Seedlings of tomato (Solanum lycopersicum L. cv. Hezuo 903) were used. Seeds were germinated in a growth medium filled with a mixture of peat and vermiculite (3:1, v/v) in trays in a growth chamber. Seedlings (except those for nutrition experiments) at two-leaf stage were transplanted into plastic pots (15 cm diameter and 15 cm depth, one seedling per pot) containing peat and vermiculite (3:1, v/v). The two-leaf stage seedlings for nutrition experiments were transplanted into plastic pots containing vermiculite and perlite (3:1, v/v). All seedlings were watered and fertilized with Hoagland’s solution (Hoagland, 1950) every 2 days. The growth conditions were as follows: 12-h photoperiod, temperature of 25°C/20°C (day/night), and photosynthetic photon flux density of 600 μmol m^-2 s^-1.

When tomato seedlings were at four-leaf stage, drought experiment was performed. The day when we started withdrawing water was set as day 1. Soil relative water content was determined by using a soil moisture meter (Takeme, GREYWELL, CHN).

When tomato seedlings were at four-leaf stage, nitrogen deficiency experiment was performed. The treatment group was watered with nitrogen deficient nutrient solution. The Hoagland’s nutrient solution contained 354 mg/L Ca(NO₃)₂⋅4H₂O, 505.5 mg/L KNO₃, 115 mg/L NH₄H₂PO₄ and 246.5 mg/L MgSO₄⋅7H₂O, and nitrogen deficient nutrient solution contained 166.5 mg/L CaCl₂, 348.5 mg/L K₂SO₄, 136 mg/L KH₂PO₄, and 246.5 mg/L MgSO₄⋅7H₂O. The day when the nutrient solution was replaced was set as day 1.

When tomato seedlings were at five- to six-leaf stage, the plants were inoculated with Botrytis cinerea (BO5-10 strain) suspensions at a density of 2 × 10⁵ spores per mL (Zhang et al., 2015). The day when the seedlings were inoculated with B. cinerea was set as day 1.

When tomato seedlings were at four-leaf stage, they were separated into three groups. Plants of the three groups were, respectively, watered with nitrogen deficiency, full strength and double strength (708 mg/L Ca(NO₃)₂⋅4H₂O, 1011 mg/L KNO₃, 230 mg/L NH₄H₂PO₄, and 246.5 mg/L MgSO₄⋅7H₂O) Hoagland’s solution. The day beginning different nitrogen strength treatment was set as day 1. From day 1 to day 7, tomato leaves were sampled for chlorophyll fluorescence measurement and total nitrogen determination.

The monitoring of plant physiology is carried out by the system integrating chlorophyll fluorescence imaging module and multispectral imaging module (Figure 1). The chlorophyll fluorescence imaging module consists of lighting part, CCD (PCO 1300, PCO, GER) imaging part and circuit control part. The lighting part contains 16 × 100 W LEDs to provide actinic light for activating the photosynthesis, saturating light pulse for inhibiting photosynthesis and excitation light pulse for measuring the fluorescence signal. The central wavelength of the LED is 460 nm. Square light pipe is used to even LED lighting spot. LED lighting spots focus on the same region to form 0.45 m × 0.34 m uniform illumination area. The CCD camera with a cut-off high-pass filter to detect fluorescence above 650 nm is surrounded by the 16 lighting tubes. The fluorescence images are captured by the signal pulse generated by the control circuit. For multispectral imaging, CCD (PCO 1300, PCO, GER) is placed behind band-pass filters. The band-pass filters are set on a rotating plate. With the rotating of the plate, CCD captures images under different wavelengths. Incandescent lamps are used to provide visible wavelengths lighting condition.

A software supporting modulated chlorophyll fluorescence measurement is developed based on Visual Studio 2010 (Microsoft, United States). In order to avoid the interference of ambient light, plants were dark adapted for at least half an hour at night before chlorophyll fluorescence measurement. First, the minimal level of fluorescence F_o was obtained after the measuring light (0.1 μmol photons m^-2 s^-1) was opened. Next, the maximal level of fluorescence F_m was obtained when the photosynthesis was temporarily inhibited by a saturating light pulse (>6000 μmol photons m^-2 s^-1). Then, the actinic light (200 μmol photons m^-2 s^-1) was opened to simulate the ambient light condition, under which the photosynthesis of the plant leaves was activated. When the photosynthesis was stable, the steady-state fluorescence F was obtained, meanwhile a saturating light pulse was applied to elicit the maximal level of fluorescence F_m′ under light condition.

