Rapid determination of leaf area and plant height by using light curtain arrays in four species with contrasting shoot architecture
© Fanourakis et al.; licensee BioMed Central Ltd. 2014
Received: 10 March 2014
Accepted: 4 April 2014
Published: 11 April 2014
Light curtain arrays (LC), a recently introduced phenotyping method, yield a binary data matrix from which a shoot silhouette is reconstructed. We addressed the accuracy and applicability of LC in assessing leaf area and maximum height (base to the highest leaf tip) in a phenotyping platform. LC were integrated to an automated routine for positioning, allowing in situ measurements. Two dicotyledonous (rapeseed, tomato) and two monocotyledonous (maize, barley) species with contrasting shoot architecture were investigated. To evaluate if averaging multiple view angles helps in resolving self-overlaps, we acquired a data set by rotating plants every 10° for 170°. To test how rapid these measurements can be without loss of information, we evaluated nine scanning speeds. Leaf area of overlapping plants was also estimated to assess the possibility to scale this method for plant stands.
The relation between measured and calculated maximum height was linear and nearly the same for all species. Linear relations were also found between plant leaf area and calculated pixel area. However, the regression slope was different between monocotyledonous and dicotyledonous species. Increasing the scanning speed stepwise from 0.9 to 23.4 m s−1 did not affect the estimation of maximum height. Instead, the calculated pixel area was inversely proportional to scanning speed. The estimation of plant leaf area by means of calculated pixel area became more accurate by averaging consecutive silhouettes and/or increasing the angle between them. Simulations showed that decreasing plant distance gradually from 20 to 0 cm, led to underestimation of plant leaf area owing to overlaps. This underestimation was more important for large plants of dicotyledonous species and for small plants of monocotyledonous ones.
LC offer an accurate estimation of plant leaf area and maximum height, while the number of consecutive silhouettes that needs to be averaged is species-dependent. A constant scanning speed is important for leaf area estimations by using LC. Simulations of the effect of varying plant spacing gave promising results for method application in sets of partly overlapping plants, which applies also to field conditions during and after canopy closure for crops sown in rows.
KeywordsLeaf area estimation Maximum height estimation Overlapping plants Phenotyping Sensor speed Sensor-to-plant system Shoot profile Shoot silhouette
The capacity to quantitatively explore plant phenotypes (from single cells to the whole plant) and their dynamic responses to a changing environment is a necessary requirement for genetic and physiological research by crop breeders, agricultural industry, and academia. Although molecular profiling technologies now enable the generation of a large amount of data with decreasing costs, the understanding of the link between genotype and phenotype still remains fragmented . Insufficient technical and conceptual capacity of the plant scientific community to probe existing genetic resources and unravel environmental effects limits faster progress in this field . To address this challenge, several automated phenotyping platforms have been developed from academia or commercial sources during the past decade (reviewed by [3–5]). The development of phenotyping applications for non-invasive assessment of the dynamics of plant biomass development is a cornerstone for this effort.
Phenotyping platforms may be operated using plant-to-sensor or sensor-to-plant measurement routines depending if plants are moved to the sensor or vice-versa, respectively. With the exception of small rosette plants [13–15], there are certain limitations in implementing imaging methods as a sensor-to-plant approach. Not only several cameras would need to be moved at a defined orientation above the plant, but also light conditions need to be strictly controlled during imaging. This need partly explains that in existing platforms, dedicated to phenotypic evaluation of plants of different sizes, a plant-to-sensor approach is implemented, where plants are moved (manually or automatically) to dedicated imaging stations (e.g. [6, 16, 17]). Contrary to this, LC can be more easily implemented using either plant-to-sensor or sensor-to-plant methodologies. For example, LC may offer an alternative in assessing large size plants (e.g., taller than 1 m) by employing different plant cultivation systems.
The aim of this study was to investigate whether or not LC can accurately estimate plant biomass in addition to maximum height (base to the highest leaf tip) in species with contrasting shoot architecture. To test the applicability of this method to field scale, we also estimated the biomass of plants spaced at different distances where increasing overlaps took place.
