
==== Front
Plant Phenomics
Plant Phenomics
PLANTPHENOMICS
Plant Phenomics
2643-6515
AAAS

10.34133/plantphenomics.0252
0252
Research Article
Fruit Water Stress Index of Apple Measured by Means of Temperature-Annotated 3D Point Cloud
https://orcid.org/0000-0002-6248-4333
Tsoulias Nikos 1
https://orcid.org/0000-0002-8887-839X
Khosravi Arash 1 2
https://orcid.org/0000-0002-6274-4596
Herppich Werner B. 1
https://orcid.org/0000-0002-8930-7855
Zude-Sasse Manuela 1 *
1 Leibniz Institute for Agricultural Engineering and Bioeconomy (ATB), Department Agromechatronic, WG Precision Horticulture, Potsdam, Germany.
2 Department of Agricultural, Food and Environmental Science, Marche Polytechnic University, 60131 Ancona, Italy.
* Address correspondence to: mzude@atb-potsdam.de
18 9 2024
2024
6 025218 1 2024
23 7 2024
20 8 2024
18 9 2024
Copyright © 2024 Nikos Tsoulias et al.
2024
Nikos Tsoulias et al.
https://creativecommons.org/licenses/by/4.0/ Exclusive licensee Nanjing Agricultural University. No claim to original U.S. Government Works. Distributed under a Creative Commons Attribution License 4.0 (CC BY 4.0).

In applied ecophysiological studies related to global warming and water scarcity, the water status of fruit is of increasing importance in the context of fresh food production. In the present work, a fruit water stress index (FWSI) is introduced for close analysis of the relationship between fruit and air temperatures. A sensor system consisting of light detection and ranging (LiDAR) sensor and thermal camera was employed to remotely analyze apple trees (Malus x domestica Borkh. “Gala”) by means of 3D point clouds. After geometric calibration of the sensor system, the temperature values were assigned in the corresponding 3D point cloud to reconstruct a thermal point cloud of the entire canopy. The annotated points belonging to the fruit were segmented, providing annotated fruit point clouds. Such estimated 3D distribution of fruit surface temperature (TEst) was highly correlated to manually recorded reference temperature (r2 = 0.93). As methodological innovation, based on TEst, the fruit water stress index (FWSIEst) was introduced, potentially providing more detailed information on the fruit compared to the crop water stress index of whole canopy obtained from established 2D thermal imaging. FWSIEst showed low error when compared to manual reference data. Considering in total 302 apples, FWSIEst increased during the season. Additional diel measurements on 50 apples, each at 6 measurements per day (in total 600 apples), were performed in the commercial harvest window. FWSIEst calculated with air temperature plus 5 °C appeared as diel hysteresis. Such diurnal changes of FWSIEst and those throughout fruit development provide a new ecophysiological tool aimed at 3D spatiotemporal fruit analysis and particularly more efficient, capturing more samples, insight in the specific requests of crop management.

Horizon 2020 http://dx.doi.org/10.13039/501100007601 SHEET 862665 Manuela Zude-Sasse Bundesministerium für Wirtschaft und Klimaschutz http://dx.doi.org/10.13039/100021130 IRRIWELL 01DH21016 Manuela Zude-Sasse
==== Body
pmcIntroduction

Due to increasing awareness of global warming-related water scarcity, water status analysis has been addressed as pivotal topic in ecophysiological studies capturing also woody plants in fruit production [1]. The temperature and water status of fruit tree is closely related to orchard yield performance [2–4]. Moreover, high fruit temperature can lead to fruit damage appearing more frequently due to climate change and resulting increase in the maximum temperature as well as appearance of heat waves [5]. Consequently, suitable methods for monitoring the crop water status are essential for ecophysiological studies and resilient food production. Current monitoring systems can be defined according to requirements on the spatial and temporal resolution of applications, sensor platforms, and sensor types [6]. For monitoring of tree and fruit water status, proximal and remote sensors can be applied. Proximal sensors can be used in individual trees [7,8], but the representation of the individual response of fruit, trees, and group of trees depends on the homogeneity of the orchard [9]. Considering older orchards or enhanced biodiversity within an orchard, the representation capacity of few individual trees and fruit data is certainly low. Consequently, various sensor types and their combinations are necessary to obtain comprehensive data [10]. In this context, e.g., various combinations of fruit gauges and leaf patch clamp pressure probes have been employed for continuous monitoring of olive (Olea europaea L.) and nectarine (Prunus persica L.) addressing the tree water status [10–12]. Apple trees (Malus x domestica Borkh.) were monitored by both fruit gauges and Scholander pressure chambers [13], while Grilo et al. [14] continuously applied a combination of fruit gauges and sap flow sensors on orange trees (Citrus sinensis). Although these combinations of proximal sensors increased accuracy, the practical use of this approach is limited by the variability within a large number of trees, complex specific installation and maintenance of sensors, and the correct interpretation of the various water status parameters under highly variable experimental or commercial orchard conditions [10,15,16].

Remote sensing at the fruit level may overcome these drawbacks, since the sensors capture many or even all trees in the orchard. Noninvasive techniques, including 2-dimensional (2D) machine vision systems combining 2D color or spectral data in short-wave up to thermal infrared wavelength ranges, have shown promising results in agriculture. Particularly, the crop water stress index (CWSI) has been widely used as a remote sensing method for water-deficit stress detection of entire canopies by means of thermal imaging [17,18]. CWSI is interesting due to the close relationship between canopy transpiration rate and surface temperature [19,20]. CWSI has, e.g., been used for large-scale monitoring via airborne systems, which made it suitable for real scenario application in commercial orchards [21]. On-the-ground and airborne canopy temperature measurements evaluated water deficits in peach trees under different irrigation strategies by means of CWSI [22–24]. CWSI calculated from both approaches was highly correlated with stem water potential, while canopy temperature closely reflected stomatal conductance. Most orchards, however, are too small for satellite measurements and provide mixed pixels of plant rows and ways disabling information gaining on the canopy. Therefore, unmanned aerial vehicles (UAVs) and ground-based thermal imaging platforms have been widely used for crop water-deficit stress detection [25,26]. Among others, CWSI was applied in peach to monitor irrigation requirements and identify areas sensitive to water deficit under low-frequency irrigation systems [24,27,28].

For various tree crops, studies investigated the correlation between CWSI and plant physiological water-deficit stress parameters such as leaf and/or stem water potential and stomatal conductance or transpiration rate [22,29–33]. CWSI is inversely correlated to stem [34] and leaf water potential [27,35]. Moreover, leaf and canopy water status were investigated remotely by estimating CWSI from aerial thermal imagery in date-palm (Phoenix dactylifera L.) trees [36]. Imaging in optical red, green, and blue (RGB)ranges and thermal cameras were applied to assess the relationship between leaf water potential and CWSI, resulting in r2 of 0.67 and 0.47, respectively, in the shaded and the sunlit side of the canopy [37]. RGB imaging and infrared thermography were applied to evaluate CWSI of pistachio (Pistacia vera L.) trees using convolutional neural networks [38]. UAV-based thermal and multispectral images were employed to analyze CWSI and normalized difference vegetation index (NDVI) in mild and moderately stressed almond [Prunus dulcis Mill. (D.A. Web)] trees [39].

CWSI has been calculated based on empirical [40] or theoretical [41] approaches. Recently, these equations were developed further [20,42]. However, analysis of canopy CWSI is based on the water deficit-induced reduction of stomatal conductance and, thus, transpiration of leaves, but assuming that the leaf water status closely reflects the fruit water status would be speculative. Up to now, an index similar to CWSI was never used to analyze the actual fruit water status.

