Using hyperspectral leaf reflectance to estimate photosynthetic capacity and nitrogen content across eastern cottonwood and hybrid poplar taxa.

Eastern cottonwood (Populus deltoides W. Bartram ex Marshall) and hybrid poplars are well-known bioenergy crops. With advances in tree breeding, it is increasingly necessary to find economical ways to identify high-performing Populus genotypes that can be planted under different environmental conditions. Photosynthesis and leaf nitrogen content are critical parameters for plant growth, however, measuring them is an expensive and time-consuming process. Instead, these parameters can be quickly estimated from hyperspectral leaf reflectance if robust statistical models can be developed. To this end, we measured photosynthetic capacity parameters (Rubisco-limited carboxylation rate (Vcmax), electron transport-limited carboxylation rate (Jmax), and triose phosphate utilization-limited carboxylation rate (TPU)), nitrogen per unit leaf area (Narea), and leaf reflectance of seven taxa and 62 genotypes of Populus from two study plantations in Mississippi. For statistical modeling, we used least absolute shrinkage and selection operator (LASSO) and principal component analysis (PCA). Our results showed that the predictive ability of LASSO and PCA models was comparable, except for Narea in which LASSO was superior. In terms of model interpretability, LASSO outperformed PCA because the LASSO models needed 2 to 4 spectral reflectance wavelengths to estimate parameters. The LASSO models used reflectance values at 758 and 935 nm for estimating Vcmax (R2 = 0.51 and RMSPE = 31%) and Jmax (R2 = 0.54 and RMSPE = 32%); 687, 746, and 757 nm for estimating TPU (R2 = 0.56 and RMSPE = 31%); and 304, 712, 921, and 1021 nm for estimating Narea (R2 = 0.29 and RMSPE = 21%). The PCA model also identified 935 nm as a significant wavelength for estimating Vcmax and Jmax. Therefore, our results suggest that hyperspectral leaf reflectance modeling can be used as a cost-effective means for field phenotyping and rapid screening of Populus genotypes because of its capacity to estimate these physicochemical parameters.

Introduction

Production of bioenergy through direct combustion [1] and/or the manufacture of liquid biofuels [2] can reduce the consumption of fossil fuels and associated emissions [3]. In the USA, bioenergy contributed about 7.3% to the energy sector in 2019, which could increase to an additional 18–55% by 2050 [4]. To bolster this increasing trend, short rotation woody crops (SRWCs) can be planted as a biomass feedstock for bioenergy production [5, 6]. Currently in the USA, Populus feedstock production for bioenergy makes up a small portion of land use, but there is the potential for large-scale plantations for bioenergy in the future because Populus is a promising bioenergy crop. Eastern cottonwood (Populus deltoides W. Bartram ex Marshall) and hybrid poplars have preferable traits for producing high biomass yield within a short period of time [7-12]. The biomass yield of Populus genotypes can reach 7.5 to 15.2 Mg/ha/year [13]. Due to their rapid growth and high biomass productivity, the United States Department of Energy classifies Populus as a potential SRWC for bioenergy production [14]. A Populus plantation can be managed with a five-year harvesting cycle [10, 15–17], and their excellent coppicing ability [18-20] eliminates the need for replanting after each harvest. Since Populus has a wide range of genetic diversity [21, 22], clonal propagation [23, 24] and the use of genetically improved plant material [3, 25] can maximize biomass yield and will accelerate the development of the bioenergy sector [3, 25]. Moreover, Populus feedstocks have desirable attributes needed for producing ethanol and other biofuels, such as high cellulose content, moderate lignin and hemicellulose content, and low ash and extractives [26].