Results and discussion
A constant scanning speed is critical for estimating leaf area
The sensor positioning system was a prototype developed for shoot phenotyping. As the scanning speed determines the system throughput and the scalability of the approach to larger number of plants, we tested by recording plant silhouettes (also referred as profiles , see Figure 1B) whether or not it affects the acquired data in the speed range of 0.9 to 23.4 m min−1. The assessment was carried out in rapeseed plants covering a range of both leaf area (64–350 cm2) and maximum height (10–20 cm). Maximum height refers to the length from the base to the highest leaf tip (Figure 1C–E). We observed that calculated pixel area exponentially decreased as scanning speed increased (Additional file 1: Figure S1A). Instead, the calculated maximum height remained constant (Additional file 1: Figure S1B).
LC have been previously used to determine canopy height in the field [11, 12]. Here we show that for maximum height estimations by using LC, a wide range of constant (0.9–23.4 m min−1) or even varying scanning speeds can be used without any noticeable effects on the obtained data. Instead, a constant scanning speed is essential for estimating plant leaf area. Moreover, in this case, a speed-specific relation between calculated pixel area and measured leaf area needs to be obtained in each case.
Leaf area estimation by means of pixel area is improved as more silhouettes are averaged and/or the larger the view angle between them
Experiments were conducted in four species with contrasting shoot architecture, including two monocotyledonous (barley, maize) and two dicotyledonous (rapeseed, tomato) ones. Plants were grown from two days following transplanting up to four weeks. Because the information that can be extracted by a single silhouette may be limited due to overlapping (i.e. not visible at a given angle) plant parts, several silhouettes were recorded. At first, plant silhouette was determined at the starting position. The pot was then rotated by 10° with respect to the horizontal and the next plant silhouette was measured. Subsequently, the same procedure was repeated for 16 times, covering in total the 0 to 170° angle range.
The maximum deviation to the mean quantifies the extent to which the calculated value of a given silhouette deviates from the average value of 16 consecutive silhouettes and expresses the maximum error that can occur. In other words, this parameter expresses the percentage of the difference of the most distant value in each vertical cloud of points (see Figure 4A), representing a single plant, with respect to the average. Although the correlation coefficient between calculated plant pixel area and measured leaf area is very high (R2 ≥ 0.938) when taking into account only one silhouette, the maximum deviation to the mean in calculated pixel area of a single silhouette is 17 and 23% for rapeseed and maize, respectively (Figure 5C, D). The maximum deviation to the mean rapidly decreased as more consecutive silhouettes are taken into account and/or using larger angles between them (Figure 5C, D). For instance, the maximum deviation to the mean was about 3% when four profiles differing by 50° were averaged in both studied species. The same trend was observed when analysing tomato or barley (Additional file 2: Figure S2C, D).
Despite the high correlation coefficient (≈0.94, and 0.85 for barley) between calculated plant pixel area and measured leaf area in all four species when considering a single silhouette, we found that the maximum deviation to the mean (expressed as percentage) ranged between 17 and 29% (corresponding to rapeseed and barley, respectively). In other words, a single silhouette may result in up to 29% mis-estimation of plant pixel area, which will lead to a respective error in the calculation of leaf area. We found that this issue is readily solved by averaging an increasing number of consecutive profiles and/or using larger angles between them. We observed that the number of such consecutive profiles that needs to be averaged was not dependent on plant size within the measured range (an example is given in Additional file 4: Figure S4). For any given maximum deviation to the mean, the number of needed profiles at several angle options between them can be calculated. Plant rotation may be performed manually or automatically (by using a rotating stage). Another possibility is to use multiple pairs of LC that capture the plant profiles at different angles. By adopting this approach, several silhouettes at different angles can be recorded within a single routine that is amenable to automation.
When considering a single silhouette the correlation coefficient (approximately 0.97, and 0.83 for barley) between calculated and measured maximum height was higher in all four species compared with the one estimating leaf area. Consequently, the maximum error in the estimation of maximum height was comparatively lower (<10%, and 15% for barley) and it was reduced using the same approach, as mentioned above.