Furthermore, spatial analysis is still lacking considering fruit data [43]. Modeling approaches were undertaken using RGB and thermal imaging to estimate the fruit temperature in 2D [44]. Easier to access weather data were applied to estimate the mean fruit surface temperature, resulting in reasonable performance of root mean square error (RMSE) <2 °C [45]. However, the spatial distribution of fruit temperature was challenging with this approach. In recent years, light detection and ranging (LiDAR) technology has been extensively used for remote sensing in arable farming and forestry, due to its ability to provide high-resolution 3D geometric information of plants in field conditions as 3D point clouds [46–48]. Terrestrial LiDAR sensors have been employed to develop estimation methods for geometric fruit parameters [49,50] as well as to map flowers and fruit in almond orchards [51], Kang et al., 2022. The advancement in terrestrial LiDAR sensors has also enabled the acquisition of the intensity of backscattered reflection at each point measured. This provides additional information for the segmentation of fruit from the 3D point cloud of whole canopy [49,52] and for the quantification of chlorophyll content of individual fruit [53,54]. Combining LiDAR and thermal imaging is proposed in this study for providing geometric and thermal data for innovative 3D tree analysis at the fruit level.

In other disciplines such as architecture and robotics [55], LiDAR-based 3D point clouds have been annotated with temperature data from thermal cameras. Having the 3D fruit visualization by LiDAR at hand, the annotation with temperature data enables to analyze 4D point clouds of canopies and fruit. Recently, a terrestrial LiDAR laser scanner was coupled with a thermal camera for reconstructing the 3D thermal point cloud in avocados and apples [56,57]. In basic ecophysiological studies capturing the variation of growth factors occurring due to global warming, such plant phenotyping is a modern technology to capture data from many individual plants. While root phenotyping [58,59] and remote sensing of whole canopies [60] is well established, the temperature of fruit surfaces (FST) was rarely tackled [45]. However, FST represents an important variable when discussing fruit growth and risk of fruit damage. Particularly, the 3D distribution within the canopies would be valuable for precise crop management. Furthermore, from the 3D thermal point clouds, a new fruit water stress index (FWSI) could be derived, providing an informative, spatiotemporally resolved parameter for analyzing fruit.

Therefore, the objectives of the present study were (a) to derive and validate the fruit surface temperature in 3D from terrestrial mobile LiDAR sensor and thermal imaging, (b) to calculate and compare estimated and manually measured FWSI, and (c) to gain first results with the new method on the course of FWSI changes during the growth season.

Materials and Methods

Experimental layout

The experiment was conducted in the Field Lab for Digital Agriculture of Leibniz Institute for Agricultural Engineering and Bioeconomy (ATB), located in Potsdam-Marquardt, Germany (latitude: 52.466°N, longitude: 12.572°E), planted with Malus × domestica Borkh. “Gala-Brookfield” on M9 rootstock with 0.95-m distance between trees, trained as slender spindle with an average tree height of 2.8 m. Trees were statically supported by horizontally parallel wires. Measurements took place on 7 trees throughout the season midday at 67, 81, and 132 days after full bloom (DAFB67, DAFB81, and DAFB132, respectively) (Table 1). Diurnal measurements were carried out on 5 trees at DAFB152 and DAFB153 at 0700, 0800, 1000, 1200, 1300, and 1800, and DAFB166. However, measurements on DAFB166, collected at midday (n = 50), have been used exclusively to complete the seasonal fruit trend analysis (Table 1).

Table 1. Variable (Var) measured during seasonal fruit development or within 2 consecutive days as diel dataset with number of trees (Tree), number of samples per tree (Sample), and frequency (f) according to date in days after full bloom (DAFB) and total number of samples (n).

Var	Period	Descript.	Number	n	
TRef	Season	DAFB	53	67	81			132			166		
Tree		7	7			7			5		
Sample		6	6			6			10		
f		2	2			2			1	302	
Diel	DAFB							152	153			
Tree							5	5			
Sample							10	10		600	
f							6	6		(300 per day)	
TLeaf
gst	Season	DAFB	53	67	74	88	109	117	123	151	165		
Tree	6	6	6	6	6	6	6	6	6	138	
Sample	3	3	3	3	3	3	2	1	2	
f	1	1	1	1	1	1	1	1	1	
TLeaf
ΨLeaf	Diel	DAFB							152	153			
Tree							5	5			
Sample							3	3		150	
f							5	5		(75 per day)	
ΨLeaf
ΨStem	Season	DAFB		61	68	81	110		124				
Tree		3	3	3	3		3			30	
Sample		2	2	2	2		2			
f		1	1	1	1		1			
Fruit quality	Season	DAFB		67		81		132			166		
Tree		4		4		4			5	34	
Sample		2		2		2			2	
f		1		1		1			1	

Further manual measurements were carried out on the trees (Table 1). Weather data (Uniklima Vario, Toss, Potsdam, Germany) were obtained in the orchard at 15 -min intervals, recording air temperature (°C), relative air humidity (%), global normal irradiance (GNI; W m−2), and precipitation (mm). The water vapor partial pressure deficit (VPD, kPa) was calculated [61] as:VPD=1−RH/100×SVP(1)

withSVPPA=610.7×107.5T/237.3+T(2)

where RH is the relative humidity, SVP is the saturated vapor pressure, and T is the temperature (°C).

Remote sensing

The phenotyping system was set up on a circular conveyor platform in the experimental apple orchard after calibrations of sensor signal intensity and position. The platform used an electrical engine (DRN71, SEW Eurodrive, Germany) operating at 50 Hz and a stainless-steel chain with mechanical suspensions to support the plant sensors. A mobile 2D LiDAR sensor emitting at a wavelength of 905 nm (LMS-511, Sick AG, Waldkirch, Germany) was mounted vertically on a metal plate placed at one suspension at 0.7 m above the ground level. The LiDAR sensor had the configuration with 0.1667° angular resolution, 25-Hz scanning frequency, and 180° scanning angle. The data were georeferenced using a real-time kinematic global navigation satellite system (AgGPS 542, Trimble, Sunnyvale, CA, USA), while orientation information was acquired using an inertial measurement unit (MTi-G-710, XSENS, Enschede, Netherlands), both of which were arranged on the sensor metal plate. The orientation information had an RMSE of 0.25° for roll, pitch, and yaw according to the manufacturer.

Additionally, a thermal camera (A655sc, FLIR Systems Inc., MA, USA) was installed, positioned 0.2 m above the LiDAR sensor at the same suspension. The camera provided the spatial resolution of 640 × 480 pixels at 50 Hz and a spectral range from 7.5 to 14 μm, with an operational temperature range from −40 °C to 150 °C and a thermal resolution of <0.05 °C. A lens (T198065, Teledyne FLIR LLC, Wilsonville, USA) with focal length of 6.5 mm (diagonal 80o) was attached to the camera. The LiDAR sensor and thermal camera underwent intrinsic and extrinsic calibrations using a lightbulb pattern [57] for determining the geometric calibration of each sensor and of both sensors in relation to each other. Points from LiDAR sensor captured also with thermal camera were annotated with thermal information (°C), resulting in spatially resolved TRaw data of canopies. All raw data are available open access [62].

LiDAR data processing for tree and fruit segmentation

The apparent return signal strength intensity (RToF) was recorded at 905 nm for each point in the 3D point cloud. Calibration algorithm for RToF was obtained with checker board coated with white powder (BaSO4, CAS number: 7727–43-7, Merck, Darmstadt, Germany), serving as 100% reference and black staining (S black, Avian Technologies, New London, USA) for 0% reference [53].

The 3D point cloud data were generated in Matlab environment (Computer Vision Toolbox, 2018b, Mathworks, Natick, USA) for positioning and alignment of 2 tree sides, for the latter using iterative closest point (ICP) algorithm, as described earlier [63]. To obtain points per tree (PPT), trees were segmented based on their stem position and planting distance. The maximum laser points in bivariate point density histogram were assumed to represent the center of the canopy. This assumption was made for the present slender spindle tree training system. For each tree point cloud, the volume of cylinder with a diameter of 0.95 m around the estimated stem position coordinates was considered.