Tree breeding has been used to produce many Populus genotypes with the aim of maximizing biomass yield [27, 28]. Tree breeding is usually followed by field trials to compare growth performance across developed genotypes. Since field experiments are laborious, costly, and time-consuming [29], identifying superior genotypes for further testing is needed as an initial screening step to optimize field trials. To this end, photosynthetic capacity can be used as a selection criterion for clonal screening because individuals with enhanced photosynthetic rate under field conditions are likely to result in higher biomass yield [30]. To evaluate photosynthetic capacity, photosynthetic CO2 response curves can be used to calculate key photosynthetic capacity parameters, such as (1) maximum carboxylation rate limited by ribulose 1-5-bisphosphate carboxylase/oxygenase (RuBisCO) activity (Vcmax), (2) electron transport-limited carboxylation rate (Jmax), and (3) triose phosphate utilization-limited carboxylation rate (TPU) [31, 32]. In addition to photosynthetic capacity parameters, nutrient content in leaves is another important parameter for plant growth because limitations of nutrients can decrease photosynthesis. In plants, nitrogen (N) is a limiting macronutrient [33-35]. N deficiency can reduce leaf area [36], leaf durability, and leaf chlorophyll content. Additionally, leaf N content tends to be linearly correlated with photosynthetic capacity [37, 38], and consequently, variations in leaf N content can affect biomass production [39-41].

Photosynthetic capacity parameters and leaf N content provide valuable information about leaf-level processes. However, measuring photosynthetic CO2 response curves in the field takes about 15–20 minutes per leaf. Analogously, measuring leaf N content requires a laboratory assay to generate results. Therefore, these measurement techniques are not cost-effective for high-throughput screening and large-scale field phenotyping. On the other hand, hyperspectral leaf reflectance can potentially be used as an alternative to quickly estimate photosynthetic capacity parameters and leaf N content. Generally, leaf spectra have low reflectance in the visible region mainly due to chlorophyll absorption and high reflectance in the near-infrared region due to internal leaf scattering with little to no absorption [42]. Therefore, there is a physiological relationship between leaf reflectance and photosynthetic capacity parameters [43-45]. For example, according to Qian et al. [44], when Vcmax increases, the reflectance at the red edge position and the green peak decreases. Leaf reflectance can also be correlated with leaf N content because N-containing bonds cause variation in spectral features [46]. Ye et al. [47] found that when leaf N content increased, spectral reflectance decreased, and this relationship was more significantly observed in the green to near infrared regions. Due to these relationships, if robust statistical models can be developed, hyperspectral leaf reflectance can be used as a proxy for rapid assessment of photosynthetic capacity [48-50], as well as non-destructive and real-time monitoring of leaf N content [51, 52]. On the other hand, creating empirical models for hyperspectral data requires sophisticated model development techniques, and model robustness depends on the quality of sample data measured in the field [53].

While existing vegetation indices (e.g., normalized difference vegetation index, photochemical reflectance index, etc.) might be sufficient for estimating physicochemical parameters [43, 54–56], many studies have also developed more robust models using multiple regression methods to improve the prediction capacity for certain plant species [50, 57, 58]. Hyperspectral data have very narrow bandwidths with tremendous amounts of redundant information leading to potential issues with multicollinearity and high-dimensionality. In this respect, statistical methods, such as stepwise regression, shrinkage (e.g., ridge regression and least absolute shrinkage and selection operator (LASSO)), and dimension reduction (e.g., principal component analysis (PCA)) may be useful for handling hyperspectral data. Hyperspectral data also contain hundreds of spectral bands, and therefore hundreds of predictor variables. If the number of predictor variables is greater than the sample size, stepwise selection of multiple linear regression likely chooses nuisance variables rather than true variables, and consequently, performs poorly for out-of-sample accuracy [59, 60]. Ridge regression includes all predictor variables in the final model, which is disadvantageous when analyzing data containing a large number of predictor variables [59]. To solve these issues, many studies suggested LASSO [61-63] and PCA [64-66] methods. The PCA method is easy and computationally efficient. Its major drawback is a difficulty in interpreting principal components because they represent linear combinations of the original features, and thus, each component uses the collection of all predictor variables [67]. LASSO can overcome this limitation of PCA. Compared to PCA, LASSO is likely to generate a simpler model because LASSO uses regularization, shrinkage, and variable selection, and the final model includes a smaller set of predictor variables [59]. However, due to being a supervised feature extraction method [68-70], LASSO is more computationally demanding than PCA. Both LASSO and PCA methods have the capacity for analyzing hyperspectral data, but each method has its own strengths and weaknesses [71-74]. Therefore, comparing both methods on the same data can help to elucidate these tradeoffs, determine where the methods provide convergent information, and identify the method with better model performance.

