Effects of mechanical defoliation and pinching applications on plant growth and root system analysis with machine learning in boxwoods

Abstract

The effects of mechanical defoliation and pinching (1 cm tip cutting) on Buxus plant growth, nutrient mobilization, and root architecture were determined. When 100% defoliation was applied, the highest increase rates of 80.3% in shoots and 88% in leaves were observed compared to the control group. In contrast, the overall effects of defoliation and pinching were negative, with 100% defoliation having the most negative effects. The chlorophyll content of the newly formed young leaves was also 50% lower with 100% defoliation. Leaves and root nutrient mobilization changed significantly, depending on the effects of defoliation and pinching. Apart from a very small increase in root length and number of forks, the effects of the treatments were negative, with 100% defoliation having the greatest negative effect on root development. Most affected was the number of crossings, which was 78% lower than in the control. In addition, machine learning (ML) algorithms were used in the study, including multilayer perceptron, J48, PART, and logistic regression. The input variables were evaluated to model and predict the root features. The performance values of the ML algorithms were noted in the following order: Logistic Regression> PART> J48> MultilayerPerceptron. As the severity of defoliation increased, the losses of the plant also increased. However, boxwood has mechanisms to compensate for these losses even when it suffers complete defoliation.

Full Article

Effects of Mechanical Defoliation and Pinching Applications on Plant Growth and Root System Analysis with Machine Learning in Boxwoods

Ömer Sarı *

The effects of mechanical defoliation and pinching (1 cm tip cutting) on Buxus plant growth, nutrient mobilization, and root architecture were determined. When 100% defoliation was applied, the highest increase rates of 80.3% in shoots and 88% in leaves were observed compared to the control group. In contrast, the overall effects of defoliation and pinching were negative, with 100% defoliation having the most negative effects. The chlorophyll content of the newly formed young leaves was also 50% lower with 100% defoliation. Leaves and root nutrient mobilization changed significantly, depending on the effects of defoliation and pinching. Apart from a very small increase in root length and number of forks, the effects of the treatments were negative, with 100% defoliation having the greatest negative effect on root development. Most affected was the number of crossings, which was 78% lower than in the control. In addition, machine learning (ML) algorithms were used in the study, including multilayer perceptron, J48, PART, and logistic regression. The input variables were evaluated to model and predict the root features. The performance values of the ML algorithms were noted in the following order: Logistic Regression> PART> J48> MultilayerPerceptron. As the severity of defoliation increased, the losses of the plant also increased. However, boxwood has mechanisms to compensate for these losses even when it suffers complete defoliation.

DOI: 10.15376/biores.19.4.7450-7477

Keywords: Buxus sempervirens; Root morphology; Leaf loss; Pinching; Plant characteristics; Nutrient distribution; Machine learning

Contact information: Black Sea Agricultural Research Institute, Çiftlik Mah. Atatürk Bulvarı No: 313 (Samsun – Ordu Karayolu 17. km), Gelemen, Tekkeköy / SAMSUN / TÜRKİYE;

* Corresponding author: omer.sari@tarimorman.gov.tr

GRAPHICAL ABSTRACT

INTRODUCTION

Boxwood, has attracted the attention of people throughout the history of civilization due to its slow growth, long life, hard and durable wood structure, leafy and evergreen nature and easy reproduction. Boxwoods have a wide range of uses due to these properties. However, boxwoods are generally used as ornamental plants today. The best-known and most cultivated species is Buxus sempervirens (Köhler 2014). Boxwoods are used as ornamental plants in single and mass plantings, hedges, potted plants, topiary, and cut greenery (Batdorf 2005; Van Trier et al. 2005; Köhler 2014). In addition to being ornamental plants, boxwoods have also been used in the production of musical instruments, writing tablets, combs, carved ornaments, paintings, and sculptures due to their hard and dense wood (Batdorf 2005; Mitchell et al. 2018). Due to these properties, they have been grown intensively for centuries (Larson 1999; Van Trier and Hermans 2007). One of the methods used to increase shoot yield in boxwood cultivation is pinching. Pinching is the removal of apical dominance to promote lateral branching and flowering (Sasikumar 2015; Ehsanullah et al. 2021). However, in natural environments, due to effects such as animals and wind, and in nurseries, during plant transportation, plant branches can break either completely or at their tips. Similarly, defoliation by anthropogenic disturbances, insects, and grazing animals affects plant growth and vegetation dynamics (Wang et al. 2020). Under natural conditions, plants can often suffer foliar damage from human activities (cutting of branches, use of pesticides that cause defoliation, etc.), as well as insects and grazing animals (Eyles et al. 2013). In addition, climate change can lead to defoliation. Climate change will increase the fitness and abundance of some forest pests and affect plant growth and vegetation dynamics (Dale and Frank 2017; Kuosmanen et al. 2018). These examples illustrate the fact that multiple biotic and abiotic stressors can cascade and alter forest vitality. The future of boxwood, one of the most important forest and ornamental plants, has been seriously threatened in recent years due to diseases and damaging influences. The most important of these are the box tree moth (C. perspectalis) (Van der Straten and Muus 2010) and the box tree borer (C. buxicola), which have seriously damaged box trees both in the wild and in the landscape in numerous European, Caucasian, and Turkish countries (Malapi-Wight et al. 2014). Both the box tree moth and the box tree borer cause leaf loss in box trees. For this reason, the plant may lose its physiological balance and therefore die. Considering the biodiversity associated with boxwood, the destruction of natural boxwood stands in limited areas by pests can have serious ecological consequences. For this reason, the extinction of box trees can indirectly cause the disappearance of many living organisms (Mitchell et al. 2018).

