Continuous Seasonal Monitoring of Nitrogen & ??Water Content in Lettuce ??Using a Dual Phenomics Systems - PlantArray & Hyperspectral

The collection and analysis of large amounts of information on a plant-by-plant basis contributes to the development of precision fertigation and may be achieved by combining remote-sensing technology with high-throughput phenotyping methods. Here, lettuce ?? plants (Lactuca sativa) were grown under optimal and suboptimal nitrogen and irrigation treatments from seedlings to harvest.

A PlantArray System was used to calculate and log weights, daily transpiration, and momentary transpiration rates throughout the experiment. From 15 d after planting until experiment termination, the entire array of plants was imaged hourly (from 09.00?h to 14.00?h) using a hyperspectral moving camera. Three vegetation indices were calculated from the plants’ reflectance signal: red-edge chlorophyll index (RECI), photochemical reflectance index (PRI), and water index (WI), and combined treatments, physiological measurements, and vegetation indices were compared. RECI values differed significantly between nitrogen treatments from the first day of imaging, and WI values distinguished well-irrigated from drought-treated groups before detecting significant differences in daily transpiration rate. The PRI, calculated hourly during the drought-treatment phase, changed with the momentary transpiration rate. Thus, hyperspectral imaging might be used in growing facilities to detect nitrogen or water shortages in plants before their physiological response affects yields.

Introduction Phenomics is the study of plant phenotype and its physical and biochemical qualities. It is considered a rapid method for selecting individual plants from an array based on their genome, interactions with the environment, and phenotype (Furbank, 2009; Zhao et al., 2019). The use of high-throughput phenotyping systems is becoming increasingly popular in plant science (York, 2019). Recent technological advances have enabled the measurement and characterization of plant phenotypic attributes on a large scale in a rapid, non-destructive, and precise manner. The use of remote-sensing technology in phenomics and agriculture, particularly in the hyperspectral domain, is widespread. While large-scale field research and applications rely mainly on multispectral satellite sensors or drones. Imaging spectroscopy is becoming more common as a tool in the laboratory, field (Chapman et al., 2014; Underwood et al., 2017), and greenhouse (Weksler et al., 2020, 2021a, b). In addition to remotesensing capabilities, whole-plant gravimetric systems are a key player in high-throughput phenotyping systems (Negin and Moshelion, 2017). Lettuce (Lactuca sativa) has been garnering much attention in this field. It is a fast-growing plant that is harvested approximately 60–90 days after planting (DAP) under field conditions (Gallardo et al., 1996; Pacumbaba and Beyl, 2011), and approximately 30 DAP in hydroponic growing systems (Donnell et al., 2011). This short growing period is appealing for research because results can be obtained in a short time. Several studies using hyperspectral technology have been performed with different varieties of lettuce, for different aims. Pacumbaba and Beyl (2011) measured lettuce leaves at the end of a 90 d macronutrient-deficiency experiment using a spectrometer and found significant changes in reflectance in the red and infrared spectral regions in stressed plants. Zhou et al. (2018) acquired 200 hyperspectral images of lettuce leaves with five levels of water content. Using wavelet transform and partial least squares regression, they estimated leaf water content and applied the model at the pixel level to show the spatial water distribution in the leaves. Murphy et al. (2019) used vegetation indices (VIs) derived from hyperspectral images of leaves to quantify variations in water absorption across the different leaf components at a very high spatial resolution. They discovered that the choice of leaf component for modeling has a marked influence on the model’s performance. Mo et al. (2015) imaged lettuce leaves with a hyperspectral camera and developed a model to distinguish discolored areas from healthy tissue. Eshkabilov et al. (2021) grew four lettuce cultivars with seven nitrogen (N) fertilizer treatments. They scanned freshly cut leaves using a hyperspectral camera and developed an algorithm to estimate nutrient levels in the measured leaves. Using cross-validation on 28 leaf samples, they achieved good predictive results. Although these studies show promise, they were all conducted at the scale of the leaf, mainly with cut leaves, and only once at the end of the experiment. Early detection of drought and N deficiency can be extremely valuable in production lines. The possibility of simultaneously measuring plant physiological properties using a gravimetric system and the impact of these properties on the signal received by proximal hyperspectral sensors is a novel concept (Weksler et al., 2020). Accordingly, the overarching aim of the present study was to use this combined phenomics system to investigate the detection of early symptoms of N deficiency and drought, to minimize the period between the appearance of physical symptoms in plants and their detection by the hyperspectral-sensing system. Specifically, we sought to discover how early the effects of N and drought treatments can be detected in whole plants without the need to cut their leaves, and how this compares with the plants’ physiological status. We investigated these questions at the canopy (plant) level using a set of carefully selected VIs that have been shown to be correlated with changes in leaf chlorophyll and water contents. These indices were evaluated daily throughout the entire experiment and at different times of the day at key stages of the experiment. Materials and methods Experimental setup and plant material The experiment was conducted in a semi-commercial greenhouse located at the Robert H. Smith Faculty of Agriculture, Food and Environment of the Hebrew University of Jerusalem in Rehovot, Israel. The temperature and relative humidity in the greenhouse were continuously monitored by the PlantArray meteorological station (Shahar Weksler et. al., (2021),?Jornal of Experimental Botany?DOI: 10.1093/jxb/erab561). The temperature and relative humidity ranged between 13 °C and 32 °C and 28% and 92%, respectively, with low values during the night. Seventy-two lettuce seedlings (Lactuca sativa var. Lior) were transplanted to 3.9 litre pots. The plants were grown from 22 October to 26 November 2020. Three different N treatments were administered to the plants: a control treatment (C), which was considered the optimal fertilization treatment (n=24 samples), a medium treatment (M), which was 70% of the control (n=24 samples), and a low treatment (L), which was 40% of the control (n=24 samples). The C, M, and L values corresponded to 70?ppm, 49?ppm, and 28?ppm N, and all other elements were administered at their optimal levels. At 25 DAP, when the plants had a daily transpiration value of approximately 250?g water, 12 plants from every treatment were subjected to progressive drought for 9 d, resulting in six treatments: the three levels of N with optimal irrigation (CW, MW, and LW), and the three levels of N with drought (CD, MD, and LD) (Fig. 1). For the drought treatment, irrigation was reduced daily to 70% of the plant’s daily transpiration value on the previous day Dalal et al., 2019). This method prevented rapid depletion of water during the drought treatment before the final day, on which no irrigation was provided. Then, the irrigation was returned to full capacity for 7 d to allow the plants to recover.