After chlorophyll fluorescence measurement of all plants on the locations of the seedling bed, the monitoring system went back to the initial position and acquired multispectral images of the same locations. Through two-dimensional scanning on the field, the monitoring system can achieve measurement in a larger scale.

In order to distinguish the fluorescence and spectral characteristic induced by drought, nitrogen deficiency, and plant disease (gray leaf mold), we choose three parameters to reflect the physiological status of plants. F_v/F_m, the maximum quantum efficiency of photosystem II (PSII), represents the potential maximum capacity of PSII reaction center transforming the photon energy absorbed by PSII to photochemical energy. Φ_PSII represents the operating quantum efficiency of PSII. 550/510, the ratio of 550 nm spectral reflectance to 510 nm spectral reflectance, represents the degree of greening of the leaves.

Image processing and data analysis is carried out using Matlab 2017b (Mathworks, United States). Resolution of images acquired is 696 × 520 and plant area should be segmented from the images. For chlorophyll fluorescence parameter images, the parameter of the non-plant area is very small as the non-fluorescing of non-plant area. So, we set threshold to segment the plant area. For multispectral images, we implemented an algorithm based on K-means clustering (Macqueen, 1965; Lucchese and Mitra, 2001), in which image can be grouped into two classes representing the plant and the background, respectively. Based on the four chlorophyll fluorescence images F_o, F_m, F, F_m′, we can calculate the maximum quantum efficiency of PSII F_v/F_m = (F_m–F_o)/F_m and operating quantum efficiency of PSII Φ_PSII = (F_m′–F)/F_m′. Based on the spectral images of 550 nm and 510 nm, we can calculate the relative reflectance 550/510.

Leaf water potential (Begg and Turner, 1970) of drought-stressed plants was measured with a water potential meter (WP4C, Decagon Devices, United States). Canopy leaves of plants under different drought condition were cut and placed inside the sample cup. The water potential was determined by the dew point method.

The total nitrogen is determined spectrophotometrically using the Nessle’s reagent. The first procedure is the digestion of the sample. Fresh plant leaves were dried to constant weight and ground to a fine power with mortar and pestle. Sample powder (0.1 g) was placed in a boiling tube. H₂SO₄ (5 mL) was added to the boiling tube and then the tube was placed overnight. All boiling tubes with the samples were heated in graphite digestion apparatus and the temperature was raised from 150 to 200°C until the appearance of white smoke. Ten drops of 30% H₂O₂ was added when the solution became dark brown and mixed the solution constantly. Then the solution was heated for 5 min. The H₂O₂ treatment and heating was repeated 3–5 times till the remained H₂O₂ was removed and the solution was clear and transparent. Then the solution was diluted to 50 mL with double distilled water (ddH₂O).

After pretreatment procedure, 1 mL sample solution was mixed with 2 mL 25% sodium potassium tartrate, 2 mL Nessler’s reagent (HgI₂, KI, and NaOH) and 45 mL non-ammonia water. The absorbance of the resulting solution at 480 nm (OD₄₈₀) was measured by spectrophotometer (TU-1900, PERSEE, CHN). For standard solution, 10 mg/L (NH₄)₂SO₄ was prepared. Calibration curve was obtained with OD₄₈₀ of 0, 2, 4, 6, 8, and 10 mL standard solution. By comparison with the calibration curve, the total nitrogen of the sample was determined.

