FW and LA displayed strong negative correlations with Fe and Cu,
implying a salinity-induced redistribution of microminerals to leaves, potentially mediated
by altered transpiration streams or competitive ion uptake at the root–soil interface. Network
topology further highlighted the role of macromineral homeostasis in pigment stability: K,
Ca, and Mg formed a cohesive, positively linked cluster, inversely aligned with Na metrics

1 3
179 Page 28 of 40
Precision Agriculture (2026) 27:179
Fig. 12 Canonical Correlation Analysis (CCA) illustrating the multivariate relationships among all stud-
ied parameters. The biplot reveals the strength and direction of canonical associations between physi -
ological and sensor-derived variables. Distinct colored circles represent samples subjected to different
salinity treatments: 0, 40, and 80 mM NaCl, enabling visual differentiation of treatment effects within
the canonical space

yet positively associated with ExR. Conversely, sodium-related traits clustered densely with
RGB and ExG variables, reflecting the pronounced optical footprint of ionic stress. The
multivariate structure of trait associations, including strong correlations among growth and
mineral parameters, is presented in the correlation matrix and trait network (Fig. S4).
Hierarchical clustering (Ward’s linkage, Euclidean distance) portrayed three coherent
biological domains: (i) mineral assimilation–green reflectance (N, Mg, NGRDI, R/G ratio),
representing nitrogen-dependent maintenance of chlorophyll and leaf greenness; (ii) pig -
ment integrity (L*, a*, b*, chroma, hue, SI, ΔE series), encapsulating the optical signature
of chlorophyll and accessory pigment loss; and (iii) integrated physiological–optical stress
responses, combining growth suppression, ionic imbalance, and broad-spectrum reflectance
shifts (Fig. 14). These clusters reflect the overall plant productivity and nutritional status,
which are typically suppressed under salinity due to ionic toxicity and osmotic stress (Rah-
man et al., 2021).
Colorimetric variables clustered tightly with ΔE metrics, indicating that changes in pig -
ment pools are directly traceable in perceptual color space. The close grouping of Na-related
variables with RGB traits corroborates their shared response pathway—chlorophyll break -
down, altered carotenoid-to-chlorophyll ratios, and structural changes in mesophyll tissues
driving shifts in visible reflectance. The proximity of 40 mM and 80 mM NaCl treatments
within the dendrogram suggest that major optical and physiological disruptions occur rap -

1 3
Page 29 of 40 179
Precision Agriculture (2026) 27:179
Fig.  13 Correlation heatmap depicting the relationships between optical sensor-derived variables and
plant physiological parameters. Data were normalized prior to clustering, and hierarchical grouping of
metabolites was performed using the complete linkage method with Euclidean distance. The matrix high-
lights distinct clusters of highly correlated variables (red), including optical sensor outputs, growth and
macromineral integrity that comprises fresh weight (FW), leaf area (LA), and concentrations of potas -
sium (K), calcium (Ca), and magnesium (Mg), all exhibiting strong intercorrelations (r = 0.91–0.99) and a
consistent decline in response to increasing salinity levels (0, 40, and 80 mM NaCl)

idly beyond a threshold, with higher salt intensities amplifying rather than fundamentally
altering the stress profile. Together with the correlation network results, the clustering pat -
terns affirm that Na⁺-driven disruption of K⁺ and Ca 2⁺ homeostasis impairs osmotic adjust-
ment, stomatal function, and thylakoid membrane stability mechanisms that collectively
degrade pigment pools and modify spectral properties. These findings align with previous
studies that emphasize the multidimensional nature of plant stress responses and the value
of integrative phenotyping approaches (Costa et al., 2024; Rahman et al., 2021).
In summary, the results of this study provide compelling evidence that RGB optical sen-
sors are not only efficient and reliable but also biologically informative proxies for salinity