Phenotyping systems

A combination of two phenotyping systems was used. The first system was the PlantArray system, a high-throughput whole-plant functional phenotyping system. Individual plants were placed in the PlantArray’s

Special load cells, which act as weighing lysimeters. Each load cell individually controlled the fertigation and logged the weight measurements. The 72 plants were randomly placed in an array on a table in the greenhouse (Fig. 1), and every 3 min, the individual weight of each plant was logged. Following the methodology described in Halperin et al. (2017), each pot was filled with an inert medium (quartz sand) and placed in a load cell. The drainage of the pots was restricted with a dedicated plastic container to ensure that irrigation to pot capacity was reached. Evaporation was restricted using a plastic cover with a hole in its center for the emerging plant stem, thus limiting the changes in pot weight to the effect of plant transpiration. During the night, the system selectively and automatically irrigated each pot with its selected treatment, using four drippers for even water distribution. Plant fresh weight (FW) was calculated using the Plantarray System: before planting, the load cells were loaded with pots filled with sand and irrigated to pot capacity, and these pots were weighed to determine the pre-planting pot weight. After planting, the daily plant weight was calculated as the weight after irrigation when drainage had stopped minus the pre-planting weight. Each plant’s final FW was calculated as its weight on the last day of the experiment. Since the irrigation was applied only at night, and evaporation from the quartz sand was restricted, any weight change was attributed to water leaving the plant via the stomata. Thus, the plant’s daily transpiration was the difference in weight between the early morning, when the plant had not transpired, and the evening, when the plant had stopped transpiring. Transpiration rate (TR) was associated with a decrease in weight throughout the daytime and was calculated by the first derivative of two consecutive weight measurements. The point at which the water content in the pot cannot sustain the demand for transpiration by drought-treated plants and stomatal conductance begins to decrease is termed the physiological critical drought point. A piecewise linear function (as described in Halperin et al., 2017) was used to find the best fit between midday TR and soil water content and, consequently, the critical drought point. On the last day of the experiment, the plants’ stems were cut and the plants’ shoots were placed in paper bags. The roots were carefully washed with tap water and placed in a separate paper bag. The bags were then dried in an oven for 4 d at 60 °C and the dry weight (DW) of the roots and shoots was recorded. The second system was based on a remote-sensing system composed of a hyperspectral camera mounted on a moving platform above the plants, connected to the greenhouse’s ceiling (Weksler et al., 2020). The system began acquiring images at 15 DAP, when the transplanted seedlings were sufficiently large. The camera (FX10, Specim, Finland) was a push-broom sensor with a 400–1000 nm spectral range. The camera was positioned 2 m above the growing table and had 512 pixels in a row and 224 bands in the spectral range. The lightweight camera produced a signal-to-noise ratio suitable for operation under the diffuse natural light conditions in the greenhouse between 09.00 h and 14.00 h. The raw images were calibrated to radiance using a pre-calibrated gain factor produced by the camera’s manufacturer and a closed shutter image (Weksler et al., 2020). Then, the radiance image was transformed to reflectance using a white reference panel (99% reflectance, Spectralon, Labsphere Inc., USA) placed in the scene at each data-acquisition session. A region of interest around each plant in the image was manually selected to extract the pixels representing the plant. A binary mask was then calculated and applied to the pixels to separate the green leaves from the background. The mask was generated by calculating the slope between spectral bands (680 nm, 740 nm), and excluding the leaves’ edges and shaded pixels using Otsu’s filter (Weksler et al., 2020). These selected pixels, representing each plant, were averaged to a mean spectral signal and smoothed using Savitzky– Golay transformation (Savitzky and Golay, 1964). The collection of mean spectral signals from each plant represented the reflectance signal of each plant at every image acquisition (Weksler et al., 2020). Figure 2 summarizes the process from image acquisition to plant mean reflectance spectrum.

