Modeling leaf color dynamics of winter wheat in relation to growth stages and nitrogen rates
2022-01-01ZHANGYonghuiYANGYubinCHENChunleiZHANGKuitingJlANGHaiyanCAOWeixingZHUYan
ZHANG Yong-hui ,YANG Yu-bin ,CHEN Chun-lei ,ZHANG Kui-ting ,JlANG Hai-yan ,CAO Wei-xing ,ZHU Yan
1 School of Computer Engineering,Weifang University,Weifang 261061,P.R.China
2 Texas A&M University System,AgriLife Research &Extension Center,Beaumont,TX 77713,United States
3 College of Information Science and Technology,Nanjing Agricultural University,Nanjing 210095,P.R.China
4 National Engineering and Technology Center of Information Agriculture,Nanjing Agricultural University,Nanjing 210095,P.R.China
Abstract The objective of this work was to develop a model for simulating the leaf color dynamics of winter wheat in relation to crop growth stages and leaf positions under different nitrogen (N) rates. RGB (red,green and blue) data of each main stem leaf were collected throughout two crop growing seasons for two winter wheat cultivars under different N rates. A color model for simulating the leaf color dynamics of winter wheat was developed using the collected RGB values. The results indicated that leaf color changes went through three distinct stages,including early development stage (ES),early maturity stage (MS) and early senescence stage (SS),with respective color characteristics of light green,dark green and yellow for the three stages. In the ES stage,the R and G colors gradually decreased from their initial values to steady values,but the B value generally remained unchanged. RGB values remained steady in the MS,but all three gradually increased to steady values in the SS. Different linear functions were used to simulate the dynamics of RGB values in time and space.A cultivar parameter of leaf color matrix (MRGB) and a nitrogen impact factor (FN) were added to the color model to quantify their respective effects. The model was validated with an independent experimental dataset. RMSEs (root mean square errors) between the observed and simulated RGB values ranged between 7.0 and 10.0,and relative RMSEs (RRMSEs)ranged between 7 and 9%. In addition,the model was used to render wheat leaves in three-dimensional space (3D). The 3D visualizations of leaves were in good agreement with the observed leaf color dynamics in winter wheat. The developed color model could provide a solid foundation for simulating dynamic crop growth and development in space and time.
Keywords:winter wheat,crop model,virtual crop,leaf color,crop phenotype,RGB Model
1.lntroduction
A virtual crop simulates crop growth and development processes in three-dimensional space (3D),integrating knowledge in crop physiology,computer programming and graphics. It has been an interesting theme of crop simulation research in recent years (Zhang et al.2014b;Chang et al.2019),and has a wide range of applications in crop production and management,crop phenotype expression,crop plant-type design,phenotype extraction,and virtual crop production experiments (Watanabe et al.2005;Cao et al.2008;Wen et al.2019). Wheat is a staple food that plays a critical role in food security all over the world. It has been a major subject in virtual crop and plant phenotype research.
Fournier et al.(2003) constructed a 3D architectural model of winter wheat,which simulates the kinetics of elongation of individual organs together with their geometric shapes. Based on the new formalism of relational growth grammars,Buck-Sorlin et al.(2005) developed a multiscaled eco-physiological model of barley development in GroIMP. Within the framework of functional-structural plant models,Dornbusch et al.(2007) presented an architectural model on the morphology of spring barley. Evers et al.(2007) analyzed the robustness of the parameterization of ADEL-wheat for spring wheat cultivars in relation to plant population density and shading. Hu et al.(2007) put forward a new image-based 3D reconstruction method for wheat,which focuses on lamina edges,growing curves,lamina twisting,and wrapping. Based on the framework of wheat morphology visualization,Wu et al.(2009)investigated the 3D geometric modeling techniques of wheat organs,and developed organ geometrical models with characteristic architectural parameters. Meng et al.(2009) reconstructed the precise 3D structure of the wheat plant through field measurements and analysis of topological and morphological structures of wheat plants.Dornbusch et al.(2011) described the plasticity of winter wheat in the dimensions and sizes of leaf blades,sheaths and internodes in relation to their position on a stem. Li et al.(2017) implemented a wheat model that integrates growth and morphology sub-models using a 3D modeling technique based on effective accumulated temperature.The visualization can demonstrate the influences of different environmental factors on wheat morphology.
RGB models that render color characteristics,have been widely applied in crop production (Ahmad et al.2006;Lee and Lee 2011),analyzing crop seed color (Dana and Ivo 2008),diagnosing crop nutrition status (Zhu et al.2009;Song et al.2010),and estimating the LAI of soybean (Roth et al.2018),the above-ground biomass of wheat (Lu et al.2019;Ma et al.2019),and potato emergence (Li et al.2019)based on RGB images of the crop canopy.
The crop leaf is a vital vegetative organ,and leaf color is a key phenotype feature of a crop,which is affected by crop growth conditions (Zhao et al.2006) and can reflect crop growth state (Yang et al.2003;Witt et al.2005),so it could be used to guide crop management decisions.Previous research has mostly focused on the geometric modeling of wheat plants or organs,while no work has been done on developing RGB-based models for winter wheat leaf color dynamics (Fournier and Andrieu 1998;Fournier et al.2003;Hanan and Hearn 2003). This study was undertaken to develop a model for predicting the phenotype dynamics of leaf color in time and space for winter wheat in relation to crop growth stages and leaf positions under different N rates. The single leaf color modeling could improve the realistic accuracy of virtual wheat and provide a solid foundation for simulating dynamic crop growth and development in space and time.
2.Materials and methods
2.1.Experiment design
Data were collected from two field experiments,one with five nitrogen (N) application rates at Pingdu in Shandong Province,and the other with four N application rates at Weifang in Shandong Province;the same two wheat cultivars were planted in each experiment (Table 1). The two cultivars,Jimai 22 (JM22) and Jinan 17 (JN17),were arranged in a completely randomized block design with three replicates. Plot size was 12 m2. N fertilizer was applied before sowing (50%) and at the stem elongation stage(50%). The amounts of P-and K-applied were based on soil tests to satisfy plant growth demand. Locally recommended crop management procedures for high yield were followed.The dataset from experiment 1 (Exp.1) was used for model calibration and the dataset from Exp.2 for validation.