The soil relative water content decreased dramatically from day 1 to day 5 after withdraw water (Figure 2). In the later stage of experiment, the soil relative water content was relatively stable but at a low level. Pseudo-color images of Φ_PSII, F_v/F_m and 550/510 parameters (Figure 3) visualize how drought stress affected the physiological status of tomato plants (see Supplementary Figures S2, S3 for more images). As soil relative water content decreased, mean Φ_PSII declined steadily from above 0.4 on day 1 to about 0.3 on day 5 (Figure 4A). When plants suffered from severe drought stress (day 7 – day 9), tomato leaves wilted and the Φ_PSII did not decrease further, remaining about 0.25. Until this time, both F_v/F_m and 550/510 showed significant declines. Therefore, Φ_PSII is a sensitive indicator to show changes in plant physiology during the initial stage of drought stress when the plant morphology didn’t change obviously. For plants in the well-watered control group, Φ_PSII and F_v/F_m remained stable around 0.4 and 0.8, respectively. Meanwhile, the value of 550/510 showed a slight rising trend (Figure 4B).

From pseudo-color images of Φ_PSII, F_v/F_m, and 550/510 parameters (Figure 5), we can see how nitrogen deficiency affected physiology of tomato plants (see Supplementary Figures S4, S5 for more images). As the degree of nitrogen deficiency increased, Φ_PSII, F_v/F_m, and 550/510 all showed declining trend (Figure 6A). The decreases of parameters first appeared in old leaves and then spread to new leaves. This resulted in an increase in parameter standard deviation, especially for F_v/F_m. For tomato plants in the control group, Φ_PSII and F_v/F_m remained around 0.4 and 0.8, respectively, and 550/510 showed a rising trend above 1.5 (Figure 6B).

From pseudo-color images of Φ_PSII, F_v/F_m, and 550/510 parameters (Figure 7), we can see how B. cinerea infection affected tomato plants (see Supplementary Figure S6 for more images). As the disease symptom aggravated, Φ_PSII, F_v/F_m, and 550/510 all showed significant decline (Figure 8). The decline rate is faster than that of nitrogen deficiency. The plant morphology changed a lot, as shown by the curled downward infected leaves.

For drought stress, both soil relative water content and Φ_PSII declined steadily during the first few days. Therefore, by sampling tomato leaves in different drought degree, we studied the correlation between water potential and Φ_PSII of tomato leaves (Figure 9). We can see the non-linear relationship between Φ_PSII and the water potential. When the drought is not severe, Φ_PSII is sensitive to the changes in water potential. Φ_PSII varies from 0.43 to 0.28, when water potential is in the range of -0.5∼-1.0 MPa. Plants experiencing a threshold levels of water deficiency (water potential <-1.0 MPa) displays considerable decrease in Φ_PSII. However, for leaves which have water potential -1.0 to -2.5 MPa, Φ_PSII only declines from 0.28 to 0.21. This correlation between Φ_PSII and water potential suggests that rapid decline of Φ_PSII is a sensitive indicator of initial drought stress. As a non-invasive means of identifying the degree of drought stress, Φ_PSII can be potentially used to monitor the plant growth status, especially for water saving irrigation.

When the plants suffer nutrient deficiency, both Φ_PSII and F_v/F_m decline with the progression of the stress (Lu and Zhang, 2000; Damatta et al., 2002). In contrast to the early stage of drought stress, the decline of F_v/F_m is more significant in response to nutrient deficiency. The increase in standard deviation of F_v/F_m (Figure 6A) also shows F_v/F_m is representative. So, by sampling tomato leaves with different nitrogen strength treatment, we studied the correlation between F_v/F_m and total nitrogen content of tomato leaves. As shown in Figure 10, the correlation between F_v/F_m and total nitrogen content is high with a coefficient of determination of 0.85. In other words, F_v/F_m can linearly reflect the nitrogen content.

Plant physiology diagnosis based on imaging has advantage of high visibility and large data volume. We try to analyze the images with simple point of view, and further to prepare for the machine automatic diagnosis. We choose chlorophyll fluorescence and multispectral parameters as the indicators to monitor the crop physiology status.

Drought, nitrogen deficiency and disease are the three types of crop stress frequently occur in agriculture. At least three indicators are required to characterize the stresses. Chlorophyll fluorescence parameters Φ_PSII and F_v/F_m are sensitive to the plant physiology and are widely used. We choose them as two indicators, and we choose the third indicator from multispectral images. In traditional measurement of multispectral parameters, absolute reflectivity is often used and the calibration board is necessary. However, the use of calibration board is not convenient for the system structure and large area measurement. Thus, relative reflectance between different wavelengths has been used in this study. The ratio change of different spectral composition can reflect the change of leaf chroma caused by the composition of the leaves. Meanwhile, the use of relative reflectance decreases the requirement of the light source for spatial domain uniformity.