1 3
179 Page 30 of 40
Precision Agriculture (2026) 27:179
Fig. 14 Hierarchical clustering of physiological and spectral traits using Ward’s linkage and Euclidean
distance. The analysis resolves three biologically distinct domains: mineral assimilation–green reflec -
tance (nitrogen (N), magnesium (Mg), normalized Green–Red difference index (NGRDI), and Red/Green
(R/G) ratio, pigment integrity (L*, a*, b*, Chroma, Hue, saturation index (SI), ΔE series), and integrated
stress responses encompassing growth suppression, ionic imbalance, and broad-spectrum reflectance
shifts. These clusters reflect coordinated physiological and optical signatures of salinity-induced stress

stress detection. The convergence of findings from PCA, CCA, correlation network analy -
sis, and hierarchical clustering, along with the robust associations between RGB-derived
parameters and physiological indicators such as ion accumulation, mineral depletion, and
water status, emphasize their capacity to serve as precise, non-destructive proxies for plant
stress assessment. By capturing these complex physiological responses through straightfor-
ward reflectance measurements, RGB sensors emerge as indispensable tools in the advanc-
ing field of digital phenotyping. These results reinforce the growing consensus that optical
sensing technologies have the potential to revolutionize stress diagnostics in precision agri-
culture (Mutanga et al., 2023; Stefanov et al., 2022; Yang et al., 2021).
DIABLO an alysis results
The DIABLO analysis proved equally adept at resolving the complex, multivariate relation-
ships between physiological parameters, CIELAB colorimetry, and RGB sensor reflectance
under increasing NaCl stress in arugula. Despite the absence of transcriptomic or metabolo-
mic datasets, the model capitalized on its capacity to identify latent structures shared across
heterogeneous data blocks, thereby revealing convergent signatures of stress. Cross-valida-
tion yielded a mean error rate of 0.083 for the first component, with zero error for the sec -
ond component across all bootstrap iterations, indicating substantial model generalizability
and stability. Such low error rates suggest that the selected latent dimensions captured the
dominant discriminatory variance, in line with previous findings where DIABLO revealed
interconnections between the different omics data layers for evergreen oak subjected to
drought stress (Guerrero-Sánchez et al., 2022).

1 3
Page 31 of 40 179
Precision Agriculture (2026) 27:179
Fig.15 Feature contribution and stability across integrated datasets in DIABLO (Data Integration Analy-
sis for Biomarker discovery using Latent cOmponents). ( a) Contribution of individual features to the
DIABLO model. Feature stability across modalities is shown for (b) physiological traits, (c) RGB optical
sensor-derived data, and ( d) CIELAB colorimetry parameters. Panels highlight the robustness and con -
sistency of selected features within each dataset

Feature tuning initially retained five physiological traits, ten CIELAB indices, and three
RGB variables. Subsequent optimization to three features per block maintained classifica -
tion accuracy, implying that a compact subset of traits is sufficient to capture the biological
divergence between treatments. This parsimony is critical for sensor-based phenotyping, as
it suggests that full-spectrum measurement may be unnecessary when a reduced panel of
highly informative indices can reliably differentiate stress severity.
Component score distributions revealed a clear treatment-dependent gradient across
all data blocks (Fig. 15a). Control plants (0 mM NaCl) consistently occupied the nega -
tive extreme of component 1, high-salinity plants (80 mM NaCl) clustered at the positive
extreme, and intermediate salinity (40 mM NaCl) samples lay near the origin. This consis -
tent directional shift, observed across the physiological (mean scores –2.67, 0.30, 2.36),
CIELAB (–1.81, 0.50, 1.32), and RGB (–1.92, 0.02, 1.89) blocks, reflects a dose-dependent
alteration in both biochemical and optical profiles. Physiologically, this gradient likely
arises from progressive Na⁺ accumulation and its associated ionic imbalances, as evidenced
by shifts in Na +/K+ and Na +/Mg2+ ratios, coupled with pigment loss and structural leaf
changes that manifest in both CIELAB and RGB metrics. Such integrated responses are
well-documented in glycophytes under salinity, where ionic toxicity precipitates chloro -