When scanning a row of (non-rotating) plants on a cultivation table in the greenhouse or in the field, one most probably will encounter a difference from the real total pixel area of these plants. Since this difference for each individual plant is sometimes positive and other times negative, most probably averaging a large number of plants will be closer to the real total pixel area than estimated by adding individual errors.
Calculated and measured values of both maximum height and leaf area were linearly related, the latter being sensitive to shoot architecture
Unlike leaf area, dicotyledonous and monocotyledonous species showed similar relation between calculated and measured maximum height (Figure 7B). In all cases, calculated maximum height was the same as the one measured manually with the exception of barley for which a slight underestimation was observed for plants around 30 cm tall.
The relation between calculated pixel area and measured leaf area apparently depends on shoot architecture. Monocotyledonous and dicotyledonous species showed distinct differences in this relation. Instead, maximum height estimations were not species-dependent. In barley, an underestimation of maximum height was observed. The pixels connecting plant parts are not always continuous (Figure 1C–E). In this species missing pixels for the very thin profile of leaf tips at the position where the maximum height is estimated may account for this underestimation.
The pixel area of several overlapping plants is indicative of total leaf area
At 12 cm spacing distance, the width of the overlapping area was generally limited (≤8 cm), resulting in a very small underestimation of total pixel area (≤5%; Additional file 6: Figure S6 and Additional file 7: Figure S7). Decreasing the plant distance from 8 cm to lower values resulted in a higher width of the overlapping area, leading to increased pixel area underestimation. At the same spacing distance, both the width of the overlapping area and the pixel area underestimation increased with plant size in rapeseed and tomato (Additional file 6: Figure S6 and Additional file 7: Figure S7). Instead, the pixel area underestimation was higher in smaller maize and barley plants.
Attempting to scale this method to a typical field plant array, one may (and will) encounter overlapping parts of neighbouring plants, reducing the accuracy of estimation. This depression of pixel area, as a result of overlaps, depends on spacing distance, plant size, and shoot geometry. Spacing distance 0 cm represents a situation in which plants are placed in two different rows behind each other, and the LC scan both rows simultaneously. Scanning two rows at the same time (i.e. spacing distance 0 cm) produced the largest underestimation in all four species. Against expectations, two plants behind each other resulted in underestimation which was far less than 100%, since their structures were never identical; meaning that in none of the different orientations of a plant covered the other plant. A very small spacing distance of 4 cm resulted in less underestimation of plant pixel area, as compared to 0 cm, but the magnitude was still considerable. At 0 or 4 cm spacing distances, the larger the plant, the higher was the underestimation of pixel area in rapeseed and tomato. Contrary to these findings, the highest underestimation was observed in small maize and barley plants. At more realistic spacing distances (≥8 cm) like those used in the field in row sowing, the underestimation was rather small in all cases for the range of plant size that we examined.
Light curtain arrays (LC) are a phenotyping tool yielding a binary data set of the shoot profile. It can be employed as either a sensor-to-plant (as in this study) or a plant-to-sensor system. Here we found that LC give a rapid estimate of leaf area and maximum height on individual plants of four species with contrasting shoot architecture. The accuracy of this estimation is improved as more plant profiles, following rotation, are averaged and/or the larger the angle between the acquired profiles. Leaf area estimation is strongly influenced by the scanning speed, whereas maximum height estimation is insensitive. Based on these experiments, we generally expect that up to early tillering stage for monocotyledonous species and until the 6th leaf stage in dicotyledonous ones, the estimation of leaf area can be sufficiently accurate. At later developmental stages, the current method may not be necessarily suitable for precise estimation of leaf area. However, regarding measurement of maximum height, LC are only limited by the vertical (and customizable) dimension of the sensor. By using LC the biomass of several plants placed in a row can be calculated, even when those partly overlap opening the window to additional field application.
