Electrical Impedance Spectroscopy Reveals Physiological Acclimation in Apple Rootstocks During Recurrent Water Stress Episodes

Abstract

Waterlogging and drought have become major challenges in many regions worldwide. Under water stress, plants exhibit a range of physiological and electrical responses, including changes measurable by electrical impedance spectroscopy (EIS). Monitoring these parameters can provide valuable insights into plant growth status under adverse conditions. This study investigated changes in relative chlorophyll content (SPAD), maximum photochemical efficiency (Fv/Fm), relative water content (RWC), non-structural carbohydrates (NSC), and EIS parameters in apple rootstocks subjected to different water stress treatments. Results indicated that all physiological indicators, except NSC, showed a declining trend under two water stress episodes. Critically, the initial water stress episode elicited significantly greater physiological disruption than its subsequent counterpart. This suggests that plants developed a degree of physiological adaptation—such as osmotic adjustment and enhanced antioxidant activity—reducing their sensitivity to subsequent stress. Correlation analysis revealed that high-frequency resistivity (r) and intracellular resistivity (r_i) were strongly associated with key physiological parameters. Thus, r and r_i may serve as effective indicators for assessing plant water stress status. Furthermore, classification algorithms—Fuzzy K-Nearest Neighbors (FKNN) and sparse Linear Discriminant Analysis (sLDA)—were applied to distinguish water status in apple rootstocks, achieving high classification accuracy. These findings provide a theoretical basis for improved water management in apple cultivation.

1. Introduction

Waterlogging and drought have emerged as critical agricultural constraints globally [1], primarily driven by precipitation anomalies, inadequate land drainage, suboptimal irrigation practices, and intensifying climate change impacts [2,3]. Under these compounded stressors, plant growth, and development are severely compromised as key physiological processes-including photosynthetic efficiency and nutrient metabolism-become significantly impaired [4]. Consequently, accurate monitoring of plant physiological status under water stress is imperative for optimizing agricultural water management and enhancing stress resilience in cultivated fruit tree systems [5]. The plant status under water stress could be observed through their appearance, which is, however, criticized as the plants could already suffer from irreversible damage [6]. The other method is through measuring their physiological indicators, such as water circulation, transpiration, and the chlorophyll content of leaves, which is feasible to detect plant status under water stress at the early stage before any observable damage. Under water stress, the chlorophyll content in plants is affected, and the SPAD value can be used to monitor the chlorophyll content in plant leaves, which reflects the plant’s photosynthetic activity and water status [7]. As water stress intensifies, the maximum photochemical efficiency (Fv/Fm) value decreases, indicating a decline in the quantum efficiency of Photosystem II, with this change often occurring before any visible symptoms of damage appear [8]. Furthermore, under drought stress, the relative water content (RWC) in plants significantly decreases, and changes in RWC are closely related to a soil water deficit. By monitoring these parameters, early warnings of water stress can be issued before severe damage occurs [9]. Additionally, under water stress, the levels of sucrose and starch in non-structural carbohydrates decrease, while glucose and fructose levels increase, and these changes typically occur before any noticeable damage to indicators like leaf RWC is observed [10]. However, these physiological indicators cannot be measured in situ, which is, therefore, time-consuming, and measuring these parameters could damage the plants [6,11]. Near-infrared spectroscopy (NIR) can provide fast and in-situ measurements [12]. Spectral reflectance from hyperspectral images can also be used to predict leaf water content [13]. However, these are the costly approaches that limits its application in the field.