Table 1 Experimental design
2.2.Data acquisition
Ten plants from each plot were randomly selected for data collection every 2-5 days. An EPSON scanner (EPSON Perfection 1200 PHOTO,Japan) with 24-bit color depth was used to scan leaves on the main stem of the selected plants.Image resolution was set at 200×200 pixels. Each scanned leaf was divided into 5 to 10 equal segments along the leaf axis;the RGB data (0-255) of leaf color for each section were then extracted from the basal to the distal segment using an in-house script. RGB values for a segment were the average values of pixels in the segment. Finally,the RGB values were calibrated with a standard whiteboard correction method (Cheng et al.2007),i.e.,when the wheat leaves were scanned with the scanner,a standard whiteboard was set as the background. The RGB values of the standard whiteboard are all 255. In one scan,if the average RGB values of standard whiteboard were r,g and b,respectively,and those of a leaf segment were r1,g1 and b1,respectively,then the calibrated RGB values of a leaf segment were:r1-255/r,g1-255/g,and b1-255/b. In normal cases,the average RGB values of the standard whiteboard were larger than the respective values of a leaf segment.
Air temperatures during the growing season were recorded at 30-min intervals with a ZDR-11 (Zhejiang University of China),and was used to calculate thermal time (TT,°C d). In addition,the dry weights and nitrogen contents of plant samples were recorded.
2.3.Data analysis
The datasets of Experiments 1 and 2 were used to calibrate and validate the models,respectively.
Regression,variance analysis,and leaf color visualization were carried out in Excel 2013,MatlabR2016a and Visual Studio 2013,respectively. The differences between the simulated and observed values were evaluated with the root mean square error (RMSE) and the relative RMSE(RRMSE) as follows (Rinaldi et al.2003):

where n is the total number of measurements;Piand Oiare simulated and observed values,respectively;andis the observed mean value.
3.Results
3.1.Modeling the leaf color dynamics with thermal time
Thermal time (TT) is the cumulative sum of Ti-Tb,where Tiis the average temperature on the ith day after the first leaf emergence,and Tbis the base temperature. The results from the experiment indicated that leaf color change in winter wheat went through three distinct stages,including early development stage (ES),early maturity stage (MS)and early senescence stage (SS). They followed a distinct pattern of light green-dark green-yellow during the three developmental stages,with corresponding changes in the RGB (red,green and blue) values (Fig.1). The R and G values gradually decreased from their initial values to a steady state,but the B value generally remained unchanged during the ES. The RGB values remained steady during the MS,but gradually increased during the SS (Figs.2-4)