1 3
179 Page 32 of 40
Precision Agriculture (2026) 27:179
phyll degradation, reduces carotenoid stability, and alters leaf optical properties (Navarro
et al., 2022).
Feature stability analysis (Fig. 15b-d) further clarified the biological foundation of the
DIABLO model. Within the physiological block, Na exhibited perfect stability (1.00), fol -
lowed closely by K, Na +/K+, Na+/Mg2+, and Na+/Ca2+ (≥ 0.95), corroborating their role as
primary indicators of ionic imbalance and mineral antagonism under saline conditions (Sha-
bala & Cuin, 2008). In the CIELAB block, b* (0.95) and Chroma(0.88) demonstrated the
greatest stability, both directly reflecting pigment concentration and color vividness. Hue,
with moderate stability (0.52), likely reflects secondary pigment shifts, while ΔE and a*
exhibited lower stability, suggesting they may be more sensitive to noise or plant-to-plant
variation at moderate stress levels. RGB stability patterns revealed White, Green, and Red
as the most reliable channels, consistent with their high correlation to chlorophyll status and
canopy light reflectance in stress phenotyping (Gitelson et al., 2009; Pastor-Guzman et al.,
2015).
The DIABLO circos-plot (Fig. 16) offered a high-resolution view of inter-block relation-
ships at a correlation threshold of r ≥ 0.7. Dense linkages between Na-related physiological
traits and RGB White, Green, and Red channels confirm the shared variance captured by
component 1, reflecting the coupling between ionic stress and changes in visible light reflec-
tance. Simultaneously, strong correlations between b*, chroma, and hue in the CIELAB
block and the same RGB channels reveal the spectral consequences of pigment degrada -
tion, as decreased chlorophyll and carotenoid concentrations shift reflectance toward longer
wavelengths. Such coordinated changes are consistent with findings in maize and spinach,
where salt-induced pigment loss leads to measurable spectral shifts in the visible domain
Liu (Liu et al., 2025; Vennam et al., 2024).
The convergence of high-stability features across blocks supports a mechanistic model
wherein Na⁺ toxicity disrupts K⁺, Ca 2⁺, and Mg2⁺ homeostasis, impairing enzymatic func -
tion and photosynthetic machinery. This ionic imbalance accelerates pigment degradation,
mediated in part by oxidative stress, leading to measurable declines in chroma, and shifts
in spectral reflectance (Vennam et al., 2024). The RGB sensor’s ability to capture these
changes in White, Green, and Red channels, probably reflects the dominance of chlorophyll
absorption in the red and blue wavelengths, combined with increased scattering in the green
and near-white range as chlorophyll content declines. A comparative representation of DIA-
BLO analysis components is provided in Fig. 17.
Collectively, the DIABLO results demonstrate that integrating physiological and optical
datasets yields a coherent, biologically grounded picture of salinity stress progression. The
stability of a minimal set of variables (Na, Na +/K+, Na+/Mg2+, b*, Chroma, White, Green,
and RED) suggests that compact, targeted sensor arrays could reliably phenotype salt stress
in leafy crops without full-spectrum imaging. This has direct implications for precision
agriculture, where rapid, non-destructive, and computationally efficient detection methods
are essential for timely intervention and resource optimization.
Limitations
The present study evaluated the ability of the TCS3200 RGB sensor to detect salinity-
associated optical changes in arugula under controlled greenhouse conditions. Accordingly,
the findings are limited to the crop, cultivar, growth stage, salinity treatments, and mea -

1 3
Page 33 of 40 179
Precision Agriculture (2026) 27:179
Fig. 16 Circos plot illustrating inter-block correlations among integrated datasets from DIABLO (Data
Integration Analysis for Biomarker discovery using Latent cOmponents) analysis. Inter-block relation -
ships among mean values (n = 4) of physiological traits, CIELAB colorimetry, and RGB sensor-derived
data are visualized using a Circos plot. Edges represent pairwise correlations at a threshold of r ≥ 0.7,
highlighting strong associations across data modalities

surement conditions investigated. The study did not include direct benchmarking against
advanced imaging platforms, large-scale deployment trials, or external validation using
independent datasets. In addition, wireless networking, cloud connectivity, and continuous
real-time monitoring were considered as potential extensions but were not experimentally
implemented. Development of a quantitative predictive model was also beyond the scope
of the present study. Future research should therefore evaluate the sensing approach across
larger and more diverse datasets, crops, developmental stages, and environmental condi -
tions, while also addressing comparative validation, predictive modeling, independent test-
ing, and practical deployment within automated or wireless greenhouse monitoring systems.