Compared to manual physiological measurements (which are destructive) and spectral techniques like NIR (which are costly), EIS provides a fast, non-destructive, and inexpensive alternative for monitoring plant physiological status under water stress [14,15]. When alternating current (AC) is applied to plants, the flow of current varies with the frequency of the AC stimulus. At low frequencies, the current flows mainly in the extracellular fluid, while at high frequencies, the current flows mainly inside the cells [16]. This frequency-dependent current can be used as an indicator of plant health and plant response to water stress. Therefore, under the action of sinusoidal excitation of a certain frequency, the parameters of the electrical impedance spectroscopy can be used to visualize the physiological and biochemical changes in the cell. The working principle of EIS is to apply small harmless electrical signals to the plant and measure the impedances within and outside the cells at different frequencies, which can more accurately reveal the plant’s response to environmental stress, such as drought and waterlogging [12,16,17,18]. As water stress could affect their physiological states, which is related to their electrical properties, the impedance results could reflect water content [19]. In addition, studies have demonstrated that EIS is widely employed to assess the responses of plant organs and tissues to various environmental stresses, including freeze-thaw injury, cold acclimation, osmotic stress, water stress, and nutrient deficiency. This technique has led to significant advancements in the field of plant stress resistance research [16]. Blanchard et al. [20] found that after inoculation with Fusarium oxysporum, the EIS of elm trunks exhibited significant changes, with resistance values able to predict disease onset 2 to 12 days before visible symptoms, thus providing a method for disease prediction in seedlings based on resistance measurements. Similarly, Wilner et al. [21] demonstrated that impedance measurements of apple roots effectively distinguished diseased plants and detected disease presence before the appearance of tomato tumor symptoms. Moreover, the AC impedance values of alfalfa plants were found to be reliable indicators for identifying plants susceptible to root-knot nematode infection. Davis et al. [22] reported significant changes in the resistance of fir trees following whitefly infestation. Levitt et al. [3] observed a reduction in the resistance of woody plants under low-temperature stress. Additionally, Serrano-Finetti et al. [12] utilized EIS to non-invasively monitor leaf water content, enabling rapid and accurate assessments of plant water status. Collectively, these studies highlight the broad potential of electrical impedance technology in evaluating plant stress damage. Overall, these studies highlight the great potential of EIS technology in assessing plant stress damage. EIS stands out for its ease of use, affordability, and minimal invasiveness, making it an ideal choice for field applications.

Traditionally, the classification of plant status under water stress based on experimental outputs often involves manual intervention, which could be subjective, inconsistent, and time- and cost-intensive. Besides, plant status under water stress can change over time, whereas tracking this dynamic change is difficult with manual assessment [23]. In these circumstances, Machine Learning (ML), such as sLDA [24,25,26] and FKNN [27,28,29], has been increasingly used for fast, real-time, and accurate plant status classification [16]. EIS allows early detection and real-time monitoring of the plant status of full plants including roots, making it useful in local irrigation management. Its combination with ML allows efficient and reasonable use of EIS data, which represents one of the most important research directions in EIS research [16]. However, few research in this field could be found.

This study investigated the variations in physiological parameters of apple rootstock leaves and Electrical Impedance Spectroscopy parameters under varying water stress treatments. The aim was to identify an optimal EIS parameter that could effectively characterize physiological changes and reflect the water status of apple rootstocks. Additionally, this parameter would serve as a basis for machine learning-based evaluation of apple rootstock water status, offering crucial technical support for future water management strategies in apple cultivation.

2. Materials and Method

2.1. Material

For this experiment, the second-year apple dwarfing plant rootstock “M9T337” with a plant height of 50 cm, and a stem thickness of 10 mm without branches, was selected from Nanzhai Town, Qianyang County, Baoji City, Shanxi Province (34°28′12″ N, 107°32′24″ E). We obtained the permission of Qianyang County Jingqian Apple Planting Company, and the collected plant samples were formally identified by Mr. Wenjun Li. These rootstocks were planted in the experimental garden of the East Campus of Hebei Agricultural University (38°51′25″ N, 115°29′13″ E), China, on 22 March 2019, with the seedling substrate containing “M9T337” as seedlings. On March 22, seedlings were transplanted into plastic pots (18.5 cm diameter × 22 cm height) containing a peat-based substrate (65% organic matter; sandy loam equivalent texture; pH 7.0; peat: perlite: vermiculite = 2:1:1 v/v) at the Experimental Garden of the East Campus. Immediately after transplanting, pots were watered to field capacity.

During the pre-stress establishment period (23 March to 8 July), volumetric soil water content was maintained at 75–85% of maximum water holding capacity through daily monitoring using a TDR100 soil moisture sensor (Spectrum Technologies, Inc., Plainfield, IL, USA). Irrigation was applied whenever soil moisture dropped below the target threshold.