Position and shape of apples within the tree point clouds were determined based on the geometric features, linearity (L), curvature (C), and RToF, considering each point of the 3D tree point cloud [49]. The temperature information was not considered in the segmentation. In the 3D point clouds of trees, local neighbors were decomposed and eigenvalues were calculated (λ1, λ2, λ3). To distinguish the 3D points of woody parts (W) from leaves and fruit, the probability density function was used to gain the mode value of the distribution of LW, CW, and RToF,W, which were used as thresholds (Rth,W, Cth,W, and Lth,W). Points that met the criteria of LW ≤ Lth,W, Cth,W ≤ CW, and Rth,W ≤ RToF,W were segmented and subtracted from the total number of PPT. Similarly, threshold values of apple points in terms of C and reflected intensity (CA and RToF,A) were used. For example, C values closer to 100 indicated a higher likelihood for the shape of neighboring point’s appearance to be curved. Neighboring points, which met the criteria of Cth,A ≤ CA, and Rth,A ≤ RToF,A, were segmented and annotated as apple. Subsequently, a density-based scan algorithm (DBSCAN) was applied (a) to find point sets with the assumed diameter (obtained as mean value from manually measured diameter of fruit) found in the neighborhood and (b) the value 10 (based on preliminary results as the minimum number of neighbors). The mean surface temperature values of segmented apples (TRaw) were considered in the further calculation. All annotated point clouds are available open access [62].

Reference temperature

During the period of fruit development, the surface temperature of apples [7 trees × 6 fruit × 2 (left and right side of each fruit) × 3 dates results in n = 252] was manually measured (TRef) with an infrared thermometer (Microscanner D501, Exergen, Watertown, USA). Similarly, during the diurnal temperature measurements of TRef, apples (n = 50) were measured 6 times (0700, 0800, 1000, 1200, 1300, and 1800) data capturing 300 readings on each of the 2 consecutive days.

TRef was compared to the corresponding temperature TRaw derived from remote sensing approach. All datasets obtained over the season in the early afternoon (dataset from seasonal readings plus 50 fruits from diurnal course at 1300 on September 21) were used for building the calibration of fruit surface temperature and its cross-validation. For this purpose, the dataset (n = 302) was split in 80% (n = 241) for calibration and 20% (n = 61) for cross-validation in block-wise design for each measuring date.

To determine the local maximum (maxT) and minimum (minT) temperatures among the apples, 2 individual fruits were covered with Vaseline and soap mixed with water, respectively. Their fruit surface temperature was additionally measured as TRaw and TRef.

FWSI estimation

Temperature points of fruit surfaces obtained by means of temperature-annotated LiDAR point cloud were exploited to estimate FWSI by 3 alternative methods (Eqs. 3, 5, and 6).

FWSII was calculated according to the approach by Irmak et al. [64]:FWSIITEst−TminEstTa+5−TminEst(3)

where TEst represents the temperature for each point of apple surface and minTEst is the minimum fruit temperature of segmented point clouds, measured at each individual measuring day. Ta is the average air temperature (Eq. 4) plus 5 °C as proposed by Irmak et al. [64].Ta=∑n−5n−2Ti4(4)

where Ti is the mean air temperature in the time range n being the exact hour of measuring fruit temperature at each individual measuring day.

FWSIJ was calculated according to [65]:FWSIJ=TEst−TWEstTDEst−TWEst(5)

where WT is the temperature of fully transpiring fruit and DT is the temperature of nontranspiring, Vaseline-coated fruit measured at each individual measuring day.

FWSIN, called normalized fruit water stress index, was calculated as:FWSIN=TEst−TminEstTmaxEst−TminEst(6)

minTEst and maxTEst represent the minimum and maximum fruit temperature of the segmented point clouds, measured at each individual measuring day.

For gaining an insight in the performance of different FWSI approaches, ∆T (°C) has been employed according to Eq. 7.∆T=TEst−Ta(7)

where both actual fruit (TEst) and air temperature Ta (Eq. 4) were considered.

For calculating the error of the method, for each of the 3 approaches, either TEst from temperature-annotated 3D point cloud was inserted as stated in Eqs. 3 to 6 or the manually measured reference fruit surface temperature TRef was used.

Water potential

Leaf and stem water potential (Ψstem) were measured on 5 neighbor trees with a portable Scholander pressure chamber (Plant Water Status Console 3000, Soilmoisture Equipment Corp., Pullman, USA) capturing direct readings or shaded leaves (n = 3) sealed in plastic bags for 30 min. The temperature from 3 neighbor leaves was manually measured (TLeaf) during each ΨStem reading. Measurements were carried out on DAFB152 and DAFB153 at 0700, 0900, 1100, 1300, and 1700. An average of 4 measurements was taken for each individual tree sampled (n = 75; Table 1).

Dendrometer and leaf gas exchange

The maximum daily shrinkage of tree stem was recorded using dendrometers (DD-L, Ecomatic GmbH, Dachau, Germany), installed 70 cm above ground, 50 cm above tree’s grafting zone. Data were recorded with data logger (CR10X and AM416 multiplexer, Campbell Scientific, Logan, USA).

Light response of steady-state leaf gas exchange was measured on 3 mature leaves selected at bearing shoots, of 3 neighbor trees from other trees subjected to LiDAR analysis, with a portable gas exchange analyzer (LI-6400 XT with LI-6400-40 red/blue LED, LI-COR Inc., Lincoln, USA). At ambient leaf temperature, relative humidity, and constant CO2 mole fraction in the reference gas (400 μmol mol−1), analyses were performed at photosynthetic photon flux rate (PPFR) of 2,000, 250, 100, 50, 20, and 0 μmol m−2 s−1 with a minimum waiting time of 100 s before each measurement. Maximum quantum yield (maxα, mol mol−1) and the rate of light-saturated CO2 gas exchange (maxJCO2, μmol m−2 s−1) were considered [16].

Fruit quality

After each measuring date with sensor system, apple samples (n = 8) were collected for fruit characterization in the laboratory. Fruit diameter (D) was manually measured by means of a digital caliper gauge considering the mean diameter of 2 measurements taken equatorially with 90o difference. The resulting mean diameter was applied in fruit segmentation process. Fresh mass (FM) was measured gravimetrically. Soluble solid content (SSC) was measured on squeezed juice obtained during firmness test, collected with pipette, and analyzed by refractometry (Pal-1, Atago, Tokyo, Japan). Stress and strain curves were measured by Texture Analyzer (TA-XT Plus, Stable Micro Systems, Godalming, UK) using a convex probe with a diameter of 11.1 mm, at constant speed of 4 mm s−1 on a peeled area in the equatorial region. Fruit flesh firmness was analyzed as the force obtained at 8-mm depth of penetration, corrected by probe size.

The chlorophyll content of apple skin and hypodermis (2 mm thickness), capturing contents of chlorophyll_a and _b, and pheophytin_a, was analyzed after acetone/diethyl ether extraction [66]. Extracts were filtered using a glass frit (pore size 3) attached to a vacuum pump, and phase separation was carried out in separating funnel by adding distilled water, transferring nonpolar pigments such as chlorophylls to nonpolar diethyl ether phase. The absorbance spectrum of the nonpolar phase was recorded spectrophotometrically (Lambda 950, Perkin Elmer Inc., Waltham, USA). The standard spectra of 3 chlorophylls were considered in the iterative multiple linear regression spectral analysis [67].

Statistical analysis

Descriptive statistics were applied to all datasets capturing minimum, maximum, mean, and standard deviation. A regression analysis was performed to quantify linear and logarithmic relationships between the manual measurements and LiDAR-derived data over the growing stages, and RMSE, mean bias error, and coefficient of determination (r2) were calculated. Descriptive statistics and graph design were performed using Sigmaplot 14.5 (Systat Software Inc., San Jose, USA).