In this study, we established two Populus plantations for field trials in upland sites of Mississippi each containing over 100 genotypes of Populus and its hybrids. We postulated that utilizing the large variability in leaf physicochemical parameters across different Populus genotypes and sites would result in robust models for Populus screening programs. Therefore, using leaf-level data from 62 Populus genotypes across seven taxa measured at two study plantations, we developed hyperspectral leaf reflectance models to estimate photosynthetic capacity parameters and leaf N content. Our main objectives were: (1) to assess whether hyperspectral leaf reflectance can adequately model photosynthetic capacity parameters and leaf N content of eastern cottonwood and hybrid poplar, (2) to compare the performance of models developed using LASSO and PCA methods, and (3) to identify spectral wavelengths that are the most sensitive to these physicochemical parameters. Successful development of predictive models from hyperspectral data can be applied to clonal screenings, enable high-throughput field phenotyping [50, 75], and thus, reduce the costs of Populus bioenergy research. Moreover, photosynthetic capacity parameters are input parameters in many terrestrial biosphere models because plant productivity depends on photosynthetic processes [76-79]. Therefore, information about these photosynthetic capacity parameters, which may be cost-effectively estimated from hyperspectral reflectance data, can be used as inputs for process-based earth system models to simulate terrestrial carbon fluxes [80-82] for improving our understanding of regional and global carbon cycles.

Materials and methods

Field and laboratory analysis

Study area

The study area consisted of two plantations containing over 100 Populus genotypes located in Monroe County (88° 17’ W, 33° 51’ N) and Pontotoc County (88° 59’ W, 34° 8’ N), MS. The plantation at the Monroe site was established in April 2018 at 2.7 × 1.8 m spacing. The average height of trees was 6.2 m at the end of the second growing season in 2019. The Monroe site has moderately well-drained, Prentiss fine sandy loam soils [83] and is surrounded by, and previously used for, row crop agriculture with 0–2 percent slopes. This site has soil carbon, soil nitrogen, organic matter, and dry bulk density of 2.29 kg/m2, 0.21 kg/m2, 1.08%, and 1.60 g/cm3, respectively. Based on climate data from a nearby weather station (USW00093862), the Monroe site has an annual temperature range of 11–23°C, and mean annual rainfall of 1,397 mm [84].

The plantation at the Pontotoc site was established in April, 2019 at 2.7 × 1.8 m spacing. The average height of trees was 5.3 m at the end of the second growing season in 2020. The Pontotoc site has well-drained, Atwood silt loam soils [83] and is also surrounded by, and previously used for, row crop agriculture with 0–2 percent slopes. This site has soil carbon, soil nitrogen, organic matter, and dry bulk density of 3.61 kg/m2, 0.44 kg/m2, 1.83%, and 1.52 g/cm3, respectively. Based on climate data from the Pontotoc Experimental Station, MS (USC00227111), the Pontotoc site has an annual temperature range of 10–22°C, and mean annual rainfall of 1,482 mm [84].