1 3
179 Page 34 of 40
Precision Agriculture (2026) 27:179
Fig.  17 Integration of spectral and physiological traits using DIABLO (Data Integration Analysis for
Biomarker discovery using Latent cOmponents) model. Latent components 1 and 2 from DIABLO are
shown for (a) CIELAB color metrics, (b) physiological parameters, and (c) RGB sensor data. Each panel
illustrates sample distribution and clustering patterns, revealing shared and dataset-specific sources of
variation across modalities

Conclusion
The objective of the present study was to propose a low-cost and non-destructive opti -
cal sensor for the early detection of salinity stress in hydroponically grown arugula. The
results demonstrated that salinity stress elicited multifaceted disruptions in physiological
function, ionic homeostasis, and optical properties of arugula leaves. The observed modula-
tion of growth parameters highlights a transient compensatory window at moderate salinity
(40 mM NaCl), beyond which the system succumbs to progressive physiological deteriora-
tion under more severe stress (80 mM NaCl). The mineral profile reinforced this trajectory,
revealing indicative features of salt-induced ionic disequilibrium, characterized by drastic
Na⁺ accumulation, K⁺ and Ca2⁺ depletion, and micromineral profile alterations. These find-
ings align with established models of glycophytic salt stress response but extend them by
providing integrated, quantitative, and temporally resolved physiological insights.

1 3
Page 35 of 40 179
Precision Agriculture (2026) 27:179
Most critically, this work supports the exceptional sensitivity of low-cost, RGB-based
optical sensing for early, non-destructive detection of salinity stress. Despite the absence
of visible symptoms, channel-specific reflectance alterations (particularly White, Green,
and Red spectra) were detectable, reflecting underlying pigment degradation and leaf struc-
tural perturbations. The integration of data integration through DIABLO framework further
amplified the diagnostic resolution, thus validating the feasibility of sensor-driven, auto -
mated stress phenotyping in controlled environments.
Nevertheless, certain limitations merit acknowledgment. While the sensor effectively
captured reflectance-based proxies of stress, its performance across diverse species, devel -
opmental stages, and environmental conditions remains to be rigorously validated. Further-
more, the influence of confounding factors such as light heterogeneity, canopy architecture,
and overlapping abiotic stressors warrants systematic investigation.
Future work should focus on: (i) expanding sensor deployment to field-grown crops
to assess real-world scalability; (ii) refining sensor algorithms to disentangle multi-stress
interactions; and (iii) integrating spectral data with physiological, molecular, and imaging
platforms for holistic, multi-modal plant stress diagnostics. Such advancements will solidify
the role of optical sensing technologies as indispensable tools for precision agriculture, con-
tributing to sustainable crop production under escalating environmental stress and climate
change.
Supplementary Information The online version contains supplementary material available at h t t p s : / / d o i . o r g
/ 1 0 . 1 0 0 7 / s 1 1 1 1 9 - 0 2 6 - 1 0 4 7 6 - w .
Acknowledgements The authors wish to thank Jacob Breon, Radhika Chalam, Scott Di Loreto, Diane Lehr,
Aline Novaski, Rishi Ravichandran, Sanchita Saha, William Spratt, and Timothy Tretter for their invaluable
technical assistance.
Author contributions Ali Ahmad: Conceptualization, Methodology, Investigation, Data curation, Formal
analysis, Visualization, Writing – original draft, and Writing – review & editing. Francisco Javier Diaz
Blasco: Writing – original draft. Sandra Sendra: Supervision, Funding acquisition, Validation, Writing –
review & editing. Jaime Lloret: Supervision, Funding acquisition, Validation, Project administration, Writing
– review & editing. Francesco Di Gioia: Conceptualization, Methodology, Supervision, Validation, Funding
acquisition, Writing – review & editing.
Funding This work was supported by the Pennsylvania Department of Agriculture Specialty Crop Block
Grant No. C940001528, the USDA-ARS Penn State Cooperative Agreement: 58‐6034‐3‐016, by the Grant
PRE2021-100809 funded by MICIU/AEI/10.13039/501100011033 and by European Social Fund Plus
(ESF +). F. Di Gioia’s contribution was supported by the USDA National Institute of Food and Agriculture
and Hatch Appropriations under Project No. PEN05002 and Accession No. 7007517. This work was also
funded in part by the PRIMA Programme supported by the European Union, project PA VER, and it is sup-
ported by MCIN/AEI/10.13039/501100011033.
Data availability Data supporting the findings of the study is available within the article or its supplementary
materials.
Declarations
Competing interests The authors declare no competing interests.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License,
which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as
you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons

1 3
179 Page 36 of 40
Precision Agriculture (2026) 27:179
licence, and indicate if changes were made. The images or other third party material in this article are
included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material.
If material is not included in the article’s Creative Commons licence and your intended use is not permitted
by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the
copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
References
Ahmad, A., Diaz-Blasco, F. J., Zaragoza-Esquerdo, M., Sendra, S., Parra, L., Viciano-Tudela, S., Lloret, J.,
Chaloupková, V ., Bados, R., & Pascual, L. S. E. (2024). LoRaWAN-based Network for Harvest Time
Estimation in Cistus ladanifer. 2024 11th International Conference on Internet of Things: Systems,
Management and Security (IOTSMS),
Ahmad, A., Liew, A. X., Venturini, F., Kalogeras, A., Candiani, A., Di Benedetto, G., Ajibola, S., Cartujo, P.,
Romero, P., & Lykoudi, A. (2024b). AI can empower agriculture for global food security: Challenges
and prospects in developing nations. Frontiers in Artificial Intelligence,7, Article 1328530.
Ahmad, A., Zaragoza-Esquerdo, M., Diaz, F. J., Sendra, S., & Lloret, J. (2026). Galvanic cell enables copper
reuse and energy harvesting for sustainable plant micronutrient delivery with potential wireless sensor
network applications. Electronics,15(14), 2992.
Ashraf, M., & Harris, P. J. (2013). Photosynthesis under stressful environments: An overview. Photosyn-
thetica,51, 163–190.
Beisel, N. S., Callaham, J. B., Sng, N. J., Taylor, D. J., Paul, A. L., & Ferl, R. J. (2018). Utilization of single‐
image normalized difference vegetation index (SI‐NDVI) for early plant stress detection. Applications
in Plant Sciences,6(10), Article e01186.
Brambilla, M., Romano, E., Buccheri, M., Cutini, M., Toscano, P., Cacini, S., Massa, D., Ferri, S., Monarca,
D., & Fedrizzi, M. (2021). Application of a low-cost RGB sensor to detect basil (Ocimum basilicum L.)
nutritional status at pilot scale level. Precision Agriculture,22, 734–753.
Cardenas-Gallegos, J. S., Lacerda, L. N., Severns, P. M., Peduzzi, A., Klimeš, P., & Ferrarezi, R. S. (2025).
Advancing biomass estimation in hydroponic lettuce using RGB-depth imaging and morphometric
descriptors with machine learning. Computers and Electronics in Agriculture,234, Article 110299.
Carter, G. A. (1993). Responses of leaf spectral reflectance to plant stress. American Journal of Botany,80(3),
239–243.
Cataldo, D. A., Maroon, M., Schrader, L. E., & Youngs, V . L. (1975). Rapid colorimetric determination of
nitrate in plant tissue by nitration of salicylic acid. Communications in Soil Science and Plant Analy -
sis,6(1), 71–80.
Cavalcante, J. S. J., Ferreira Neto, M., Peixoto, T. D. C., Pereira, K. T. O., Umbelino, B. F., Rodrigues Filho,
R. A., Moreira, R. C. L., Fernandes, P. D., da Silva Dias, N., & da Silva Ribeiro, J. E. (2025). Produc -
tion, Growth, Photosynthetic Response, and Ionic Responses of Arugula Cultivars Under Salt Stress
Elicitors. Journal of Plant Growth Regulation, 1–17.
Chakraborty, K., Mondal, S., Bhaduri, D., Mohanty, A., & Paul, A. (2022). Interplay between sodium and
chloride decides the plant’s fate under salt and drought stress conditions. Plant nutrition and food secu-
rity in the era of climate change (pp. 271–314). Elsevier.
Corti, E., Falsini, S., Schiff, S., Tani, C., Gonnelli, C., & Papini, A. (2023). Saline stress impairs lipid storage
mobilization during germination in eruca sativa. Plants,12(2), 366.
Costa, L. P., Mendonça, V ., d. Oliveira, Fd. A., Pinto, F. F., d. Carlos, KGd. S., de Oliveira, M. K., d. Medeiros,
J. F., & d. Dias, Vd. L. (2024). Salt stress and calcium nitrate in arugula in soilless cultivation using
substrate. Revista Brasileira De Engenharia Agrícola e Ambiental,29, Article e285670.
Cruz, L. R., Fernandes, Â., Di Gioia, F., Petropoulos, S. A., Polyzos, N., Dias, M. I., Pinela, J., Kostić, M.,
Soković, M. D., & Ferreira, I. C. (2020). The effect of nitrogen input on chemical profile and bioactive
properties of green-and red-colored basil cultivars. Antioxidants,9(11), Article 1036.
Darlan, D., Ajani, O. S., An, J. W., Bae, N. Y ., Lee, B., Park, T., & Mallipeddi, R. (2025). SmartBerry for
AI-based growth stage classification and precision nutrition management in strawberry cultivation. Sci-
entific Reports,15(1), Article 14019.
de Carvalho Oliveira, G., Machado, C. C. S., Inácio, D. K., da Silveira Petruci, J. F., & Silva, S. G. (2022).
RGB color sensor for colorimetric determinations: Evaluation and quantitative analysis of colored liq -
uid samples. Talanta,241, Article 123244.
Deinlein, U., Stephan, A. B., Horie, T., Luo, W., Xu, G., & Schroeder, J. I. (2014). Plant salt-tolerance mecha-
nisms. Trends in Plant Science,19(6), 371–379.