Results and Discussion

Fruit surface temperature estimation

Slender spindles form the major training system of apple trees in worldwide production, providing a 3D structure in which the fruits are more or less evenly distributed according to the success of the crop load management. Apple surface temperature was monitored employing the LiDAR sensor to gain geometric position data in the format of 3D point cloud of the trees. Furthermore, a thermal camera was used to annotate the points with temperature information. The extrinsic geometric calibration was applied to acquire the thermal 3D point cloud of trees (Fig. 1), revealing an RMSE of 1.82% pixel−1. The estimated temperature (TRaw) in the annotated 3D point cloud of the trees varied (Fig. 1), with lower temperatures in the upper parts of canopies at all measuring dates. Mean of TRaw of stem points considering all measuring dates was 20.6 ± 0.65 °C standard deviation. The segmented leaf area showed a mean value of TRaw = 18.3 ± 1.4 °C.

Fig. 1. 3D spatial temperature distribution in the canopy analyzed by means of LiDAR and thermal sensors at DAFB153.

It was often assumed that fruit surface T is equal to air T, which was found to be untrue when analyzing the spatial distribution of apple temperature by means of few thermocouple readings and modeling the fruit temperature [68]. In the present study, application of the fruit segmentation routine [49] allowed to exclude leaf area and wood structure and segment the temperature-annotated point clouds of all fruit in the present study. The LiDAR system was used as a pilot sensor to segment the fruit point clouds from the overall canopy point clouds, revealing F1 scores of 80.7, 85.6, 90.2, and 91.4 for the measuring dates at DAFB67, DAFB81, DAFB132, and DAFB166, respectively. The bigger fruit at later growth stages provided better results as expected according to previous results [49]. The segmented, annotated fruit point clouds (Fig. S2) obtained were processed further as temperature means per fruit.

Throughout the seasonal measurements, fruit temperature estimated by means of T-annotated 3D point clouds (TRaw) ranged from 7.55 to 40 °C with a mean of 23.4 °C, while TRef measured manually on apple surfaces varied between 8.94 and 43.18 °C. Considering all measuring days, mean TRef was 24.3 °C. TRaw revealed the highest mean value of 25.41 ± 5.34 °C at DAFB67 beginning of July and the lowest mean value of 11.14 ± 0.94 °C at DAFB166 after harvest. For the same dates, TRef showed slightly enhanced values of 31.85 ± 2.55 °C and 11.43 ± 0.60 °C, respectively. Moreover, TRaw depicted lower minimum temperatures on DAFB67 (15.42 °C) and DAFB166 (8.94 °C), in contrast to reference measurements 24.67 and 10.37 °C, respectively. Similarly, TRaw measurements noted enhanced maximum values found between TRaw and TRef, revealing 43.18 and 34.76 °C at DAFB67 and 14.49 and 12.05 °C at DAFB166, indicating potential overestimation. However, the 2 variables were highly and linearly correlated (r2 = 0.91; RMSE = 2.54%; Table 2 and Fig. 2).

Table 2. Descriptive statistics of the relationship between fruit temperatures estimated from LiDAR and thermal sensors (TRaw) data and those measured manually (TRef). Given are the coefficients of determination for calibration (ΤEst) and cross-validation (TVal), root mean square error (RMSE), and bias. For calibration, 80% of data (n = 241) were used, whereas the remaining 20% (n = 61) were analyzed in cross-validation.

	Min	Max	Mean	Bias (°C)	r 2	RMSE (%)	
T Ref	8.94	43.18	24.30				
T Raw	7.55	40.00	23.40	−0.77	0.91	2.54	
T Est	8.94	43.18	22.87	−0.99	0.93	1.59	
T Val	10.37	41.10	24.14	−1.23	1.00	1.91	

Fig. 2. Scatterplots of (A) raw LiDAR-derived temperature (TRaw) (n = 302), (B) estimated from linear regression (n = 241) (TEst) and (C) cross-validation results (TVal) (n = 61). All measuring dates were included and marked in different colors.

TRaw data were calibrated based on TRef using 80% of the fruit data to achieve TEst. The robustness of calibration was assessed by means of cross-validation considering 20% of fruit data. The measuring uncertainty of calibration was 1.59% with r2 = 0.93, which was confirmed by slightly enhanced coefficient of determination appearing in the smaller dataset used in the cross-validation. The mean bias error increased 0.24 °C and RMSE by 0.32 °C, comparing calibration (Fig. 2B) and cross-validation results (Table 2 and Fig. 2C). To further enhance the robustness and generalizability of these findings, future studies could benefit from increasing both the sample size and the number of measurement days. This expansion would allow for more comprehensive validation and spatial analysis across varying climate conditions.

Despite the high correlation, problems with sensor drift can lead to inaccurate temperature readings. However, the blockwise cross-validation, taking 80%/20% from each measuring date, showed no systematic error (Fig. 2). Environmental factors such as ambient conditions, including humidity, wind, and direct sunlight, and variations in surface emissivity, e.g., by free water on the surface, can affect sensor readings [69]. In the present study, again, the errors appeared without change in the bias over the season (Table 2 and Fig. 2). More importantly, manual temperature readings may have introduced errors, because the time gap between the remote and manual readings may have caused deviation. On the other hand, the measurements in remote mode may have caused issues. Occlusions, where parts of the fruit or relevant surfaces are blocked by leaves, branches, or other fruit, can result in incomplete data capture and contribute to errors in temperature estimation [52]. Furthermore, some pixels may point to the sky instead of the tree surface. Also, wind can affect the actual surface that is captured with both sensors. Finally, data processing algorithms and interpolation methods used to create 3D point clouds can also introduce errors if they do not accurately account for all variables [57].

Comparison of fruit water stress index approaches

Based on TEst, alternative fruit water stress indices (FWSI) were obtained (Table 2) with established equations for calculating the crop water-deficit stress index of canopies based on 2D thermal imaging. Variation of FWSIJ, calculated according to [65], was generally high, with FWSIJ,Est ranging from −2.81 to 6.06 and FWSIJ,Val from −1.69 to 5.39. The wet and dry references obviously failed in apple fruit. Similar results were found in leaves [70], showing that Vaseline may have different effects than simply hindering transpiration, whereas the soap solution frequently dried before the remote reading was finalized.

Additionally, the coating may have affected the LiDAR readings, since the bias and RMSE were high when comparing FWSIJ based on remote sensing and based on manually recorded fruit temperature. Since the error of temperature estimation was low in untreated apples, such influence can be assumed. Accurately measuring of the temperature on a fully transpiring (WT) and nontranspiring (DT) apple surface may be an arduous challenge due to the need to maintain the correspondent status uniformly within the fruit surface. Similarly, applying Vaseline on vine leaves to estimate the water stress index have shown that the method is far from practical for large-scale agriculture [71]. This can potentially lead in lack of uniformity of water stress index not only within the tree canopy but also on the fruit surface during the growing stages. Artificial reference surfaces, such as green hemispherical cellulose surfaces, obtained high correlations with DT and could potentially replace the coated Vaseline approach [72]. In the present study, measuring uncertainty was enhanced for FWSIJ,Est, as indicated by r2 = 0.66 and RMSE = 2.27% (Table 3 and Fig. 3B and E). RMSE values for FWSII,Est and FWSII,Val (Fig. 3A, C, D, and F) were lower (0.11% and 0.24%, respectively). Therefore, FWSIJ,Est was not considered in the later steps of this study.

Table 3. Coefficient of determination (r2), root mean square error (RMSE, %), and bias (%) of fruit water stress indices calculated from remotely measured, calibrated fruit temperature data and manual fruit temperature readings for FWSIEst considering the calibration dataset (n = 241) and for FWSIVal capturing the dataset, not used in the calibration, but cross-validation (n = 61).

	Min	Max	Mean	Bias (%)	RMSE (%)	r 2	
FWSI I,Est	−0.01	1.58	0.48	−0.08	0.11	0.99	
FWSI I,Val	0.00	1.98	0.50	−0.16	0.24	0.99	
FWSI N,Est	−0.01	1.00	0.44	0.01	0.02	0.99	
FWSI N,Val	0.00	1.00	0.46	−0.02	0.05	0.98	
FWSI J,Est	−2.81	6.06	0.53	−0.89	2.27	0.66	
FWSI J,Val	−1.69	5.39	0.91	−1.02	2.00	0.80	

Fig. 3. Fruit water stress index (FWSI) calculated by 3 approaches according to Jones [65], Irmak et al. [64], and normalized index considering a data split for calibrated fruit surface temperature data (80% of entire dataset, n = 241) shown in the left column (A to C) and results from the remaining 20% of dataset used for cross-validation (n = 61) in the right column (D to F).