Plant material and growth conditions
Experiments included two dicotyledonous (Brassica rapa cv. Campino, Lycopersicon esculentum cv. Harzfeuer) and two monocotyledonous (Zea mays cv. Badischer Gelber, Hordeum vulgare cv. Barke) species. Seeds were sown, germinated, and plants were grown up to the cotyledons stage (i.e. fully open cotyledons and before the appearance of the first leaf) or the second leaf stage for the dicotyledonous and monocotyledonous species, respectively. This period ranged from twelve to fourteen days, depending on the species. Subsequently, seedlings were transplanted in 2 L pots containing commercially available soil (‘Nullerde’ Archut Erzeugnisse GmbH, Vechta, Germany), which were randomly distributed over a glasshouse compartment, located in central part of Germany (Jülich, 50.9°N). The day and night set points were 19 and 17°C for air temperature, while for relative air humidity these were 60 and 50%, respectively. Supplementary light was provided by high-pressure sodium lamps (SON-T Agro, 250 W, Philips, Eindhoven, The Netherlands) at 50 μmol m−2 s−1 photosynthetic photon flux density for 16 h per day (from 0600 to 2200 hours). This light intensity was recorded during the nocturnal period at 50 cm from the root-to-shoot interface.
The plants were watered weekly with 100 ml of a full strength Hoagland solution . Two to three days following watering with Hoagland solution, plants were flushed with rainwater (approximately 10% drainage) to prevent salt accumulation. Measurements were conducted between two days and four weeks after transplanting.
Sensor positioning system
The sensor positioning system, employed for the experiment, was a prototype developed for phenotyping purposes. It consists of five laterally movable table elements and an overhead loadbearing system. The layout of the tables enables their movement into different arrangements in which they are closer or further apart, allowing an optimal use of the available space, as explained below. During operation, the tables are automatically arranged in a way that gaps on both sides of a given table are created. In this way, it is possible for the overhead loadbearing system to scan that table, carrying the plants designated for the measuring cycle. The remaining four tables are positioned without gaps. The same principle is used to enable access for the personnel. In this way the space, which would be required for passageways between conventional greenhouse tables, is made available for experimentation. The overhead load bearing system consists of an aluminum framework, designed to minimize shadowing. It has a payload capacity of 50 kg. It can move in both x and y directions (accuracy of ± 1 mm), as well as in the z direction (accuracy of ± 2 mm). The x and y movement is due to a laterally movable crossbeam, equipped with a longitudinal carriage rail. The movement in the z direction is enabled through two cantilever arms, equipped with carriage rails. The technical conception and realization was performed by a company (MK-Maschinenbau Kitz GmbH, Troisdorf, Germany).
The LC system is a parallel set of light barriers at a distance of 2.5 mm, which are arranged on two vertical poles (INFRASCAN 5000, Sitronic GmbH, Austria). The vertical support poles are 1.59 m long (1.43 m of which contain light barriers), spaced at a (horizontal) distance of 55 cm (Figure 1A). One set of light barriers is emitting infrared light (950 nm), while the other one is receiving the emitted light. In this way, the sensor detects if any of these beams are interrupted by an object placed between the emitting and receiving sides.
Before measurement, sixteen plants were placed in a row on a bench (Figure 1A). To minimize overlaps, pots were spaced at 30 cm. The light barriers were guided along the row of plants, in a way that the emitter was placed from one side of the measurement bench (perpendicular to the horizontal), while the receiver was placed on the other side (Figure 1A). The bottom edge of the light barriers was located at the bench level, allowing monitoring of both pots and plants. Care was taken that no plant parts were below the upper pot edge. All measurements took place at a constant sensor speed.
At first, the plant silhouette was recorded (Figure 1B). Subsequently, the plants were rotated by 10° to the horizontal, and another scan was conducted. In total 18 scans were acquired, starting from the initial position (0°) up to 170° with a 10° rotation for each run. Plants were not rotated by 180°, since this would result in the same silhouette as the initial position (0°). View angles larger than 180° were also not acquired, since the yielded data have been already recorded (e.g. 190° the same silhouette as 10°, 200° the same silhouette as 20°, etc.).