Studying the effects of biodegradation on young trees and plant responses can also provide a theoretical basis for the selection and management of species to restore vegetation (Wang et al. 2020). In addition to natural conditions, this damage can also occur under controlled conditions and when transplanting plants. For this reason, defoliation experiments are often used to simulate biodegradation in plant species (Wiley et al. 2017; Wyka et al. 2017). Leaf removal reduces the photosynthetic area of leaves and affects root viability and nutrient concentration of plants (Quentin et al. 2010; Barry et al. 2012; Eyles et al. 2013; Jacquet et al. 2014; Wiley et al. 2017). It has been reported that the photosynthetic capacity of the remaining leaves increases after defoliation, and plants rapidly produce new leaves (Korpita and Orians 2014; Qiu et al. 2016; Eyles et al. 2016). Plants can store and reactivate reserves to restore leaf area (Moot et al. 2021). Previous studies have shown that plants tend to allocate more biomass to the aerial parts after defoliation (Mukherjee et al. 2015; Wiley et al. 2017; Chen et al. 2017). To illustrate, physical leaf damage alters the pattern of resource allocation to different vegetative and reproductive organs (Marshall et al. 2005; Stevens et al. 2014; Wang et al. 2020). In general, species, even within the same genus, respond differently to defoliation, and marked differences in compensation time, biomass, and chemical content can occur. While the responses of Eucalyptus globulus and Eucalyptus nitens to defoliation were similar (Barry and Pinkard 2013), white oak (Quercus alba) and black oak (Quercus velutina) responded differently to defoliation (Rieske and Dillaway 2008). The plants preferentially distribute more carbon to the shoots to compensate for the reduced leaf area due to the loss of carbon (C) as a result of defoliation (Wang et al. 2020). It has been reported that the production of new leaves after defoliation leads to a decrease in root mass and increased root mortality in most plants (Kosola et al. 2001; Hikosaka 2005). Pinching is also used in addition to pruning in annual cut flower species such as African marigold (Tagetes erecta), gypsophila (Gypsophila oniculata), and lisianthus (Eustoma grandiflorum) to stimulate the formation of new shoots (Cheong et al. 2002; De Pascale et al. 2005; Badge et al. 2014), and perennial ornamental plants such as Buxus (Buxus spp.), rose, and holly (Brum et al. 2007). However, it has been reported that a significant part of the N produced by the plant is consumed in flowers and reproductive organs (Zhang et al. 2021). Root growth of potted plants is a key element in the overall performance of the plant (Wraith and Wright 1998). A strong root structure has a positive effect on plant productivity by increasing plant water and nutrient uptake as well as resistance to diseases and pests. Under stress conditions, the underground parts of plants are most affected (Comas et al. 2013; Bucksch et al. 2014). Root architecture varies greatly in response to different nutrient deficiencies and shows the ability to adapt to ever-changing growth conditions through structural flexibility (Sun et al. 2017). Morphological changes in the root system are regulated by the plant’s nutritional status and its interaction with the environment, which are detected by localized signals from the roots (Giehl et al. 2014; Razaq et al. 2017). However, it is not very easy to study the root structure, which is naturally located underground. This is the main reason why not many studies have based their results on the phenotypic characteristics of the root. In recent years, numerous advances have been made in the measurement of roots, in addition to the development of techniques such as software to analyze plant images that can describe root growth in a simpler, faster, more reproducible and descriptive way (Judd et al. 2015; Paez-Garcia et al. 2015). In addition, modeling techniques for the structure and activity of root features based on multivariate and machine learning methods have been investigated. However, further studies are still needed to determine the importance of root traits in influencing aboveground biomass (Moon et al. 2018; Awika et al. 2021; Tütüncü 2024).