Seasonal course of FWSI

Throughout the season, maximum air T reached 43 °C, whereas minimum air T was 12.6 °C. Global radiation and VPD were maximal 600 W m−2 and 3.2 kPa, measured at DAFB73 and DAFB86, respectively. Heatwaves occurred between measuring days 2 and 3, DAFB90 and DAFB110, with maximum values of air T and global radiation higher than 40 °C and 500 W m−2, respectively (Fig. S1). Accordingly, VPD reached a peak of 3.5 kPa at DAFB90. Fruit development took place according to the typical increase of fresh mass and changes in soluble solid content, and decrease in fruit flesh firmness (Table S1).

Cross-validation of fruit surface temperature estimation (Table 1) confirmed that, after calibration, remote sensing data provided accurate T data of the fruit surface and can be applied in fruit monitoring. Thus, this approach was applied on all fruit data, remotely measured throughout the season 2022 considering measurements in the early afternoon. Subsequently, FWSII according to approaches by Irmak et al. [64] and the normalized FWSIN were selected due to low measuring uncertainty when compared to the reference data. In comparison of both approaches, the mean FWSII,Est was slightly higher (0.52) than FWSIN,Est (0.38) at DAFB67. Both indices showed marginal differences between DAFB81 and DAFB132. FWSII,Est showed decreased values of 0.62 ± 0.1 (mean ± SD) on DAFB132 to 0.18 ± 0.02 on DAFB166 (Fig. 4 and Fig. S2). On the other hand, variability of FWSIN,Est was high on DAFB166, with similar means as on DAFB132. Specific impacts of development and physiological and physical properties of fruit may explain the distinct temporal variation of both indices. For example, the increased leaf temperature and reduced stomatal conductance, which was observed between DAFB81 and DAFB132, may have resulted in the increased FWSI (Fig. S5). In contrast to FWSIN,Est, FWSII,Est illustrated a similar pattern with the leaf temperature and higher variation due to the consideration of Ta in the equation. Therefore, FWSII,Est decreased during fruit development (Fig. 5A).

Fig. 4. Box plot (A) of FWSII,Est as derived according to Irmak et al. [64] and (B) of the normalized FWSI (FWSIN,Est) calculated from measured local maxTfruit and minTfruit in TEst data of fruit considering 42 apples per measuring date, during fruit development in days after full bloom.

Fig. 5. Relationships between the difference of air temperature and fruit surface temperature (ΔΤ) in relation to the fruit water stress indices (A) calculated according to Irmak et al. [64] with airT plus 5 °C (FWSII,Est) and (B) normalized on fruit surface temperature data (FWSIN,Est), analyzed on 4 days after full bloom (DAFB; 67, 81, 132, and 166) during fruit development.

On DAFB67, ΔΤ was high, as indicated by maxT = 14.1 °C and minT = −14.8 °C (Fig. 5). Values of FWSII,Est remained unchanged DAFB81 and DAFB132. The range of ΔΤ was reduced on DAFB166, varying from 5.0 to 9.8 °C. FWSII,Est (Fig. 4A) may be dominated by the difference of air temperature and FWSII,Est as shown earlier for CWSI considering leaves [73]. Consequently, FWSII,Est may reflect the impact of high air T and possibly risk of heat damage and/or need for irrigation water. However, more work is requested to evaluate FWSI in specific experiments. Considering the present seasonal data, in temperate climate, the correlation between ΔΤ and FWSII,Est points to a close relationship from DAFB67 to DAFB166 (Fig. 5A). Similar correlation revealed that with FWSIΝ,Est (Fig. 5B), however, the overall correlation was slightly weaker with 0.95 and 0.79 for FWSII,Est and FWSIΝ,Est, respectively.

On the other hand, it is well established that stomata of apples become closed lenticels with progressing fruit development, thus affecting transpiration rate of apple fruit [74–76]. The normalized FWSIN,Est appeared lower at the first measuring date compared to later measurements and may, therefore, reflect such development, resulting in enhanced FWSIN,Est values when transpiration rate was limited during fruit ripening (Fig. 4B). Again, the results provide a first view on the seasonal development of the index.

The relationship between ΔT and FWSI was not studied before. However, the difference between canopy temperature and Ta and its relationship with VPD has been investigated in several perennial trees. Enhanced and moderate correlations were observed in peach, nectarine, and orange, resulting in r2 of 0.70, 0.65, and 0.5, respectively [27,31,73]. In high-density olive orchards, r2 varied from 0.56 to 0.82 over the growing period [77]. Consistent with the previous observations made in several orchards that were adequately irrigated, no temperature difference or slightly cooler canopy temperature occurred considering measurements over several days [31,73,78]. Expanding data collection across different seasons and climatic conditions could provide a more comprehensive understanding of these dynamics, exploring the relationship between environmental factors and the FWSIEst behavior.

Diel course of FWSI

Diel course was measured on ripe fruit during the commercial harvest window of apples over 2 consecutive days. The high resolution of LiDAR sensor allowed the spatial analysis of the variability of FWSI within the tree canopy (Fig. 6). The FWSI of apples declined in upper parts of canopy, with highest values found between 2.3 and 3 m. In this parts of canopy, FWSII,Est ranged between 0.4 and 0.8 (Fig. 6A), while values of FWSIN,Est (Fig. 6B) spanned a slightly lower range (0.4 to 0.6). This finding needs to be evaluated further, as it may be due to methodological challenges, when measuring fruit slightly moving due to wind, and resulting lower temperature, when an object in the back or the sky was measured in the mixed pixel. As an additional factor, crop load influences the stomatal conductance of apple leaves, which is decreased by a reduction in fruit number [79]. For example, enhanced midday leaf CWSI values were observed in apple trees with no crop load, resulting in a smaller difference in canopy and air temperature [78]. However, due to exposure to global radiation, this would not mean that such exposed fruit are necessarily less affected by heat damage. However, it highlights the necessity of understanding the local variation in water relationships within the canopy as changes can have substantial effect on fruit quality and yield.

Fig. 6. Spatial distribution of (A) FWSII,Est and (B) FWSIN,Est within apple trees. Fruit temperatures necessary to estimate FWSI were measured remotely by means of T-annotated point clouds, presenting an example point cloud measured around noon of DAFB153.

CWSI and FWSI showed no correlation (Fig. S3). The dataset acquired on last measuring date DAFB166 varied from other measuring dates, since the first 3 measurements during the season were taken at 1500, while the last one was conducted at 1300, when the fruits were still colder from the night time. Fruit temperature (TEst) followed the diurnal course of air temperature (Ta) with a certain lag time (Fig. 7). On both measuring days, average Ta was approximately 23.5% and 28.6% higher during dawn (0700) compared to TEst. Moreover, variation of TEst was low at this time, since fruit reached equilibrium with Ta during the night. The difference between Ta and TEst was lowest around noon (1300). Only when Ta declined in the late afternoon (1800), fruit started to cool and ΔT became negative. Similar to fruit, temperature of leaves (TLeaf) was highest at 1800 on both days (Fig. S4A). At the same time, stem water potential (ΨStem) was lowest (Fig. S4B). Generally, variation of ΨLeaf increased after 11 h. A combination of midday stem water potential, predawn leaf water potential, and cumulative transpiration rate enabled the early prediction of water stressed in peach orchards [80].

Fig. 7. Diurnal courses of air (Ta) and fruit surface temperature (TEst) measured on 2 successive DAFB152 and DAFB153 (September 21 and 22).