By comparing with the control groups, we can identify the threshold of stressful conditions. Φ_PSII is sensitive to plant physiological conditions, and has fluctuation even in the control group. Under lighting condition of 200 μmol m^-2 s^-1 photosynthetic photon flux density, the normal range of Φ_PSII should be around 0.4. When Φ_PSII decreases close to 0.3, the plants may be in some stressed conditions. The fluctuation of F_v/F_m is relative small and the normal range of F_v/F_m should be around 0.8. When F_v/F_m decreases below 0.75, PSII reaction center has been affected. For 550/510, it increases with the accumulation of chlorophyll. The normal range of 550/510 should be above 1.5.

We found that different types of stresses can induce distinct changes of chlorophyll fluorescence and multispectral signals, which can be used as indicators of plant physiology. For the drought stress, stomatal closure results in the decline of photosynthetic rate and the decrease in Φ_PSII. Φ_PSII is proportional to the carbon assimilation rate. In general, Φ_PSII decreases in response to different stresses (Baker, 2008). Drought stress does not affect F_v/F_m sharply when the plants are not in extreme drought condition (Giardi et al., 1996; Woo et al., 2008). A long-term drought stress will cause a decrease in F_v/F_m, potentially due to a secondary nutrient stress. Therefore, Φ_PSII can show the early warning of drought, and we use Φ_PSII to evaluate the degree of drought stress.

F_v/F_m was commonly used for assessing plant performance under stress (Bresson et al., 2015; Awlia et al., 2016). Nitrogen deficiency affects PSII photochemistry and decreases the quantum yield of PSII electron transport. The decline of F_v/F_m is mainly due to the decrease of the maximal fluorescence F_m (Lu and Zhang, 2000). The photoinhibition caused by nitrogen deficiency is associated with the inactivation of PSII reaction centers. In addition, nitrogen deficiency induced a significant decrease in CO₂ assimilation capacity. CO₂ assimilation acts as a major sink for the reducing equivalents (ATP and NADPH) generated by the primary photochemical reactions.

Relative reflectance is also an index reflecting different stresses. Several indices have been used for estimating chlorophyll or nitrogen content (Yendrek et al., 2017). For plants exposed to nitrogen deficiency, the synthesis of chlorophyll is hindered, resulting in the change of leaf spectral distribution (Kuckenberg et al., 2009; Bürling et al., 2011). We use relative reflectance 550/510 to represent the change of leaf chroma. As nitrogen content decreases, the inhibition of chlorophyll synthesis will cause decline of relative reflectance 550/510 (Figures 5, 6A). However, during the early stage of drought stress, the inhibition of chlorophyll synthesis will not be obvious and the relative reflectance 550/510 will not decline (Figure 4A). The different changing pattern of 550/510 values under nitrogen deficiency as compared to other stresses can be used to identify the occurrence of nitrogen deficiency.

Pathogens acquire nutrient through interfering the plant metabolism, resulting in similar changes in chlorophyll fluorescence and multispectral signals as the nitrogen deficiency (Rolfe and Scholes, 2010; Ivanov and Bernards, 2016). However, the development rate of pathogen infection is fast, and plant leaves will curl downwards. To distinguish these two stresses, the analysis based on images should not be ignored (Garciaruiz et al., 2013; Wetterich et al., 2016).

Statistics analysis about the indicator images can be carried out for drought and nitrogen deficiency stress. However, plant disease has the characteristic of sudden incidence and heterogeneity of infection area. Statistics may cover the disease/healthy area. Thus, analysis based on images should not be ignored. On the other hand, overall growth condition of the crop is more important in practical agricultural production. If the infection area is small and local, that means the disease doesn’t affect the crop broadly yet. Before the disease is spreading out, effective management should be taken.

There are also some other possibilities that affect the plant photosynthesis. In the process of agriculture production, apart from the Φ_PSII decrease caused by drought stress, another common stress is temperature fluctuation. This will cause interference to the judgment of plant physiology. However, in practice the temperature factor can be excluded by temperature recording.