Hyperspectral analysis

Three different indices were carefully selected to test the effects of, and hypotheses regarding, the different treatments (Table 1). The first was the red-edge chlorophyll index (RECI) (Gitelson et al., 2003), which utilizes the low sensitivity of the infrared region and the high sensitivity of the red-edge region to assess chlorophyll content. The second was the water index (WI) (Pe?uelas et al., 1993), which utilizes the water absorption band at ρ970 nm and a low sensitivity band outside the water absorption region at ρ900 nm. The third VI was the photochemical reflectance index (PRI) (Gamon et al., 1992), which utilizes a reflectance band that is affected by physiological changes (ρ531 nm) in relation to changes in photosynthetically active radiation (PAR) and a reference band (ρ571 nm). Three key experimental days were chosen for calculation of the indices on an hourly basis. These days represented pre-drought status (well-irrigated day; 23 DAP), the last day of drought (33 DAP), and recovery status (post-drought; 35 DAP).

Statistical analysis

ANOVA was performed to check for significant differences between the groups on an hourly or daily basis. Post hoc differences were calculated by the Tukey–Kramer test. A Welch test was used when the ANOVA assumptions of normal distribution and homogeneity of variance were not met. Post hoc comparison of differences between groups was performed using the Games–Howell post hoc test, and single pairwise comparisons were performed by using the Mann–Whitney test.

Daily outlier samples were removed using the 1.5 interquartile method (Ghasemi and Zahediasl, 2012). Using this method, the maximum number of outliers removed per group each day was found to be four, leaving a minimum of eight samples per group and a maximum of 12. All statistical differences were calculated using α=0.05 and were considered significant at P≤0.05. Statistical analysis, algorithms, and estimation were performed using Python with dedicated modules (Vallat, 2018; Terpilowski, 2019).

Following the hourly analysis, the most representative hours of the daily values were selected for the index calculations. The daily values of RECI and WI were calculated for the entire experiment and compared. In addition, the PRI was calculated explicitly on an hourly basis from 27 DAP to 30 DAP.

Results

The experiment was performed during the autumn. The atmospheric conditions throughout the imaging phase of the experiment (15–38 DAP) are presented in Fig. 3. They show typical PAR and vapor pressure deficit (VPD) values for the season. There were some cloudy and rainy periods when radiation was obscured by the clouds, which caused the VPD to drop. The final plant FW and DW measurements are presented in Fig. 4. Significant weight differences (P<0.05) were found between the CW and LW groups (well-irrigated plants). The same significant differences were found between the plants that received the additional drought treatment from 25 DAP. Comparing the dried plants’ biomass as DWs of shoots, roots, and shoots and roots together showed different behaviors. Shoot DW was significantly different between the CW and LW groups and between the CD group and both the MD and LD groups. Interestingly, the root DW of CW plants was significantly different from that of MW but not LW plants, most likely because of the relatively large standard deviation for the LW value. Root DW for CD plants was significantly different from that of plants subjected to the other two drought treatments. The DWs of shoots and roots combined tell a slightly different story: values for the CW treatment differed significantly from those of the other well-irrigated treatments. The same significant difference was calculated for the drought treatments (significant difference for CD versus MD and LD treatments).

Read More...