2.2. Experiment Design

Seedlings were subjected to treatments from 9 July to 19 August 2019. A total of 360 (4 × 3 × 3 × 10 = 360) seedlings were required for 4 treatments, 3 replications, 3 parallel measurements, and 10 times sampling. The experiment was conducted in a randomized block design with four treatments, namely, the waterlogging treatment (WL, water level is even with the surface of the soil in the pots), the flooding treatment (FD, water level is 2–3 cm above the surface of the soil in the pots), the drought treatment (DR, the soil moisture content ranges from 25% to 35% of the soil’s maximum field capacity), and the control treatment CK (soil volumetric water content 75–85% of the maximum field water holding capacity of the soil). In addition, 3 replications represent three samples for each treatment, 3 parallel measurements mean that each sample has three parallel identical samples for parallel measurements to reduce experimental error, 10 times sampling represents sampling once every two days from day 1 to day 7 and from day 29 to day 35, once on day 14 and once on day 42. The current solar radiation, air temperature, and air humidity in the experimental environment were also initiated at the beginning of the experiment, as shown in Figure 1.

For WL and FD, two pools of 600 cm × 115 cm × 28 cm (L × W × H) were constructed in the experimental area. For DR, a rain shelter of 600 cm × 120 cm × 150 cm (L × W × H) was constructed. CK area is a natural open air without cover on top. To simulate the historical precipitation distribution, two 7-day water stress episodes (Wt) were imposed on the seedlings: the initial water stress episode from 9 July to 15 July 2019, and the subsequent water stress episode from 6 August to 12 August 2019. The rest of the time was the daily maintenance period (Dm). All treatments were watered slowly on July 1 until water seeped out of the bottom of the pots, after which watering was stopped for the DR area while the rest were watered normally.

Two water stress episodes were conducted in this experiment to increase results applicability. This is because of the frequent occurrence of extreme weather; the probability of plants being subjected to the subsequent episode increases in nature. As the effects of the subsequent episode increases on plants are different from the initial, the study of the physiological state of plants subjected to the subsequent episode increases is crucial for water management in agriculture and for improving the resistance of cultivated trees [5].

The initial and subsequent water stress episodes and sampling strategies are presented in Figure 2. Note that watering was performed once a day at 9:00 a.m. for the CK and DR; the WL and FD groups were watered only when the soil water content was lower than 75% and the watering time is the same as CK and DR.

2.3. Indicators Measurement and Machine Learning Classification

2.3.1. Indicator Measurement

During each sampling, three seedlings were randomly selected from three replicates of three blocks, and one leaf was randomly taken from each selected seedling.

At the end of the initial water stress episode (day 7) and at the end of the subsequent episode (day 35), 10 leaves were randomly selected from each seedling for the measurement. Note that the measurement was performed on the leaf margins and the middle of the leaf veins, while leaf veins were avoided.

The relative chlorophyll content was first measured using a hand-held Soil Plant Analysis Development meter (SPAD-502Plus, Konica Minolta Sensing Singapore Pte Ltd). The samples were fixed with dark treatment clips, dark-adapted for 20 min, and measured with a portable fluorometer (HandyPEA, Hansatech, King’s Lynn, UK) to determine the basic fluorescence value of the samples (F₀), the maximum fluorescence value (Fm) (after excitation by a saturating light pulse of 0.8 s 3000 mmol · m⁻² · s⁻¹), and the maximum photochemical efficiency of photosystem II was determined using the ratio Fv/Fm (Fv is variable fluorescence), where Fv = Fm − F₀.

After the above parameters were measured, leaves were picked up from the plant for Relative water content (RWC) measurement. The removed leaves were first washed with tap water to remove impurities, then with deionized water 3 times. After drying the surface water, a 1.5 cm × 0.5 cm hole punch was used to take part of the leaf (not veins) for EIS measurement. A sample of 5 g leaves was selected from the remaining leaves for water content measurement according to Equation (1). The sample was first weighed and recorded as fresh weight. It was then placed in a closed Petri dish and soaked in water. After 24 h, its weight was measured and recorded as saturated weight. Next, the sample was placed in a numbered paper envelope and baked in an oven at 80 °C for more than 48 h until the weight reached a constant weight, which was recorded as the dry weight.

$$RelativeWaterContent\left(\%\right)={\displaystyle \frac{Freshweight\left(g\right)-Dryweight\left(g\right)}{Saturatedweight\left(g\right)-Freshweight\left(g\right)}}\times 100\%$$

(1)

The remaining samples were ground into powder and 0.1 g was accurately weighed to determine the content of Non-Structural Carbohydrate (NSC) through the anthrone colorimetric method [30].