Machine learning is used to predict the effects of many applications in agriculture, especially crop yields. It is one of the techniques used. Artificial neural networks, support vector machines, linear and logistic regression, decision trees and Naïve Bayes are some of the algorithms used for prediction. Difficulty in selecting the algorithms lies in not knowing which of the existing algorithms is suitable for the plant under study (Palanivel and Surianarayanan 2019). However, there is very little information on boxwoods regarding the effects of pruning on the roots, plant parts and nutrient content.

It is important to ensure the sustainable cultivation of crops. The effects of defoliation and pinching on plant development, nutrient content of roots and leaves in boxwood and architectural features of roots were evaluated by image analysis. In addition, the study attempted to model and predict the effects of applications on root architecture by using methods such as artificial neural network analysis and machine learning based on data mining.

EXPERIMENTAL

Materials

In this study, 3-year-old boxwoods taken from Samandağ’ı district of Hatay province (36º11’15.98” N 35º55’55.18” E, altitude 725-1250 m), Türkiye in 2020 and transferred to pots after rooting were used.

Experimental Design

Mechanical defoliation and pinching applications were carried out on March 15, 2023, before the shoot period. In the study, 30 plants were used for each group and all leaves of 30 plants were counted for each group before leaf defoliation. In the first group the leaves were not removed, 25% of total leaves in the second group, 50% of the total leaves in the third group, and 100% of total leaves are in the fourth group were removed by hand. In the pinching application, the tips of the lateral and terminal branches of each of the 30 plants were measured with a ruler and cut off with pruners at 1 cm. The trees were regularly maintained (watering, weeding, etc.). No fertilizer was applied during the experiment. The average temperature in the greenhouse was 27.5 °C and the humidity was 65 °C in 2023. During the experimental period, the total sunshine duration was 1120 h.

Measuring Plant Growth Characteristics

Plant height (cm), plant height (cm), number of leaves, number of shoots, shoot length (mm), shoot diameter (mm), leaf width (cm), and leaf length (cm) were measured on June 30. While the plant height, leaf width and leaf length were measured with a ruler, the shoot length and shoot diameter were measured.

Chlorophyll Content

The effect of the applications on the leaves chlorophyll content was determined. For this purpose, measurements were made at the beginning of the vegetation period and at the end of the vegetation period.

First measurement: Chlorophyll content in plants to be included in each application, before applications to boxwoods measured on March 15 in old leaves.

Second measurement: Measurements were made on both old and newly formed leaves on June 30, the end of the first vegetation period after the applications. The measurements were taken on 10 leaves from each plant. The relative chlorophyll content in the leaf were measured using the SPAD-502 chlorophyll meter (Minolta Camera Co., Ltd., Japan). The upper four rows of leaves were used to measure the SPAD index.

Nutrient Content Analysis of Leaves and Roots

Old leaves and the youngest leaves that had completed their development (the 8^th leaf from the growing tip) were removed. The plants with roots were removed from the pots on June 30. They were washed and dried with paper towels. Then the roots were cut from the healthy tip and the lateral parts of the root. The leaves and roots were dried at 65 °C for 48 h, and then three plants were randomly selected from each replicate of each application. The leaf and root samples taken from the plants were washed for chemical analysis, dried and ground at 65 °C until they reached a constant weight. Total nitrogen in the ground samples was determined using the modified Kjeldahl method (Kacar and Inal 2008). For the analysis of P, K, Ca, Mg, Fe, Mn, Zn and Cu, the plant samples were wet-burned (4:1, HNO₃:HClO₄) and read in the ICP-OES instrument (Soltanpour and Workman 1981).

Rooting Potential and Phenotypic Root Development Examinations

The root analysis program WinRhizo (Regent Instruments, Quebec, Canada) was used to study root architecture. The plants with roots were removed from the pots on June 30. The roots of the removed plants were carefully washed and cleaned. Then, the roots were transferred to the computer in three dimensions using the scanner of the device (Epson Expression 10000XL, Epson America Inc., Long Beach, CA, USA) and computerized. The following parameters of the root structure and the degree of rooting were examined using the WinRhizo program: WinRhizo software made it possible to determine the total root length (cm), root surface area (cm²), root volume (cm³), average root diameter (mm), number of tips, number of forks, and number of crossings.