Photosynthetic A/Ci curves

Gas exchange measurements were conducted in the field using a LI-COR 6400 XTP portable photosynthesis system (LI-COR Biosciences Inc., Lincoln, NE, USA) with a red/blue light source attached. At the Pontotoc site, leaves were within reach, and measurements were made on live leaves still attached to branches on the mid canopy sunlit side of trees. At the Monroe site, leaves could not be accessed due to the height of trees, and a 7.8 m long pole pruner was used to cut branches from the mid to upper parts of crowns on the sunlit side of trees. The cut branches were immediately submerged in water, and approximately 20 cm of the branch end was recut underwater to minimize embolisms [85, 86]. Then, sample leaves attached to cut branches were measured in a sunlit location outside of the field plot. Fully expanded leaves with minimal disease located near the top of branches were selected for these measurements. Photosynthetic A/Ci response curves (A = net CO2 assimilation and Ci = sub-stomatal CO2 concentration inside the leaf) of the sample leaves were measured by setting ambient CO2 levels at 400 ppm and then, 300, 200, 100, and 50 ppm before being increased to 400, 600, and 800 ppm. During measurements, stomatal conductance was monitored. If low stomatal conductance occurred as a result of stress due to removal from the tree (in the case of the Monroe site), a new branch was collected and measured. Measurements were made between 10:00 AM and 2:00 PM on clear sunny days. For each individual during each measurement period, photosynthetic photo flux density, leaf temperature, and relative humidity inside the leaf chamber were controlled under saturating light conditions at 1500 μmol/m2/s, 30°C, and 40 to 60%, respectively [85-87]. From the A/Ci curves, we calculated three main photosynthetic capacity parameters–Vcmax, Jmax, and TPU in a Microsoft Excel solver program that contained non-linear curve-fitting equations from Sharkey et al. [88], which was based on Farquhar et al. [31]. During calculation, the parameters were scaled to a standardized temperature of 25°C.

We measured a total of 105 leaves (n = 105). Among them, we measured 57 leaves at the Monroe site in July and early September in 2019 and 48 leaves at the Pontotoc site in July in 2020. In total, about 85% of leaf samples were measured in July and 15% in early September. September measurements were made on previously measured genotypes from July because some spectral response curves from leaves measured in July were more variable than the majority of collected data, and thus, these genotypes were re-measured. At the Monroe site, we measured leaf samples from 21 genotypes and five taxa, including P. deltoides × P. deltoides (D×D, or eastern cottonwood), P. deltoides × P. maximowiczii A. Henry (D×M), P. deltoides × P. nigra L. (D×N), P. deltoides × P. trichocarpa Torr. & Gray (D×T), and P. trichocarpa × P. deltoides (T×D) (S1 Table). At the Pontotoc site, we measured leaf samples from 48 genotypes and six taxa, including D×D, D×M, D×N, the three-way hybrid Populus deltoides × Populus nigra × Populus maximowiczii ((D×N)×M), D×T, and T×M (P. trichocarpa × P. maximowiczii) (S1 Table).

Hyperspectral leaf reflectance

Immediately after gas exchange measurements of a leaf, hyperspectral leaf reflectance was measured on the same leaf so that both measurements were made in close proximity temporally and under the same environmental and light conditions. The spectral measurements were made with a portable, single-beam GER 1500 field spectroradiometer (Spectra Vista Corp., Poughkeepsie, NY, USA). The GER 1500 can detect a spectral range of 286.90 to 1089.62 nm and contains 512 spectral bands. It has full-width at half-maximum of 3.20 nm and a nominal bandwidth of approximately 1.50 nm covering parts of the ultraviolet, visible, and near-infrared wavelength regions of the electromagnetic spectrum. Decimal values of spectral wavelengths were rounded to integer values when reporting results.

To calculate the remote sensing reflectance of leaves, we collected two types of spectral data—radiance of the leaf and irradiance of the sky under sunny, cloud-free conditions. From approximately 40 cm above the leaf, radiances were measured on the leaf adaxial surface. Excluding the midrib, four radiance measurements were taken from various locations on each leaf lamina, and these readings were then averaged to obtain mean radiance for individual leaves. Irradiance of the sky was measured by holding the spectroradiometer and its attached fiber-optic cable with diffuser vertically toward the cloud-free sky. Six irradiance readings were recorded between individual radiance measurements, and these readings were then averaged to obtain mean irradiance of the sky. To calculate the reflectance of leaves, mean sky irradiance measured in close temporal proximity to individual radiance readings was chosen. The remote sensing reflectance for each individual leaf was then calculated by dividing mean radiance of the leaf with mean irradiance of the sky.