EIS was measured using an impedance analyzer (HP4284A, Agilent, Santa Clara, CA, USA). The Ag/AgCl electrode (RC1, WPI Ltd., Sarasota, FL, USA) was first connected to the impedance analyzer, and then the electrode gel was added to remove the polarization resistance of the electrode. The leaf surface was carefully cut using scissors at the tip of the leaf avoiding the proximity to the veins. The cuts should be as small as possible and able to be inserted into the electrode. The cross-section of the cut leaf samples was connected to the electrode gel. Although the EIS process requires cutting small sections of leaves for measurement, it is important to emphasize that this method is minimally invasive and does not significantly harm the plant or negatively impact its overall health and growth. Studies have indicated that the effects of cutting small leaf sections are negligible concerning the plant’s physiological well-being (Figure 3).

The EIS values were determined at 42 frequencies. The frequency of electrical impedance parameter determination was set to 80 Hz–1M Hz. According to the measurement results of the electrical impedance of apple rootstock leaves, there is only one arc, and its impedance analysis pattern is suitable for the single-DCE model (belonging to the distribution model), as shown in the following Figure 4:

$$Z=R+{\displaystyle \frac{R_1}{1+{\left(i\tau \omega \right)}^{\psi }}}$$

(2)

First, the following five parameters are obtained: high-frequency resistance (R), low-frequency resistance (R₁), relaxation time (τ), Relaxation time distribution coefficient (ψ), and Sinusoidal voltage frequency (ω).

Through Equations (3) and (4), calculate the extracellular resistance (R_e) and intracellular resistance (R_i). Normalize according to the cross-sectional area and length of each sample, and calculate the corresponding resistivity by the Equation (5): high-frequency resistivity (r) low-frequency resistivity (r₁), extracellular resistivity (r_e), and intracellular resistivity (r_i) [12,31]:

$$R_e=R+R_1$$

(3)

$$R_i=R\left(1+{\displaystyle \frac{R}{R_1}}\right)$$

(4)

$$r_{x\left(\Omega m\right)}=R_x{\displaystyle \frac{A}{l}}$$

(5)

2.3.2. Machine Learning Classification Methods

In this study, shrinking Linear Discriminant Analysis (sLDA) and fuzzy K-Nearest Neighbor (FKNN) algorithms were employed to classify EIS parameters, providing valuable insights into the plant’s physiological status.

sLDA is based on Wilks’ lambda criterion [32] to analyze the effect of variables on the grouping discrimination, so as to introduce or exclude variables. Firstly, the feature vectors are separated from the known classes using a linear hyperplane. Then the mean feature vector and covariance matrix are calculated. The feature vector x is classified according to Equations (6)–(8):

$$a{x}^T+b$$

(6)

$$a=C^{-1}\left(\mu 1-\mu 2-\mu 3-\cdots \dots -\mu i\right)$$

(7)

$$b=-2^{-1}\left(\mu 1+\mu 2+\mu 3\dots \dots +\mu i\right)a^T$$

(8)

where $\mu 1,\mu 2,\mu 3,\dots \dots \mu i$ are the mean eigenvector and C⁻¹ is the covariance matrix for all classes.

The basic principle of the Fuzzy K Nearest Neighbours (FKNN) algorithm is that by assigning different weighting coefficients to the k nearest neighbors of a test sample, the nearest neighbor with the largest weighting coefficient is computed using the fuzzy decision-making method, and the test sample is attributed to the class with the largest nearest neighbor value [33]. FKNN adds classification heuristic algorithms, and the optimal solution is obtained through heuristic algorithms by constantly searching for the optimal solution, which can improve the classification accuracy [2].

2.4. Data Analysis

The significance of the difference between the treatments at different sampling times was tested by contrasts using Bonferroni-corrected significance levels. The normality of residuals was checked using the Shapiro–Wilk test, and the homogeneity of variance was checked using Levene’s test. The results indicated that the assumptions of normality and homogeneity were satisfied for all analyses. The data of this experiment were analyzed by one-way analysis of variance (ANOVA) and least significant difference (LSD) using IBM SPSS v26 to deal with the significant differences analysis of physiological data and EIS data. SigmaPlot 12.5 was used to eliminate outliers and figure the data. The Jupyter Notebook V7.0.0 development environment was used to write the code of sLDA and FKNN algorithm and output the classification accuracy and visualization results. Before ML classification, the data is normalized using the Standard Scaler method for better robustness.

3. Results

3.1. Effect of Water Stress on Physiological Parameters

After the initial water stress episode (day 14), RWC values in the WL group showed no significant change compared to CK, while both DR and FD groups exhibited significant decreases in RWC with statistically significant differences from the control (p < 0.05). Following the subsequent water stress episode (day 42), only the FD group displayed significant differences in RWC compared to CK (p < 0.05) (Figure 5A).