Fig.1 Color changes of a main stem leaf in winter wheat cultivar Jimai 22 at different developmental stages.
Each leaf was divided into 5-10 equal segments along the leaf axis from the distal to the basal ends,depending on leaf length. For leaves at the same main stem nodal position under different N rates,the change trends of RGB values in the leaf distal segment were similar (Fig.2),and no significant differences were found in the initial and final RGB values for leaves at different nodal positions and under different N rates,respectively (P>0.05),but the steady RGB values of the MS obviously decreased with increasing N rates (Fig.2). The dynamics of RGB values in the distal segments at different main stem nodal positions under the same N rate were also similar (Fig.3),and no significant differences were observed in the initial,steady and final RGB values of the distal segments at different leaf positions under the same wheat cultivar and N rate,respectively(P>0.05). Furthermore,the changes in RGB values of different segments of the same leaf were identical along the leaf axis from the distal to the basal segments (Fig.4),and no significant differences among different segments of the same leaf were observed for the initial,steady,and final RGB values,respectively (P>0.05). According to the above analysis,a linear piecewise function of eq.(1) was used to quantify the RGB changes of the distal segment of leaf n with TT (and the color model of the whole leaf is developed in the following section).

Fig.2 RGB (red,green and blue) changes of the top segment of the 10th leaf on the main stem of Jimai 22 (JM22) and Jinan 17 (JN17) under different nitrogen (N) rates with thermal time (TT). TT durations at early development stage (ES),early maturity stage (MS) and early senescence stage (SS) for the 10th leaf on the main stems of JM22 and JN17 approximately ranged over[902,1 168],[1 168,1 346],[1 346,1 479.5],and [848,1 097],[1 097,1 263] and [1 263,1 387.5],respectively. N1JM22-R,R value of wheat cultivar JM22 under N1 rate in Exp.1,and the meanings of the other labels follow this pattern. N1,N2,N3,N4,and N5 are equal to 0,80,160,240,and 320 kg N ha-1,respectively.

Fig.3 RGB (red,green and blue) changes of the distal segments of the 5th and 10th leaves on the main stems of Jimai 22 (JM22)and Jinan (JN17) with thermal time (TT) under the N4 rate of 240 kg N ha-1 in Exp.1. N4JM22-5R,R value of the 5th leaf color in wheat cultivar JM22 under N4 rate in Exp.1,and the meanings of the other labels follow this pattern.

Fig.4 RGB (red,green and blue) changes of the distal (D),the middle (M) and the basal (B) segments of the 10th leaf on main stems of Jimai (JM22) and Jinan 17 (JN17) with thermal time (TT) under the N4 rate of 240 kg N ha-1 in Exp.1. JM22-DR,the R value of the top segment of the leaf in JM22;the meanings of other labels follow this pattern.

where n is the main stem nodal position. TT denotes the thermal time,calculated by eq.(2). ITTnis the TT at the initial elongation of leaf n,calculated by eq.(3). PHYLL,a cultivar parameter,is the phyllochron which denotes the thermal time between two successive leaf appearances,89 and 83°C for JM22 and JN17,respectively,according to our experimental data and related models (Yan et al.2000,2001;Zhu et al.2008). D1,D2and D3are the coefficients of the durations of the ES,MS and SS,respectively. According to our experimental records and previous studies (Jing 1996;Zhu et al.2008),D1,D2and D3are equal to 3.0,2.0 and 1.5,respectively. NT1and NT2denote the normalized TT at the ES and SS,respectively,as estimated by eq.(4). TRGB(n,TT) is a vector composed of R,G and B values of the distal segment of leaf n at TT. TRGBI,TRGBSand TRGBFdenote the initial,steady and final RGB vectors of the distal segment of leaf n at the ES,MS and SS,calculated by eqs.(5)-(7),respectively.

where Ti,the average temperature on the ith day;Tb,a basal temperature;Tb,0°C before the double ridge stage;Tb,3.3°C after the double ridge stage (Yan et al.2000).