Leaf N content

Leaves measured for both photosynthetic capacity parameters and hyperspectral reflectance were placed in sealed plastic bags, kept inside a cooler, and transported to the laboratory for N determination. Fresh leaf area was measured using a LI-3100C Area Meter (LI-COR Biosciences Inc., Lincoln, NE, USA) in the laboratory. Leaves were then oven-dried at 60°C for at least 72 h until all water was evaporated from samples, and their dry weight was measured. Next, dried leaves were ground to a fine powder that could pass through a 0.25 mm sieve and analyzed for N content via elemental combustion analysis (ECS 4010 CHNOS Elemental Analyzer, Costech Analytical Technologies Inc., Valencia, CA, USA). Standards, duplicates, and blanks were included in the analysis for quality assurance. Since all photosynthetic capacity parameters were calculated on an area basis, we calculated leaf N content on a per unit area basis (Narea; g N/m2) by using leaf mass per area (LMA; calculated as the ratio of dried leaf weight and fresh leaf area) as follows:

Tree biomass

To determine if leaf physicochemical parameters were correlated with woody biomass and productivity, the measured Populus trees were harvested to estimate their biomass at the end of the second growing season. The freshly harvested trees were separated into leaf and wood components with fresh wood weights measured in the field and fresh leaf weights measured in the laboratory. A wood sample from each tree was also returned to the laboratory and its fresh weight was recorded. Wood samples were oven-dried at 65°C, and their dry weight was measured and used to calculate the dry mass of the field-weighed wood samples to determine dry woody biomass (kg) for each tree.

Statistical analysis

Taxa comparisons and correlations across parameters

We made statistical comparisons at the taxa level. We grouped the measured leaf sample data based on taxa. We added T×D genotypes into D×T, and the(D×N)×M genotype into D×N due to their small sample sizes (T×D, n = 2 leaf samples, and (D×N)×M, n = 1 leaf sample) (S1 Table). Therefore, we had five taxa groups. Since analysis of variance (ANOVA) assumptions, such as normal distribution of model residuals and homogeneity of variance across groups, were not met, a non-parametric Kruskal-Wallis test was performed to compare the differences in photosynthetic capacity parameters (Vcmax, Jmax, and TPU), Narea, and tree biomass among the five taxa groups and between the Monroe and Pontotoc sites using R version 4.0.4 [89]. If taxa groups differed significantly, a post-hoc Dunn’s Kruskal-Wallis multiple comparison test using the Bonferroni method was then performed to identify significant differences among the groups. To determine the relationship among Vcmax, Jmax, TPU, Narea, and tree biomass, we used all measured leaf sample data (n = 105) and calculated a Pearson correlation coefficient matrix across these parameters.

Model development: Least absolute shrinkage and selection operator

We used all measured leaf sample data (n = 105) to develop models for estimating leaf physicochemical parameters from hyperspectral data across measured Populus taxa and genotypes. We followed the least absolute shrinkage and selection operator (LASSO) model development method to create hyperspectral leaf reflectance models for each leaf physicochemical parameter. To address the high-dimensional features of hyperspectral data and to reach a sparse solution, LASSO cross validation [90] (i.e., an L1-regularized regression) was applied for variable selection. Predictor variables were hyperspectral wavelength bands, and the response variables were photosynthetic capacity parameters (Vcmax, Jmax, and TPU) and Narea.

All statistical analyses were performed in R version 4.0.4 [89]. Using R package ‘glmnet’ version 2.0.18 [91], k-fold cross validation for ‘glmnet’ was performed by specifying the number of folds as 10. Deviance, which used mean square error for Gaussian models, was assigned as the loss for cross validation. From the cross validation output, the tuning parameter lambda (λ) at minimum mean cross-validated error was selected. Once the optimal λ was determined, data were fit with a generalized linear model through penalized maximum likelihood using LASSO regularization [91]. LASSO minimized overfitting by shrinking the coefficients of variables that were not related with the response variable to zero and thus, the predictor variables with non-zero coefficients were selected for the linear model. Then, R2, root mean square error (RMSE), and root mean square percentage error (RMSPE; calculated as ) of the selected models were calculated.