Contrasting the strong dependence of TEst from Ta, the FWSI approaches appeared with different patterns. From 7 to 8 h, means of FWSII,Est rapidly increased from 0.15 (± 0.01) to 0.45 (± 0.03) and then only gradually to 0.62 (±) until 13 h (Fig. 8A). In the late afternoon (18 h), value and variation of FWSII,Est were high, neither following Ta nor VPD due to fruit mass and its temperature holding capacity serving as a buffer. A similar pattern was noticed on DAFB153. On that day, values of FWSII,Est again doubled within an hour, revealing 0.3 at 36 h and 0.6 at 37 h. On the other hand, means of FWSIN,Est showed less pronounced differences between 7 and 8 h and throughout the day (Fig. 8B). Variation of FWSIN,Est was only slightly enhanced on DAFB153, fluctuating between 0.14 and 0.62 without clear patterns.

Fig. 8. Diurnal courses of the water vapor pressure deficit (VPD) and FWSI calculated (A) according to CWSI proposed by Irmak et al. [64] (FWSII,Est) and (B) by normalized index (FWSIN,Est).

So far, the principle of water-deficit stress indices was developed under the concept of the canopy temperature and leaf transpiration, which are sensitive to air temperature and VPD. With the present approach, the variation of fruit surface temperature can be monitored spatially with annotated point clouds, potentially providing new insight at the yield-relevant fruit scale. Similar approaches have been used to model fruit temperature at different locations [68], but relying on a model approach, while the method proposed in the present study would enable modeling work based on empirical analysis. However, the accuracy of noninvasive remote sensing techniques, such as LiDAR laser scanning in the field or orchard, is often compromised by occlusions and the presence of coinciding leaf surfaces. This issue is prevalent across various plant species as highlighted in recent studies by Deery et al. [46] and Keller et al. [81]. Therefore, potential differences between manual and LiDAR estimated parameters may appear to a varying extent in different tree architectures. Future studies should target the FWSI measurement in various production systems.

Overall, FWSI can be gained as a spatially and temporally resolved variable, enhancing the opportunities to analyze the fruit response to varying climate conditions. Furthermore, the efficiency and number of fruit analyzed could be enhanced by the remote sensing approach. This may increase the accuracy of analysis and modeling approaches through detailed spatiotemporal data.

The implementation of such remotely obtained plant data in existing agricultural systems can be streamlined due to its compatibility with remote sensing platforms currently developed for large-scale operations. This is crucial in large orchards and fields where manual monitoring of each individual plant or tree is impractical, while the low requirement for manual intervention makes it particularly suitable for extensive areas, where traditional monitoring methods may be less efficient or cost-prohibitive [82]. The application of FWSIEst in real-world scenarios is supported by CWSI estimation at the canopy scale in pilot studies capturing several large orchards, but also by its feasibility in informing and optimizing irrigation scheduling [83]. Consequently, (i) more efficient water management practices should include FWSI at the crop scale to advance physiological models and (ii) also practical solutions can be effectively deployed in production systems based on the close-range remote sensing approach. Furthermore, the methodology can provide new insight in studies on sunburn damage [84].

Conclusion

T-annotated 3D point clouds of fruit were acquired seasonally and in diel course under field conditions. Calibrated TEst resulted in high coefficient of determination with reference measurements (r2 = 0.93). These thermal point clouds enabled the recording of spatiotemporally resolved water stress indices capturing leaves and newly introduced, considering segmented fruit data. The fruit water stress index showed reliable results, when calculated by means of air T plus 5 °C (FWSII,Est). The diurnal course of fruit temperature evaluated from thermal point cloud followed the air temperature, revealing similar temperature at midday for 2 successive days. FWSII,Est provided patterns in both seasonal and diurnal courses. Overall, this study proves the applicability of thermal point cloud for FWSI analysis and the quantification of the temporal and spatial variation of this parameter. LiDAR-derived data on FWSI in combination with weather data and leaf or stem water potential may find application in irrigation scheduling and allow further ecophysiological studies on fruit response to varying climate conditions.

Acknowledgments

Funding: The project SHEET is part of the ERA-NET Cofund ICT-AGRI-FOOD, with funding provided by national sources (BLE) and co-funding by the European Union’s Horizon 2020 research and innovation program, grant agreement number 862665.

Project IRRIWELL (A novel plant-based approach to estimate irrigation water needs and the application of optimal deficit strategy, grant number 01DH21016, 2021-2024) funded the investigation through a PRIMA initiative of Members States, Associated Countries, and Participating Countries. German funding source: BMWK (Bundesministerium für Wirtschaft und Klimaschutz), Berlin; German funding agency: DLR (Deutsche Zentrum für Luft- und Raumfahrt e. V.), Bonn.

Author contributions: All authors reviewed the manuscript.

Competing interests: The authors declare that they have no competing interests.

Data Availability

Data underlying the results presented in this paper are available open access [62], with data in the supplement available from the authors upon request.