where TT0is the accumulated TT from seeding to leaf appearance,101°C for two cultivars of JM22 and JN17.

where MRGBis a leaf color matrix,which is calculated by eq.(8) and denotes the average RGB values of all leaves on main stem in winter wheat under an adequate N rate;MRGB(:,i) denotes the ith column of matrix MRGB,MRGBis a cultivar-specific parameter;and FN is the N impact factor,calculated by eq.(9):



where ANCSH is the actual N content;MNCSH is the optimum N content,calculated by eq.(10):

where DM is the dry matter of above-ground wheat plant;a and b are equation coefficients of 5.35 and -0.44,respectively (Justes et al.1994).
3.2.Modeling the changes of leaf color along the leaf axis
Leaf color varied along the leaf axis with an identical trend of light green-dark green-yellow,but the timings of the changes were asynchronous among the different segments during the ES and SS (Fig.4). The color differences along the leaf axis from the distal to the basal segments were mainly caused by the differences in elongation timings of the different leaf segments. However,there was little difference in leaf color along the direction perpendicular to the leaf axis and the values at any point E on line segment CD,which is perpendicular to AB,were assumed equal (Fig.5). Eq.(11)was used to calculate the RGB values of point (E) in leaf n at TT time.where LRGB(r,n,TT) denotes the RGB values of any point(E) in leaf n at TT;LLnis the length of leaf;r denotes the distance from the line segment (CD) to the basal point of the leaf (B);other parameters are the same as above. All symbols used in this study are listed in Table 2.

Table 2 List of symbols used in this study

Fig.5 Relation graph of leaf structure in winter wheat.

3.3.Model validation
The datasets of the distal,middle,and basal segments of the 5th-9th leaves on the main stem under N2 rate at tillering stage,and the 7th-11th leaves on the main stem at stem elongation stage under N4 rate in Exp.2 were selected to validate the leaf color model. Comparisons of the simulated and measured RGB data indicated that RMSEs were 9.62(Fig.6-A) and 7.98 (Fig.6-B) at tilling stage for JM22 and JN17,7.52 (Fig.7-A) and 7.76 (Fig.7-B) at stem elongation stage for JM22 and JN17,respectively;while RRMSEs were 8.49% (Fig.6-A) and 7.24% (Fig.6-B) at tilling stage for JM22 and JN17,and 7.08% (Fig.7-A) and 7.10% (Fig.7-B)at stem elongation stage for JM22 and JN17,respectively.These results showed that the models developed here presented a good performance for predicting the changes of leaf color in time and space.

Fig.6 Comparisons of the measured and simulated RGB (red,green and blue) values of distal (D),middle (M) and basal (B)segments of leaves on main stems of Jimai 22 (JM22) (A) and Jinan 17 (JN17) (B) under the N2 rate of 90 kg N ha-1 at tillering stage in Exp.2. JM-DR,R value of the top segment of wheat leaf,the meanings of the other labels follow this pattern.

Fig.7 Comparisons of the measured and simulated RGB (red,green and blue) values of distal (D),middle (M) and basal (B)segments of leaves on main stems of Jimai 22 (JM22) (A) and Jinan (JN17) (B) under the N4 rate of 270 kg N ha-1 at tillering stage in Exp.2. JM-DR,R value of the top segment of wheat leaf,the meanings of the other labels follow this pattern.
3.4.Model application
Integration of the leaf color model and 3D visualization of leaf color dynamics in time and space were realized using the C# and OpenGL on the .Net platform. The visualization results of the 10th leaf on the main stem in JM22 under N1 and N3 rates,respectively are shown in Figs.8 and 9. The results were relatively realistic,had a good prediction for the leaf color changes and embodied the differences among the various N rates.
4.Discussion
Leaf color is a vital phenotype trait of crops,and has been widely applied in crop nutrition diagnosis (Yang et al.2003;Yadav et al.2010;Li et al.2019),and monitoring crop growth status (Roth et al.2018;Lu et al.2019;Ma et al.2019). Leaf color models could be integrated with diagnosis tools for thecritical nitrogen dilution curve (Yao et al.2014),chlorophyll content (Yadav et al.2010),and lipid content (Su et al.2008)to quantitatively assess leaf N status in winter wheat. These advances would extend the application of virtual wheat in agricultural production. Crop leaf color is readily affected by environmental factors,thus the data of leaf color are difficult to obtain. In our work,RGB data for different leaf segments along the leaf axis were collected during leaf growth stages using a scanner and calibrated by the whiteboard method(Cheng et al.2007). These measurements could reduce the errors in the RGB data compared to the previous studies(Zhu et al.2010;Lee and Lee 2011).
Our experimental data and previous study (Zhao et al.2006) indicated that crop leaf color is affected by cultivars.The initial,steady and final RGB values of leaf color for the nth main stem leaf for a specific wheat cultivar under an adequate N condition were more or less steady. Therefore,a cultivar parameter MRGBwas introduced into the leaf color model to quantify the color differences among the cultivars. MRGBwas useful for improving the robustness of the model for different cultivars of winter wheat. Crop leaf color varied with changes in the N rate (Liu et al.2019),but no quantitative results had been reported in previous works. In present work,the RGB values of leaf color during the MS were generally steady under the same N rate,and decreased with increasing N rate,but the initial and final RGB values of leaf color were only slightly affected by N rate.Therefore,FN was added into the color model to quantify the effect of N rate on RGB values of leaf color in the MS. The FN was useful for improving the robustness of the model for winter wheat under different N rates.