Fig. 1. The MLP structure with 7 inputs, 5 outputs, and 10 hidden neurons

Modeling Procedures and Classification Techniques

To model and predict the effects of treatments on root characteristics after defoliation and pinching of B. sempervirens, different data mining algorithms available in WEKA 3.9.6 (Machine Learning Group, University of Waikato) (Bouckeart et al. 2016) were applied to the dataset. The results obtained were compared. A model was created by selecting the algorithm with the highest success rate among these algorithms. Four machine learning methods – Multilayer Perceptron, J48, PART and logistic regression – were used in the study. The input variables consisted of one species and seven different root characteristics are measured (root length, root surface area, root volume, average root diameter, number of tips, number of forks and number of crossings) The target variables (output) control, 25%, 50%, 100% defoliation and pinching included (Fig. 1).

Data Evaluation

The study was conducted using a completely randomized design, each containing a single seedling, and 30 replicates were evaluated for each treatment. Analysis of variance was performed using SPSS statistical software version 20.0 and differences between treatments were compared using Duncan’s multiple comparison test (within 5% and 1% error limits).

RESULTS AND DISCUSSION

Characteristics of Plant Growth

The applications significantly influenced plant development. Pinching had the biggest effect, as it increased plant height (14.8%), plant width (4.8%), and shoot length (9.2%) compared to the control. The highest results were found in the number of shoots (80.3%) and number of leaves (88%) with 100% defoliation compared to the control. The highest results (18% and 11%) were found for the increase in shoot diameter at 25% and 50% defoliation. The applications for leaf width and leaf length showed similar effects, except for 100% defoliation. Plant height (28.6%), shoot length (42%), shoot diameter (11%), leaf length (18.2%), and leaf width (43.3%) were the applications that produced the lowest results at 100% defoliation. The lowest results were found for plant width (14.5%) and number of leaves (47.3%) compared to the control at 25% defoliation. The number of shoots (88%) was lowest when pinching was applied compared to the control (Fig. 2).

Fig. 2. Rates of change of the characteristics of the upper part of the plant compared to the control values of the analysis results after defoliation and pinching (25%, 50%, 100% defoliation and pinching).

Changes in the nutrient content of the roots and leaves

The applications had a significant effect on the amount of plant nutrients in the roots. According to the results of the application, the N, P, and Ca contents increased in all three applications. The K content and Zn content were lower in all treatments compared to the control. The Mg content was lower at 25% and 100% defoliation compared to the control, but a slight increase was observed at 50% and pinching. Again, Fe and Mn contents were lower at 100% defoliation compared to the control, while they increased in the other applications. While the Cu content was lower in the 50% and pinching treatments than in the control, an increase was observed in the 25% and 100% treatments. In all other treatments, the nutrient element contents increased compared to the control. The greatest decrease was observed for Zn. Among the applications, the greatest loss of nutrients was found in 100% defoliation, while the least loss was observed in 25% defoliation (Figs. 3 and 4).

Fig. 3. The rate of change of nutrient content in the roots and leaves of the plants compared to the reference control of the analysis results after fertilizer application (a: %25 defoliation, b: %50 defoliation, c: %100 defoliation, d: pinching)

The applications had a significant effect on the amount of plant nutrients in the leaves. According to the results of the applications, N and P increased with all four applications, but least with the application of pinching. While K, Zn, and Mn levels increased in 25%, 50%, and 100% of the applications, respectively, compared to the control, it decreased in the pinching application. The increasing trend in nutrient content varied according to the severity of leaf loss. Ca and Cu levels were found to be lower in all treatments compared to the control. While Mg content decreased in 50% defoliation and pinching applications compared to the control, it increased in 100% defoliation. While Fe increased in 25%, 50% defoliation and pinching applications, it was also found to be lowest in 100% defoliation. Pinching was generally the application with the lowest nutrient content. The application with the lowest levels of N, K, Mg, Zn, and Mn compared to the control was only a pinching application. In contrast, the application with the highest levels of P, K, and Zn showed 25% defoliation. Fe and Mn were highest at 50% defoliation compared to the control. While the Mg content was highest at 100% defoliation, the N content had the same effect as the other 25% and 50% defoliation. The highest value in terms of foliar nutrient content was found for Mn application, followed by the amount of N. The lowest value was recorded for Ca and Cu, while the highest decrease was recorded for pinching application (Figs. 3 and 4).