Supplementary Materials

Supplementary 1 Figs. S1 to S6

Tables S1 and S2
==== Refs
References

1. Dietz KJ, Zörb C, Geilfus CM. Drought and crop yield. Plant Biol. 2021;23 (6 ):881–893.34396653
2. Bahat I, Netzer Y, Grünzweig JM, Alchanatis V, Peeters A, Goldshtein E, Ohana-Levi N, Ben-Gal A, Cohen Y. In-season interactions between vine vigor, water status and wine quality in terrain-based management-zones in a ‘Cabernet Sauvignon’ vineyard. Remote Sens. 2021;13 (9 ):1636.
3. Fernandes RDM, Cuevas MV, Diaz-Espejo A, Hernandez-Santana V. Effects of water stress on fruit growth and water relations between fruits and leaves in a hedgerow olive orchard. Agric Water Manag. 2018;210 :32–40.
4. Measham PF, Wilson SJ, Gracie AJ, Bound SA. Tree water relations: Flow and fruit. Agric Water Manag. 2014;137 :59–67.
5. Munné-Bosch S, Vincent C. Physiological mechanisms underlying fruit sunburn. Crit Rev Plant Sci. 2019;38 :140–157.
6. Zude-Sasse M, Fountas S, Gemtos TA, Abu-Khalaf N. Applications of precision agriculture in horticultural crops—Review. Eur J Hortic Sci. 2016;81 :78–90.
7. Khosravi A, Zucchini M, Mancini A, Neri D. Continuous third phase fruit monitoring in olive with regulated deficit irrigation to set a quantitative index of water stress. Horticulturae. 2022;8 (12 ):1221.
8. Stagakis S, González-Dugo V, Cid P, Guillén-Climent ML, Zarco-Tejada PJ. Monitoring water stress and fruit quality in an orange orchard under regulated deficit irrigation using narrow-band structural and physiological remote sensing indices. ISPRS J Photogramm Remote Sens. 2012;71 (3 ):47–61.
9. Wulfsohn D, Aravena Zamora F, Potin Téllez C, Zamora Lagos I, García-Fiñana M. Multilevel systematic sampling to estimate total fruit number for yield forecasts. Precis Agric. 2012;13 :256–275.
10. Marino G, Scalisi A, Guzmán-Delgado P, Caruso T, Marra FP, Lo Bianco R. Detecting mild water stress in olive with multiple plant-based continuous sensors. Plan Theory. 2021;10 (1 ):131.
11. Scalisi A, Marino G, Marra FP, Caruso T, Lo Bianco R. A cultivar-sensitive approach for the continuous monitoring of olive (Olea europaea L.) tree water status by fruit and leaf sensing. Front Plant Sci. 2020;11 :340.32265975
12. Scalisi A, O’Connell MG, Stefanelli D, Lo Bianco R. Fruit and leaf sensing for continuous detection of nectarine water status. Front Plant Sci. 2019;10 :805.31333685
13. Boini A, Manfrini L, Bortolotti G, Corelli-Grappadelli L, Morandi B. Monitoring fruit daily growth indicates the onset of mild drought stress in apple. Sci Hortic. 2019;256 :108520.
14. Grilo FS, Scalisi A, Pernice F, Morandi B, Lo Bianco R. Recurrent deficit irrigation and fruit harvest affect tree water relations and fruitlet growth in ‘Valencia’ orange. Eur J Hortic Sci. 2019;84 (3 ):177–187.
15. Fernández JE. Plant-based sensing to monitor water stress: Applicability to commercial orchards. Agric Water Manag. 2014;142 (4 ):99–109.
16. Matyssek R, Herppich WB. Experimentelle pflanzenökologie. Berlin (Germany): Springer Spektrum; 2020.
17. Gerhards M, Schlerf M, Mallick K, Udelhoven T. Challenges and future perspectives of multi-/hyperspectral thermal infrared remote sensing for crop water-stress detection: A review. Remote Sens. 2019;11 (10 ):1240.
18. Meron M, Tsipris J, Orlov V, Alchanatis V, Cohen Y. Crop water stress mapping for site-specific irrigation by thermal imagery and artificial reference surfaces. Precis Agric. 2010;11 :148–162.
19. Larcher W, Wagner J. Temperaturgrenzen der CO2-Aufnahme und Temperaturresistenz der Blätter von Gebirgspflanzen im vegetationsaktiven Zustand. Oecol Plant. 1976;11 :361–374.
20. Sánchez-Piñero M, Martín-Palomo MJ, Andreu L, Moriana A, Corell M. Evaluation of a simplified methodology to estimate the CWSI in olive orchards. Agric Water Manag. 2022;269 :107729.
21. Zare M, Drastig K, Zude-Sasse M. Tree water status in apple orchards measured by means of land surface temperature and vegetation index (LST–NDVI) trapezoidal space derived from Landsat 8 satellite images. Sustain For. 2020;12 :70.
22. Agam N, Cohen Y, Berni JAJ, Alchanatis V, Kool D, Dag A, Yermiyahu U, Ben-Gal A. An insight to the performance of crop water stress index for olive trees. Agric Water Manag. 2013b;118 :79–86.
23. Wang D, Gartung J. Infrared canopy temperature of early-ripening peach trees under postharvest deficit irrigation. Agric Water Manag. 2010;97 (11 ):1787–1794.
24. Ramírez-Cuesta JM, Ortuño MF, Gonzalez-Dugo V, Zarco-Tejada PJ, Parra M, Rubio-Asensio JS, Intrigliolo DS. Assessment of peach trees water status and leaf gas exchange using on-the-ground versus airborne-based thermal imagery. Agric Water Manag. 2022;267 : Article 107628.
25. Beauchêne K, Leroy F, Fournier A, Huet C, Bonnefoy M, Lorgeou J, de Solan B, Piquemal B, Thomas S, Cohan JP. Management and characterization of abiotic stress via PhénoField®, a high-throughput field phenotyping platform. Front Plant Sci. 2019;10 :904.31379897
26. Käthner J, Ben-Gal A, Gebbers R, Peeters A, Herppich WB, Zude-Sasse M. Evaluating spatially resolved influence of soil and tree water status on quality of European plum grown in semi-humid climate. Front Plant Sci. 2017;8 :1053.28676810
27. Bellvert J, Marsal J, Girona J, Gonzalez-Dugo V, Fereres E, Ustin SL, Zarco-Tejada PJ. Airborne thermal imagery to detect the seasonal evolution of crop water status in peach, nectarine and Saturn peach orchards. Remote Sens. 2016;8 (1 ):39.
28. Blanco V, Willsea N, Campbell T, Howe O, Kalcsits L. Combining thermal imaging and soil water content sensors to assess tree water status in pear trees. Front Plant Sci. 2023;14 :1197437.37346137
29. Agam N, Cohen Y, Alchanatis V, Ben-Gal A. How sensitive is the CWSI to changes in solar radiation? Int J Remote Sens. 2013a;34 (17 ):6109–6120.
30. Berni JAJ, Zarco-Tejada PJ, Sepulcre-Cantó G, Fereres E, Villalobos F. Mapping canopy conductance and CWSI in olive orchards using high resolution thermal remote sensing imagery. Remote Sens Environ. 2009;113 :2380–2388.
31. Testi L, Goldhamer DA, Iniesta F, Salinas M. Crop water stress index is a sensitive water stress indicator in pistachio trees. Irrig Sci. 2008;26 (5 ):395–405.
32. Zhou Z, Majeed Y, Naranjo GD, Gambacorta EM. Assessment for crop water stress with infrared thermal imagery in precision agriculture: A review and future prospects for deep learning applications. Comput Electron Agric. 2021;182 :106019.
33. Gonzalez-Dugo V, Zarco-Tejada PJ, Intrigliolo DS, Ramírez-Cuesta JM. Normalization of the crop water stress index to assess the within-field spatial variability of water stress sensitivity. Precis Agric. 2021;22 :964–983.
34. Sepúlveda-Reyes D, Ingram B, Bardeen M, Zúñiga M, Ortega-Farías S, Poblete-Echeverría C. Selecting canopy zones and thresholding approaches to assess grapevine water status by using aerial and ground-based thermal imaging. Remote Sens. 2016;8 (10 ):822.
35. Cohen Y, Alchanatis V, Meron M, Saranga Y, Tsipris J. Estimation of leaf water potential by thermal imagery and spatial analysis. J Exp Bot. 2005;56 (417 ):1843–1852.15897226
36. Cohen Y, Alchanatis V, Prigojin A, Levi A, Soroker V. Use of aerial thermal imaging to estimate water status of palm trees. Precis Agric. 2012;13 (1 ):123–140.
37. Zhou Z, Diverres G, Kang C, Thapa S, Karkee M, Zhang Q, Keller M. Ground-based thermal imaging for assessing crop water status in grapevines over a growing season. Agronomy. 2022;12 (2 ):322.
38. Pantelidakis M, Panagopoulos AA, Mykoniatis K, Ashkan S, Eravi RC, Pamula V, et al. Identifying sunlit leaves using convolutional neural networks: An expert system for measuring the crop water stress index of pistachio trees. Expert Syst Appl. 2022;209 (1 ):118326.
39. Gutiérrez-Gordillo S, de la Gala González-Santiago J, Trigo-Córdoba E, Rubio-Casal AE, García-Tejero IF, Egea G. Monitoring of emerging water stress situations by thermal and vegetation indices in different almond cultivars. Agronomy. 2021;11 (7 ):1419.
40. Idso SB, Jackson RD, Pinter PJ, Reginato RJ, Hatfield JL. Normalizing the stress-degree-day parameter for environmental variability. Agric Meteorol. 1981;24 :45–55.
41. Jackson RD, Idso SB, Reginato RJE, Pinter PJ. Canopy temperature as a crop water stress indicator. Water Resour Res. 1981;17 :1133–1138.