Regarding SPAD values, after the initial stress episode (day 14), neither WL nor FD groups showed significant changes compared to CK, with no statistically significant differences observed (p < 0.05). Conversely, the DR group demonstrated a significant decrease in SPAD relative to CK (p < 0.05). After the subsequent stress episode (day 42), all treatment groups (DR, WL, FD) showed non-significant SPAD variations with no significant differences from CK (p < 0.05) (Figure 5B).

Fv/Fm measurements revealed that after the initial stress episode (day 14), DR showed no significant difference from CK (p < 0.05). Both WL and FD groups exhibited significant Fv/Fm reductions with statistically significant differences versus CK (p < 0.05). Following the subsequent stress episode (day 42), FD showed a significant decrease differing from CK (p < 0.05), while DR and WL displayed non-significant differences (Figure 5C).

For NSC content after the initial stress episode (day 14), all treatment groups (DR, WL, FD) showed significant increases with statistically significant differences from CK (p < 0.05). After the subsequent stress episode (day 42), no significant NSC differences were observed between any treatment group and CK (p < 0.05) (Figure 5D).

3.2. Effect of Water Stress on EIS

Following the initial water stress episode (day 14), the r-value exhibited significant increases in both DR and FD groups compared to CK, whereas WL showed no significant change relative to the control (p < 0.05). After the subsequent water stress episode (day 42), FD demonstrated a significant increase with statistically significant differences from CK, while both DR and WL displayed non-significant changes with no significant differences versus control (p < 0.05) (Figure 6A).

Regarding r₁-values post-initial stress (day 14), both WL and FD showed significant decreases with statistically significant differences from CK, whereas DR maintained non-significant variation (p < 0.05). Following the subsequent stress episode (day 42), none of the treatment groups (DR, WL, FD) exhibited significant differences in r₁-values compared to CK (p < 0.05) (Figure 6B).

For r_i-values after the initial stress episode (day 14), significant differences from CK were observed in DR and FD groups, while WL showed no significant difference (p < 0.05). Post-subsequent stress (day 42), all experimental groups (DR, WL, FD) displayed non-significant differences in r_i-values relative to CK (p < 0.05) (Figure 6C).

Analysis of r_e-values revealed significant differences from CK in DR and FD groups after initial stress (day 14), whereas WL showed non-significant variation (p < 0.05). Following subsequent stress (day 42), no significant differences in r_e-values were detected between any treatment group and CK (Figure 6D).

3.3. Correlation Analysis

As shown in

Table 1. Correlation between electrical impedance parameters and physiological parameters.

EIS Parameters	Relative Chlorophyll Content SPAD	Maximum Photochemical Efficiency Fv/Fm	Relative Water Content RWC	Non-Structural Carbohydrate NSC
re	0.532 *	0.341	0.471 *	−0.389
ri	−0.684 **	−0.575 *	0.613 *	0.520 *
r	−0.523 *	−0.584*	−0.641 **	0.552 *
r1	0.485 *	0.314	0.495 *	−0.367

Note: * indicates significant correlations at the 0.05 level, ** indicates significant correlations at the 0.01 level and fonts are in boldface when the correlations are significant.

, during water stress in apples, RWC, SPAD, and Fv/Fm exhibit a positive linear correlation with apple electrical impedance parameters (r, r_i), while NSC shows a negative linear correlation with these impedance parameters. The correlation between the electrical impedance parameters r, r_i and RWC, SP1AD, NSC, Fv/Fm is stronger than the correlation between the impedance parameters r_e, r₁ and RWC, SPAD, NSC, Fv/Fm.

3.4. Classification Results

A total of 360 samples were used for each classification. The training set and test set were divided according to the ratio of 7:3.

The categorized results through sLDA for r, and r_i parameters are illustrated in Figure 7. Figure 7A achieves an accuracy of 86.1%, which is 4.6% higher than the 81.5% recorded in Figure 7B. In Figure 7A, three sample points from DR are misclassified as CK, while two from CK are categorized as DR. Additionally, five FD samples are misclassified as WL, and one WL sample is assigned to FD. Notably, there is no cross-misclassification between the state pairs WL-FD and CK-DR. In Figure 7B, two sample points belonging to DR are categorized into CK; three sample points belonging to CK are classified into DR; five sample points belonging to FD are categorized into WL; and four sample points belonging to WL are categorized into FD. However, there is no confusion of sample points between the two states of WL and FD and the two states of CK and DR. Despite some degree of sample overlap, all four moisture states (CK, DR, WL, FD) remain clearly distinguishable. However, Figure 7A exhibits a higher classification accuracy, demonstrating a more precise differentiation of states.