Fig. 4. Change in nutrient content in roots and leaves depending on defoliation and pinching application

Chlorophyll content

The effect of the applications on chlorophyll content was found to be statistically significant. The amount of chlorophyll in the plant leaves increased by 13% compared to the first measurement (15 March), from 78.6 CCI in the control to 89 CCI. It was found to increase by 13.4% from 77.7 CCI to 88.1 CCI at 25% defoliation. At 50% defoliation, it increased by 28% from 75 CCI to 96.2 CCI and at pinching application by 56% from 69.3 CCI to 108.1 CCI (Table 1). A small increase in the amount of chlorophyll was observed in the control plants.

The amount of chlorophyll in the newly formed leaves was compared with the amount of chlorophyll in the leaves at the first measurement. Compared to the first measurement (15 March), the amount of chlorophyll in newly formed leaves after 107 days (30 June) was 16% less in the control, 9% less in 25% defoliation, 14% less in 50% defoliation, and 50% less in 100% defoliation (Table 1).

Table 1. Effect of Defoliation and Pinching on the Chlorophyll Content of Leaves

Morphological and architectural characteristics of the root

The architectural characteristics of roots were found to be significantly affected by applications. Root length was 7.7% higher in the 50% defoliation and 5.3% higher in the pinching application than in the control.

The number of forks was 2.3% higher in the pinching application than in the control. The effect of the applications on root root characteristics other than root length and number of forks was generally found to be lower than the control. At 25% defoliation, root length was 6.3% less, root surface area was 15.5% less, root volume was 13.6% less, number of tips was 5.3% less, number of forks was 7.7% less, and number of crossings was 22% less, while the root diameter value was the same as the control. It was found that at 50% defoliation the root surface area was 7.1%, the root volume 20.3%, the root diameter 33.3%, the number of tips 4.4%, the number of forks was 5.5%, and the number of crossings was 24.5%, which was significantly lower than the control. The lowest results were achieved with 100% defoliation. Accordingly, with 100% defoliation, root length was 18.8%, root surface area 31%, root volume 32.2%, root diameter 33.3%, number of tips 43.2%, number of forks was 33%, and number of crossings 78% lower than the control (Fig. 5).

Fig. 5. The rate of change of the architectural root characteristics of the roots compared to the control values of the analysis results after defoliation and pinching (25%, 50%, 100% defoliation and pinching)

ML modeling analysis

The values classified by canopy clustering are divided into four groups. The values classified into 4 groups showed that 100% defoliation had the most effective (negative) effect on the architectural features of the roots, with 40% in the 2^nd cluster. In cluster 0, pinching was 16% successful, in cluster 1, control was 16% successful, in cluster 3, defoliation was 25% to 12% successful and in cluster 4, defoliation was 50% to 16% successful. This grouping made it possible to verify the results of this study with artificial intelligence. A 100% defoliation is generally the most effective application achieved with classical methods. In addition, root length was determined as the most important result by using the select attribute result in the CfsSubsetEval algorithm and the BestFirst search method.

Artificial neural networks

The study considered 25 %, 50 %, and 100 % defoliation and pinching. After defining the total root length (cm), root surface area (cm²), root volume (cm³), average root diameter (mm), number of tips, number of forks, and number of crossings as output variables, the output variable was predicted using the input variables. The decision tree for the decision is shown in Fig. 6.

Fig. 6. Decision tree obtained by J.48 method

As shown in Fig. 6, to classify the applications of defoliation and pinching in terms of their impact on root architecture, it can be seen in the decision tree that the most important impact in the context of the study was on the number of crossings, and this is due to the impact of 100% defoliation. Indeed, the application and root trait reduced the number of crossings the most at 100% defoliation with 78% compared to the control. In this respect, ML made the correct prediction.

Choosing the most suitable model

Algorithms frequently mentioned in the literature (multilayer perceptron, J48, PART, and logistic regression) were used to select the most appropriate model. In this context, the most successful algorithm was selected based on the correct prediction rate. The selected algorithms were used to create models one after the other. As a result, it was decided to apply the logistic regression algorithm, which has the highest accuracy of 84%, to the data set. In selecting this algorithm, the accuracy value, duration and average absolute error were considered. Accordingly, the performance levels of the models created using the Multilayer Perceptron, J48, PART and Logistic Regression algorithms were compared and the resulting performance levels were noted in the following order: Logistic Regression> PART> J48> Multilayer Perceptron (Table 2).

Table 2. Predictive Power of Machine Learning Models Depicts the Relationship Between the Variables Defoliation and Pinching and the Change in Root Architecture Characteristics of B. sempervirens