For model validation, the data were divided into training and testing datasets using the ‘trainControl’ function available in R package ‘caret’ version 6.0–84 [92]. Training and testing data included 70% and 30% of the entire data, respectively. The control parameters for cross validation were defined as 10-fold cross validation with 20 repetitions [93]. Then, the averaged values of repeated 10-fold cross-validated RMSE and R2 of the model were calculated.

Model development: Principal component analysis

We used all measured leaf sample data (n = 105) for principal component analysis (PCA) model development. Using the ‘prcomp’ function in R, we conducted PCA to create hyperspectral leaf reflectance models for each physicochemical parameter. An unrotated P-mode PCA was conducted on the data matrix (spectral bands × leaves) to produce orthogonal principal components (PCs), of which those with eigenvalues greater than 1.0 were retained [94]. Two PCs were retained as they explained 98.2% of the variance (PC1 = 85.8% and PC2 = 12.4%). Next, linear regression was performed using PC1 and PC2 as predictor variables for each of the photosynthetic capacity parameters and Narea. Then, the performance of these linear models was evaluated with training and testing data using the methods described above for repeated 10-fold cross validation.

Results

Photosynthetic parameters, Narea, biomass, and leaf reflectance

Vcmax ranged from 67.8 to 368.8 with a mean of 184.4 ± 15.7 μmol/m2/s; Jmax ranged from 71.5 to 402.4 with a mean of 193.5 ± 17.3 μmol/m2/s; TPU ranged from 5.3 to 25.8 with a mean of 13.2 ± 1.2 μmol/m2/s; Narea ranged from 1.1 to 4.7 with a mean of 2.5 ± 0.1 g/m2; and age two, whole-tree dry woody biomass ranged from 779.4 to 7288.4 with a mean of 4001.9 ± 341.7 g. These were the results calculated from all measured leaf sample data (n = 105).

Among the five taxa groups, D×N had 31% lower Vcmax than D×T; D×D had 32% lower Jmax than D×T; and D×D had 31% lower TPU than D×T. For tree biomass, D×N and T×M were 39% and 54% lower than D×T, and 40% and 54% lower than D×D, respectively (Fig 1). There was no difference in Narea among the taxa (Fig 1). Generally, T×M had lower leaf reflectance especially when compared with D×N and D×T (Fig 2). The positions of peak and trough leaf reflectance were mainly observed within the range of 700–800 nm and 900–1000 nm, respectively (Fig 2 and S1 Fig).

Fig 1

Box plots showing the variability among five different Populus taxa (left panels) and between the Monroe and Pontotoc sites (right panels).

Bars with different letters indicate statistically significant differences (P-value < 0.05). Parameters include maximum carboxylation rate of Rubisco (Vcmax) by (a) taxa and (b) site, maximum electron transport limited carboxylation rate (Jmax) by (c) taxa and (d) site, triose phosphate utilization limited carboxylation rate (TPU) by (e) taxa and (f) site, nitrogen per unit leaf area (Narea) by (g) taxa, (h) site, and tree biomass by (i) taxa and (j) site. D×T and T×D taxa were combined and shown as D×T, and (D×N)×M was included in D×N (S1 Table). D×D = P. deltoides × P. deltoides, D×M = P. deltoides × P. maximowiczii, D×N = P. deltoides × P. nigra, D×N×M = P. deltoides × P. nigra × P. maximowiczii, D×T = P. deltoides × P. trichocarpa, T×D = P. trichocarpa × P. deltoides, and T×M = P. trichocarpa × P. maximowiczii.

Fig 2

Leaf reflectance curves of five different Populus taxa.