Similar to the sLDA analysis, Figure 8 presents the classification results using the FKNN algorithm. Figure 8A has a classification accuracy of 87%, which is higher (4.6%) than that of Figure 8B (82.4%). Figure 8A has three sample points belonging to DR divided into CK; one sample point belonging to WL is divided into FD; two sample points belonging to FD are divided into WL. But there is no confusion of sample points between CK and WL or FD, same as DR, which has no confusion of sample points with WL or FD. In Figure 8B, two sample points belonging to DR were assigned to CK; two sample points belonging to CK are assigned to DR; five sample points belonging to FD are assigned to WL; and three sample points belonging to WL are assigned to FD. However, there was no confusion of sample points between CK and WL or FD, same as DR, which has no confusion of sample points with WL and FD. In sum, although some of the sample points in Figure 8A,B overlap, the four water states CK, DR, WL, and FD can be identified in both classifications.

4. Discussion

4.1. Physiological Parameters Selection

For the SAPD parameter, under waterlogging and flooding stress, the lack of oxygen in the root system affects plant growth, chlorophyll synthesis, and photosynthesis. Under drought stress, chloroplasts are destroyed, which could also inhibit chlorophyll synthesis and photosynthesis [34]. Compared with the CK group, the SPAD values of DR, WL, and FD decreased.

For the NSC parameter, under waterlogging and flooding stress, plants may increase starch synthesis and storage for growth and metabolism, while decreasing soluble sugar synthesis and accumulation to reduce energy consumption and maintain cellular metabolism [35,36]. Under drought stress, plants need to utilize starch to provide energy to maintain metabolism and growth, which may decrease starch synthesis and storage, resulting in a decrease in starch content [37]. Meanwhile, drought stress may increase the synthesis and accumulation of soluble sugars to meet the needs of metabolism and cell wall water maintenance [34]. As a consequence, NSC increases along with the water stress in this experiment.

For the RWC parameter, under waterlogging and flooding stress, the apple root system is hypoxic and its water transport capacity is reduced, resulting in a decrease in the RWC of the leaves [38]. Under drought stress, insufficient water supply causes cells to lose water continuously, and water transfer between roots and leaves is blocked, which also leads to a decrease in RWC [39].

For the Fv/Fm parameter, under waterlogging and flooding stress, excessive water in the soil leads to root hypoxia and a large accumulation of reactive oxygen species (ROS, such as superoxide anions and hydrogen peroxide) [40,41]. This directly attacks the core protein D₁ of PSII, destroying the electron transfer function of PSII, thereby significantly reducing the Fv/Fm [42]. Under drought stress, plant stomata are closed, inhibiting the synthesis of photosynthetic pigments, reducing the activity of Rubisco enzymes, and causing damage to PSII, causing a continuous decrease in Fv/Fm [43].

Physiological data revealed that, compared to the initial water stress episode, the subsequent episode exerted a reduced overall impact on plants, with key physiological parameters remaining statistically unchanged. After the subsequent episode of water stress, only FD and CK had significant differences in RWC and Fv/Fm parameters, while other indicators and treatments did not show significant differences. This may be because after experiencing the initial water stress episode, plants developed certain physiological adaptability by regulating cell osmotic potential and enhancing the antioxidant system, thereby reducing their sensitivity to the subsequent episode [44]. In addition, the initial water stress episode may cause cell damage, destroy the stability of the cell membrane, store too much water in the cell, cause cell membrane rupture, and affect the normal growth of the plant [45]. Therefore, plants can establish a memory of the subsequent episode through physiological regulation and reduce the impact of repeated stress [46,47].