1 3
Page 37 of 40 179
Precision Agriculture (2026) 27:179
Di Gioia, F., Avato, P., Serio, F., & Argentieri, M. P. (2018a). Glucosinolate profile of Eruca sativa, Diplotaxis
tenuifolia and Diplotaxis erucoides grown in soil and soilless systems. Journal of Food Composition
and Analysis,69, 197–204.
Di Gioia, F., Rosskopf, E. N., Leonardi, C., & Giuffrida, F. (2018b). Effects of application timing of saline
irrigation water on broccoli production and quality. Agricultural Water Management,203, 97–104.
Di Gioia, F., Signore, A., Serio, F., & Santamaria, P. (2013). Grafting improves tomato salinity tolerance
through sodium partitioning within the shoot. HortScience,48(7), 855–862.
Diaz, F. J., Ahmad, A., Parra, L., Sendra, S., & Lloret, J. (2024). Low-cost optical sensors for soil composi-
tion monitoring. Sensors,24(4), 1140.
Faehn, C., Konert, G., Keinänen, M., Karppinen, K., & Krause, K. (2024). Advancing hyperspectral imaging
techniques for root systems: A new pipeline for macro-and microscale image acquisition and classifica-
tion. Plant Methods,20(1), 171.
Fallahi, H.-R., Fad

论文原文与出版信息

Precision Agriculture
查看论文原文

继续探索