D×T and T×D taxa were combined and shown as D×T, and taxa (D×N)×M was included in D×N (S1 Table). D×D = P. deltoides × P. deltoides, D×M = P. deltoides × P. maximowiczii, D×N = P. deltoides × P. nigra, (D×N)×M = P. deltoides × P. nigra × P. maximowiczii, D×T = P. deltoides × P. trichocarpa, T×D = P. trichocarpa × P. deltoides, and T×M = P. trichocarpa × P. maximowiczii.

Trees at the Monroe site had higher photosynthetic capacity and leaf N content than those at the Pontotoc site. Trees at the Monroe site had 37% higher Vcmax, 44% higher Jmax, 45% higher TPU, 12% higher Narea, and 34% higher tree biomass than those at the Pontotoc site (Fig 1). Tree biomass was positively correlated (P-value < 0.05) with Vcmax, Jmax, and TPU, but not with Narea (P-value = 0.22). Narea was positively correlated with Vcmax, Jmax, and TPU (P-value < 0.05) (Table 1).

Table 1

Pearson correlation coefficient (r) matrix among tree biomass, Vcmax, Jmax, TPU, and Narea.

	Tree biomass	V cmax	J max	TPU	N area
Tree biomass	1.00
V cmax	0.33	1.00
J max	0.35	0.97	1.00
TPU	0.36	0.96	0.99	1.00
N area	0.12 ns	0.39	0.38	0.36	1.00

Non-significant correlations are denoted by “ns” superscript, and the rest with “bold” font are significant at P-value < 0.05.

Variable selection and model performance

The LASSO model selected the same two wavelengths, 758 and 935 nm to estimate both Vcmax and Jmax (Table 2 and S1 Fig). The TPU model selected three wavelengths, 687, 746, and 757 nm; and the Narea model selected four wavelengths, 304, 712, 921, and 1021 nm (Table 2 and S1 Fig). The LASSO model for Narea had the lowest RMSPE (21%), followed by TPU (31%), Vcmax (31%), and Jmax (32%). In terms of R2, the LASSO model for TPU was the highest (R2 = 0.56), followed by Jmax (R2 = 0.54), Vcmax (R2 = 0.51), and Narea (R2 = 0.29) (Table 2 and Fig 3).

Table 2

Selected predictor variables in the models and averaged 10-fold cross validation results.

Model	Model form	R 2	RMSPE	RMSE (Entire data)	Cross validated RMSE
					Training data	Test data
LASSO: Vcmax	-30.2 + (601.1 × ρ758.29) + (302.7 × ρ935.71)	0.51	31%	57.20	59.07	57.37
LASSO: Jmax	-44.6 + (796.2 × ρ758.29) + (200.57 × ρ935.71)	0.54	32%	60.98	61.68	60.64
LASSO: TPU	-6.4 + (-62.0 × ρ687.03) + (35.2 × ρ745.99) + (59.9 × ρ756.76)	0.56	31%	4.02	4.02	4.37
LASSO: Narea	2.0 + (28.3 × ρ303.51) + (-24.4 × ρ711.97) + (0.7 × ρ920.68) + (5.9 × ρ1021.44)	0.29	21%	0.51	0.53	0.51
PCA: Vcmax	184.4 + (-1.9 × PC1) + (-5.6 × PC2) + (-0.03 × PC1 × PC2)	0.52	31%	56.99	56.68	59.54
PCA: Jmax	193.5 + (-2.4 × PC1) + (-5.3 × PC2) + (-0.01 × PC1 × PC2)	0.53	32%	61.92	62.46	64.72
PCA: TPU	13.2 + (-0.2 × PC1) + (-0.3 × PC2) + (-0.0007 × PC1 × PC2)	0.50	33%	4.30	4.20	4.65
PCA: Narea	2.5 + (0.0008 × PC1) + (-0.04 × PC2) + (-0.0005 × PC1 × PC2)	0.15	23%	0.56	0.60	0.55

Training and testing data contained 70% and 30% of the entire data, respectively. ρ is spectral reflectance at the corresponding wavelengths.