After all plant silhouettes were recorded, the number of leaves (≥1 cm), leaf area, leaf and plant (aboveground) dry mass were recorded. Leaf area was determined with a leaf area meter (LI-COR Model Li-3100, LI-COR, Lincoln, NE), and dry weight was assessed after drying the tissue for 24 h at 80°C.
Plant pixel area and maximum height estimation
The LC yield a binary data set, where the values 0 and 1 correspond to continuous or interrupting light path of a given photodiode. The methodology employed to extract a single shoot silhouette from a scan of several plants is described in Figure 2. At first, individual plant position was identified from the scans including several plants (Figure 1B). This was done by first defining the pot center. Secondly, small size plant parts (e.g. stems and petioles) that were not connected to the silhouette were included (see Figure 1C–E). To identify which data belong to a given plant, a radial search within a radius of 25 pixels was conducted, beginning at the bottom center of each pot position. In this way, all pixels belonging to that area are added to the silhouette. Finally, the pixels belonging to the pot were removed from the silhouette (Figure 1B). Starting at the bottom of the silhouette, we defined that the upper edge of the pot is at the position where the number of pixels in two consecutive rows differed more than 80%. With this approach, any plant parts below the upper edge of the pot are also removed. Plant pixel area was taken as the number of pixels belonging to a given silhouette, while maximum plant height was taken as the maximum pixel value in the y-axis (Figure 1C–E). The analysis was conducted by using the MATLAB program (R2012b, MathWorks, Natick, MA).
Simulation of spacing distance
To evaluate the potential of LC in assessing plant leaf area of overlapping plants, simulations were conducted at variable spacing distances. Two representative small, medium and large plants per species were selected. In each pair of plants, the spacing distance was decreased from 20 cm (where no overlaps occurred) down to 0 cm (maximum overlapping areas), with a step of 4 cm. In each spacing distance, simulations included all different angle combinations (18 × 18 view angles). In each spacing distance and angle combination, the pixel area of the two plants was assessed and this was compared to the sum of pixel areas of the two individual plants.
Light curtain arrays.
This research received the financial support of the Forschungszentrum Jülich in the Helmholtz Association. This work was partly funded by the European Plant Phenotyping Network (EPPN -European Union FP7 Capacities Programme) through a post-doctoral grant to DF (grant agreement 284443), by the German Plant Phenotyping Network (DPPN - BMBF Fz. 031A053), and by the AgroHort med project (EU fund for regional development, EU-EFRE-NRW AZ: B0833.01.09).
- Fiorani F, Schurr U: Future scenarios for plant phenotyping. Annu Rev Plant Biol. 2013, 64: 267-291. 10.1146/annurev-arplant-050312-120137.View ArticlePubMedGoogle Scholar
- Pieruschka R, Poorter H: Phenotyping plants: genes, phenes and machines. Funct Plant Biol. 2012, 39: 813-820. 10.1071/FPv39n11_IN.View ArticleGoogle Scholar
- Furbank RT, Tester M: Phenomics – technologies to relieve the phenotyping bottleneck. Trends Plant Sci. 2011, 16: 635-644. 10.1016/j.tplants.2011.09.005.View ArticlePubMedGoogle Scholar
- Araus JL, Cairns JE: Field high-throughput phenotyping: the new crop breeding frontier. Trends Plant Sci. 2014, 19: 52-61. 10.1016/j.tplants.2013.09.008.View ArticlePubMedGoogle Scholar