4.2. EIS Parameters Selection

For r and r_i parameters, under waterlogging and flooding situations, the increase in water content inside the cell results in the restriction of ion channels inside the cell [48]. In this context, low-frequency current cannot pass through the cell membrane, while high-frequency current can not only flow from the cell gap but also can pass through the cell membrane from the intracellular flow [16,49]. Therefore, r (i.e., high-frequency resistivity) and r_i (i.e., intracellular resistivity) could better represent the status of the plants under water stress. Similarly, r and r_i are more representative in the case of drought stress. This could be reasoned by the fact that the reduction in the internal water content of the cell leads to an increase in the concentration of ions. Under this circumstance, the cell membrane and the cell wall of conductivity decrease [50], limiting the low-frequency current passing through the cell membrane.

Water stress can damage the metabolic activity of plant cells and trigger the production of reactive oxygen species (ROS), which accelerates membrane lipid peroxidation, destroys the integrity of the cell membrane, and disrupts the cellular ion balance [51]. Since the cell membrane is selectively permeable, its structural damage can lead to ion leakage, reduce the intracellular ion concentration, and increase the ion concentration in the extracellular space, increasing intracellular resistance (r_i) [16,33].

Under water stress, chloroplasts are damaged, photosynthetic efficiency is reduced [52], SPAD and Fv/Fm are reduced, and plants synthesize NSC to adapt to water shortage. Photosynthesis usually depends on ion exchange and electron transfer within photosystems I and II. When metabolic activity decreases under stress, ion movement within the cell becomes sluggish [53], further reducing conductivity and causing an increase in intracellular resistance (r_i).

As water stress intensifies, the rise in r_i reflects the gradual damage to the cell membrane and ion imbalance. However, plants may initiate compensatory mechanisms to mitigate this damage, such as activating repair processes to increase cell wall thickness. Thickening of the cell wall helps limit ion leakage, slow down the increase in resistance, and restore cell function, ultimately enhancing the plant’s resistance to stress [54]. The EIS parameter trend graph shows that the overall stress condition of the rootstock under the subsequent episode of drought and flood stress was less than the initial, which is consistent with the changing pattern of physiological parameters. This may be because the initial water stress episode has a training effect on the rootstock and enhances its ability to adapt to adversity. Under the subsequent episode of drought stress, plants may maintain the balance of intracellular and extracellular ion concentrations by regulating ion channel activity [47], while synthesizing and accumulating osmotic regulating substances, such as proteins, sugars, and organic acids, to resist water loss [55].

Notably, the classification results of the four apple rootstocks statuses using EIS parameters are relatively positive. Moreover, the classification of apple rootstock status by r and r_i can greatly save time for data collection, this method avoids the time-consuming and potentially contaminating process of preparing chemicals for determining physiological parameters, so that the status of apple rootstock can be quickly detected. These results evidence the advantages of using EIS for apple rootstock’s status detection.

It should be noted that this work focused on the responses of leaves to water stress, because the leaf is the most sensitive part of the plant to adversity, and that leaf could directly reflect the plant’s ability to adapt to adversity, e.g., water stress [56]. This may be because when the plant body is subjected to water stress, the root system regulates the water pathways in the stem and leaves to maintain homeostasis, transports water to the leaves through the stem and regulates the plant’s response to water stress by successive signaling to the stem and leaves through abscisic acid and gibberellins [6,57]. Future work could be conducted to test other parts of the plants, i.e., root and stem, together with the leaf, and examine their individual reaction and collective responses to water stress. These findings provide a basis for comparison with other apple rootstocks available nationally and internationally, and future studies could extend this approach to evaluate the responses of a wider range of rootstocks. Moreover, further research is needed to establish the relationships between EIS parameters and leaf/stem/soil water potential dynamics, which would provide a more comprehensive understanding of plant water status under different stress conditions and enhance the applicability of EIS in both scientific and agricultural contexts.

5. Conclusions

This study investigated the variations in impedance spectroscopy parameters (r, r_i, r₁, r_e) and physiological parameters (RWC, SPAD, Fv/Fm, NSC) of apple rootstocks under interval water stress. The results indicated that the initial water stress episode event had a more pronounced impact on plant responses compared to the subsequent episode of water stress. Moreover, a strong correlation was observed between electrical and physiological parameters, with r and r_i exhibiting the highest correlation. Therefore, r and r_i can serve as reliable indicators for assessing the water status of apple rootstocks under water stress conditions. Furthermore, FKNN and sLDA algorithms were used to classify the water status of apple rootstocks with high classification accuracy, proving that machine learning is feasible for classifying water status. Further research is needed to establish the relationships between EIS parameters and leaf/stem/soil water potential, which would strengthen the applicability of EIS as a practical tool for monitoring plant water status in different environments.