42. Osroosh Y, Peters RT, Campbell CS, Zhang Q. Automatic irrigation scheduling of apple trees using theoretical crop water stress index with an innovative dynamic threshold. Comput Electron Agric. 2015;118 :193–203.
43. Anderson NT, Walsh KB, Wulfsohn D. Technologies for forecasting tree fruit load and harvest timing—From ground, sky and time. Agronomy. 2021;11 (7 ):1409.
44. Chandel AK, Khot LR, Osroosh Y, Peters TR. Thermal-RGB imager derived in-field apple surface temperature estimates for sunburn management. Agric For Meteorol. 2018;253 :132–140.
45. Li L, Peters T, Zhang Q, Zhang J, Huang D. Modeling apple surface temperature dynamics based on weather data. Sensors. 2014;14 :20217–20234.25350507
46. Deery DM, Smith DJ, Davy R, Jimenez-Berni JA, Rebetzke GJ, James RA. Impact of varying light and dew on ground cover estimates from active NDVI, RGB, and LiDAR. Plant Phenomics. 2021.
47. Höfle B, Hollaus M, Hagenauer J. Urban vegetation detection using radiometrically calibrated small-footprint full-waveform airborne LiDAR data. ISPRS J Photogramm Remote Sens. 2012;67 :134–147.
48. Yun T, Li J, Ma L, Zhou J, Wang R, Eichhorn MP, Zhang H. Status, advancements and prospects of deep learning methods applied in forest studies. Int J Appl Earth Obs Geoinf. 2024;131 :103938.
49. Tsoulias N, Paraforos DS, Xanthopoulos G, Zude-Sasse M. Apple shape detection based on geometric and radiometric features using a LiDAR laser scanner. Remote Sens. 2020;12 (15 ):2481.
50. Kang H, Wang X, Chen C. Accurate fruit localisation using high resolution LiDAR-camera fusion and instance segmentation. Comput Electron Agric. 2022;203 (10 ):107450.
51. Underwood JP, Hung C, Whelan B, Sukkarieh S. Mapping almond orchard canopy volume, flowers, fruit and yield using lidar and vision sensors. Comput Electron Agric. 2016;130 :83–96.
52. Gené-Mola J, Gregorio E, Guevara J, Auat F, Sanz-Cortiella R, Escolà A, Calveras JL, JRM R, Ruiz-Hidalgo J, Vilaplana V, et al. Fruit detection in an apple orchard using a mobile terrestrial laser scanner. Biosyst Eng. 2019;187 (34 ):171–184.
53. Saha KK, Zude-Sasse M. Estimation of chlorophyll content in banana during shelf life using LiDAR laser scanner. Postharvest Biol Technol. 2022;192 (2022 ): Article 112011.
54. Tsoulias N, Saha KK, Zude-Sasse M. In-situ fruit analysis by means of LiDAR 3D point cloud of normalized difference vegetation index (NDVI). Comput Electron Agric. 2023;205 (4 ):107611.
55. Borrmann D, Elseberg J, Nüchter A. Thermal 3D mapping of building façades. Adv Intell Syst Comput. 2013;193 :173–182.
56. Narváez FJY, del Pedregal JS, Prieto PA, Torres-Torriti M, Cheein FAA. LiDAR and thermal images fusion for ground-based 3D characterisation of fruit trees. Biosyst Eng. 2016;151 :479–494.
57. Tsoulias N, Jörissen S, Nüchter A. An approach for monitoring temperature on fruit surface by means of thermal point cloud. MethodsX. 2022;9 :101712.35572811
58. Klein SP, Schneider HM, Perkins AC, Brown KM, Lynch JP. Multiple integrated root phenotypes are associated with improved drought tolerance. Plant Physiol. 2020;183 :1011–1025.32332090
59. Müllers Y, Postma JA, Poorter H, van Dusschoten D. Stomatal conductance tracks soil-to-leaf hydraulic conductance in faba bean and maize during soil drying. Plant Physiol. 2022;190 :2279–2294.36099023
60. Cotrozzi L, Peron R, Tuinstra MR, Mickelbart MV, Couture JJ. Spectral phenotyping of physiological and anatomical leaf traits related with maize water status. Plant Physiol. 2020;184 :1363–1377.32907885
61. Monteith J, Unsworth M. Principles of environmental physics: Plants, animals, and the atmosphere Cambridge (MA): Academic Press; 2013.
62. Zude-Sasse M, Jörissen S, Bignardi M, Regen C, Tsoulias N. Fruit surface temperature data analysed by means of temperature-annotated 3D point clouds of apple fruit trees. Data Brief. 2024;55 : Article 110762.39149715
63. Tsoulias N, Paraforos DS, Fountas S, Zude-Sasse M. Estimating canopy parameters based on the stem position in apple trees using a 2D LiDAR. Agronomy. 2019;9 :740.
64. Irmak S, Haman D, Bastug R. Determination of crop water stress index for irrigation timing and yield estimation of corn. Agron J. 2000;92 :1221–1227.
65. Jones H. Plants and microclimate, a quantitative approach to enviromental plant physiology. 2nd and 3rd ed. Melbourne: Cambridge University Press; 1992.
66. Wellburn AR. The spectral determination of chlorophylls a and b, as well as total carotenoids, using various solvents with spectrophotometers of different resolution. J Plant Physiol. 1994;144 (3 ):307–313.
67. Pflanz M, Zude M. Spectrophotometric analyses of chlorophyll and single carotenoids during fruit development of tomato (Solanum lycopersicum L.) by means of iterative multiple linear regression analysis. Appl Opt. 2008;47 :5961–5970.19002219
68. Saudreau M, Marquier A, Adam B, Monney P, Sinoquet H. Experimental study of fruit temperature dynamics within apple tree crowns. Agric For Meteorol. 2009;149 (2 ):362–372.
69. Meron M, Sprintsin M, Tsipris J, Alchanatis V, Cohen Y. Foliage temperature extraction from thermal imagery for crop water stress determination. Precis Agric. 2013;14 (5 ):467–477.
70. Park S, Ryu D, Fuentes S, Chung H, Hernández-Montes E, O’Connell M. Adaptive estimation of crop water stress in nectarine and peach orchards using high-resolution imagery from an unmanned aerial vehicle (UAV). Remote Sens. 2017;9 (8 ):828.
71. Belfiore N, Vinti R, Lovat L, Chitarra W, Tomasi D, de Bei R, et al. Infrared thermography to estimate vine water status: Optimizing canopy measurements and thermal indices for the varieties Merlot and Moscato in northern Italy. Agronomy. 2019;9 (12 ):821.
72. Apolo-Apolo OE, Martínez-Guanter J, Pérez-Ruiz M, Egea G. Design and assessment of new artificial reference surfaces for real time monitoring of crop water stress index in maize. Agric Water Manag. 2020;240 :106304.
73. Gonzalez-Dugo V, Zarco-Tejada PJ, Fereres E. Applicability and limitations of using the crop water stress index as an indicator of water deficits in citrus orchards. Agric For Meteorol. 2014;198 :94–104.
74. Bishnu PK, Yiru S, Knoche M. Lenticels and apple fruit transpiration. Postharvest Biol Technol. 2020;167 :111221.
75. Pieniazek SA. Physical characters of the skin in relation to apple fruit transpiration. Plant Physiol. 1944;19 (3 ):529–536.16653934
76. Singh V, Gamrasni D, Arie RB, Naschitz S, Friedman H. Identification of open lenticels in apples after harvest in relation to lenticel breakdown development during storage. Postharvest Biol Technol. 2016;121 :165–170.
77. Egea G, Padilla-Díaz CM, Martinez-Guanter J, Fernández JE, Pérez-Ruiz M. Assessing a crop water stress index derived from aerial thermal imaging and infrared thermometry in super-high density olive orchards. Agric Water Manag. 2017;187 :210–221.
78. Osroosh Y, Peters RT, Campbell CS. Daylight crop water stress index for continuous monitoring of water status in apple trees. Irrig Sci. 2016;34 (3 ):209–219.
79. Wünsche JN, Palmer JW, Greer DH. Effects of crop load on fruiting and gas-exchange characteristics of Braeburn'/M. 26 apple trees at full canopy. J Am Soc Hortic Sci. 2000;125 (1 ):93–99.
80. Remorini D, Massai R. Comparison of water status indicators for young peach trees. Irrig Sci. 2003;22 :39–46.
81. Keller B, Zimmermann L, Rascher U, Matsubara S, Steier A, Muller O. Toward predicting photosynthetic efficiency and biomass gain in crop genotypes over a field season. Plant Physiol. 2022;188 (1 ):301–317.34662428
82. Hunt ER Jr, Daughtry CS. What good are unmanned aircraft systems for agricultural remote sensing and precision agriculture? Int J Remote Sens. 2018;39 (15-16 ):5345–5376.
83. Ben-Gal A, Agam N, Alchanatis V, Cohen Y, Yermiyahu U, Zipori I, Presnov E, Sprintsin M, Dag A. Evaluating water stress in irrigated olives: Correlation of soil water status, tree water status, and thermal imagery. Irrig Sci. 2009;27 :367–376.
84. Torres CA, Sepúlveda A, Leon L, Yuri JA. Early detection of sun injury on apples (Malus domestica Borkh.) through the use of crop water stress index and chlorophyll fluorescence. Sci Hortic. 2016;211 :336–342.