- Dhondt S, Wuyts N, Inzé D: Cell to whole-plant phenotyping: the best is yet to come. Trends Plant Sci. 2013, 18: 428-439. 10.1016/j.tplants.2013.04.008.View ArticlePubMedGoogle Scholar
- Golzarian MR, Frick RA, Rajendran K, Berger B, Roy S, Tester M, Lun DS: Accurate inference of shoot biomass from high-throughput images of cereal plants. Plant Methods. 2011, 7: 2-10.1186/1746-4811-7-2.PubMed CentralView ArticlePubMedGoogle Scholar
- Walter A, Studer B, Kolliker R: Advanced phenotyping offers opportunities for improved breeding of forage and turf species. Ann Bot. 2012, 110: 1271-1279. 10.1093/aob/mcs026.PubMed CentralView ArticlePubMedGoogle Scholar
- Fiorani F, Rascher U, Jahnke S, Schurr U: Imaging plants dynamics in heterogenic environments. Curr Opin Biotechnol. 2012, 23: 227-235. 10.1016/j.copbio.2011.12.010.View ArticlePubMedGoogle Scholar
- Green JM, Appel H, MacNeal Rehrig E, Harnsomburana J, Chang JF, Balint-Kurti P, Shyu CR: PhenoPhyte: a flexible affordable method to quantify 2D phenotypes from imagery. Plant Methods. 2012, 8: 1-12. 10.1186/1746-4811-8-1.View ArticleGoogle Scholar
- Paproki A, Sirault X, Berry S, Furbank R, Fripp J: A novel mesh processing based technique for 3D plant analysis. BMC Plant Biol. 2012, 12: 63-10.1186/1471-2229-12-63.PubMed CentralView ArticlePubMedGoogle Scholar
- Montes J, Technow F, Dhillon B, Mauch F, Melchinger A: High-throughput non-destructive biomass determination during early plant development in maize under field conditions. Field Crops Res. 2011, 121: 268-273. 10.1016/j.fcr.2010.12.017.View ArticleGoogle Scholar
- Busemeyer L, Mentrup D, Möller K, Wunder E, Alheit K, Hahn V, Maurer HP, Reif JC, Würschum T, Müller J, Rahe F, Ruckelshausen A: BreedVision – A Multi-Sensor Platform for Non-Destructive Field-Based Phenotyping in Plant Breeding. Sensors. 2013, 13: 2830-2847. 10.3390/s130302830.PubMed CentralView ArticlePubMedGoogle Scholar
- Granier C, Aguirrezabal L, Chenu K, Cookson SJ, Dauzat M, Hamard P, Thioux JJ, Rolland G, Bouchier-Combaud S, Lebaudy A, Muller B, Simonneau T, Tardieu F: PHENOPSIS, an automated platform for reproducible phenotyping of plant responses to soil water deficit in Arabidopsis thaliana permitted the identification of an accession with low sensitivity to soil water deficit. New Phytol. 2006, 169: 623-635. 10.1111/j.1469-8137.2005.01609.x.View ArticlePubMedGoogle Scholar
- Jansen M, Gilmer F, Biskup B, Nagel KA, Rascher U, Fischbach A, Briem S, Dreissen G, Tittmann S, Braun S, De Jaeger I, Metzlaff M, Schurr U, Scharr H, Walter A: Simultaneous phenotyping of leaf growth and chlorophyll fluorescence via GROWSCREENFLUORO allows detection of stress tolerance in Arabidopsis thaliana and other rosette plants. Funct Plant Biol. 2009, 36: 902-914. 10.1071/FP09095.View ArticleGoogle Scholar
- Arvidsson S, Perez-Rodriguez P, Mueller-Roeber B: A growth phenotyping pipeline for Arabidopsis thaliana integrating image analysis and rosette area modeling for robust quantification of genotype effects. New Phytol. 2011, 191: 895-907. 10.1111/j.1469-8137.2011.03756.x.View ArticlePubMedGoogle Scholar
- Reuzeau C, Pen J, Frankard V, Wolf J, Peerbolte R, Broekaert W, van Camp W: TraitMill: a discovery engine for identifying yield-enhancement genes in cereals. Mol Plant Breed. 2005, 3: 753-759.Google Scholar
- Hartmann A, Czauderna T, Hoffmann R, Stein N, Schreiber F: HTPheno: an image analysis pipeline for high-throughput plant phenotyping. BMC Bioinforma. 2011, 12: 148-10.1186/1471-2105-12-148.View ArticleGoogle Scholar
- Hoagland DR, Arnon DI: The water-culture method for growing plants without soil. Calif Agric Exp Sta Cir. 1938, 3: 346-347.Google Scholar
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.