Fig.8 Leaf color dynamics of the 10th leaf on the main stem of Jimai 22 (JM22) at different days after leaf emergence under the N1 rate of 0 kg N ha-1 in Exp.2.
With the progression of wheat leaf growth,leaf color changed from light green to dark green along the leaf axis from the distal to the basal segments at the ES. Leaf color generally remained dark green in the MS. Finally,leaf color changed from dark green to yellow along the leaf axis at the SS.Similar results on crop leaf color dynamics had been reported in previous studies (Zhu et al.2008;Chang et al.2011;Zhang et al.2014a). During the ES and SS,the timing of biomass transfers into and out of different leaf parts are asynchronous,thus the changes of leaf color were not uniform along the leaf axis,and these have not been reported in previous works (Fournier et al.2003;Watanabe et al.2005;Meng et al.2009;Wu et al.2009;Li et al.2017).In our model,the leaf color change along the leaf axis was quantified,and 3D dynamic rendering of leaf color was realized,providing a capability that could more realistically capture leaf color dynamics in time and space,and under different N rates,than in previous studies (Fournier et al.2003;Meng et al.2009;Wu et al.2009;Li et al.2017).

Fig.9 Leaf color dynamics of the 10th leaf on the main stem of Jimai 22 (JM22) at different days after leaf emergence under the N3 rate of 180 kg N ha-1 in Exp.2.
The RGB dynamics of leaf color on tillers were not quantified in our work,but those RGB values could be calculated based on the synchronous growth rules between tillers and main stem in winter wheat. Further investigations are needed to capture the effects on leaf color of other factors,such as water and other nutrient stresses.
5.Conclusion
An RGB Model was developed and evaluated that captures the color dynamics of wheat leaf on the main stem in time(crop growth process) and space (along the leaf axis) for different wheat cultivars and N rates. This model could provide a solid foundation for simulating dynamic crop growth and development in space and time.
Acknowledgements
The work was supported by the National Natural Science Foundation of China (31872847),the Higher Educational Science and Technology Program of Shandong Province,China (J18KA130),the Science and Technology Benefiting People Plan Project of Weifang High-Tech Zone,Shandong Province,China (2019KJHM13),and the Natural Science Foundation of Shandong Province,China (ZR2019PF023).
Declaration of competing interest
The authors declare that they have no conflict of interest.
杂志排行
Journal of Integrative Agriculture的其它文章
- Effect of harvest time on the chemical composition and antioxidant capacity of Gannan navel orange (Citrus sinensis L.Osbeck ‘Newhall’) juice
- New geographic distribution and molecular diversity of Citrus chlorotic dwarf-associated virus in China
- The relationship between farm size and fertilizer use efficiency:Evidence from China
- Does Internet use promote the adoption of agricultural technology?Evidence from 1 449 farm households in 14 Chinese provinces
- Polypropylene crisper and 1-MCP delay the softening,lignification and transcription levels of related enzyme genes of golden needle mushrooms (Flammulina velutipes)
- Effects of land use/cover change (LUCC) on the spatiotemporal variability of precipitation and temperature in the Songnen Plain,China