Fig 3

Scatterplots showing the linear relationship between measured and estimated Vcmax, Jmax, TPU and Narea for the LASSO models: (a), (b), (c), and (d) and the PCA models: (e), (f), (g), and (h).

The solid black line is the regression line, and the dotted red line is the 1:1 line.

In the PCA method, PC loading in each spectral band indicated that PC1 did not have much variability, however, distinguishability mainly occurred in PC2 (Fig 4). For PC2, the wavelengths that had the highest positive loadings were 429, 431, 437, 439, and 414 nm, and those that had the highest negative loadings were 931, 928, 930, 934, and 927 nm. Both PC1 and PC2 loadings indicated that the wavelengths near 935 nm were significant (Fig 4), and the same wavelength was also selected by the LASSO models for estimating Vcmax and Jmax (Table 2).

Fig 4

Lollipop plot showing the loadings of all spectral wavelengths in (a) PC1 and (b) PC2.

As shown by vertical green dashed line, a significant peak and trough occurred at or near 935 nm, which was also selected by the LASSO models for estimating Vcmax and Jmax.

The PCA models had comparable RMSPE and R2 as the LASSO models for Vcmax, Jmax, and TPU. Only the PCA model for Narea had substantially lower R2 than the LASSO model for Narea (PCA = 0.15 vs. LASSO = 0.29) but RMSPE was comparable (PCA = 23% vs. LASSO = 21%) (Table 2). Among the PCA models, the PCA model that estimated Narea had the lowest RMSPE (RMSPE = 24%), followed by Vcmax (RMSPE = 31%), Jmax (RMSPE = 32%), and TPU (RMSPE = 33%) (Table 2 and Fig 3). The PCA model for Jmax had the highest R2 (R2 = 0.53), followed by models for Vcmax (R2 = 0.51), TPU (R2 = 0.50), and Narea (R2 = 0.15).

The model validation using repeated 10-fold cross validation indicated that there was a very slight difference in RMSEs among the training data, test data, and entire dataset for both LASSO and PCA models (Table 2). Therefore, both LASSO and PCA models were not over-fit. Cross-validation also indicated that RMSEs were generally higher, albeit modestly, for the PCA models compared with the LASSO models.

Discussion

Model performance

Our results showed promise in estimating photosynthetic capacity parameters (Vcmax, Jmax, and TPU) using hyperspectral leaf reflectance (Table 2). In both LASSO and PCA, the models for estimating Narea had the lowest R2, but RMSPE was smaller for Narea than that of the estimates of Vcmax, Jmax, and TPU (Table 2). The Narea model likely had the smallest RMSPE because of having smaller variance compared to that of Vcmax, Jmax, and TPU (Fig 1). The Narea model had the smallest R2 probably because Narea was weakly correlated with photosynthetic capacity parameters (Table 1) suggesting that a large portion of N in leaves was not involved in photosynthesis, and consequently, might not generate a strong spectral signal. In addition to Narea, we also developed a model for leaf N content per unit mass (i.e., Nmass (%); calculated by dividing leaf N mass with leaf sample mass and then, multiplying with 100). Nonetheless, the Narea model (R2 = 0.29 for LASSO and 0.15 for PCA) fit the data better than that of Nmass (%) (R2 = 0.07 for LASSO and 0.02 for PCA). Narea resulted in a better model likely because Narea incorporated the information of both Nmass and leaf mass per area in its calculation.

For screening, new genotypes are developed by breeding within and potentially between taxa. Therefore, creating models from taxa- or genotype-specific data may not be robust enough because new genotypes that need to be screened for productivity potential are being developed. In this study, we used all leaf sample data (n = 105) measured across 7 taxa and 62 genotypes while developing LASSO and PCA models, with many of these genotypes commonly selected for establishing short rotation plantations across different geographic regions. Our measurements at both study plantations represented a single growing season, but we measured a wide variety of taxa and genotypes to generate large variability in physicochemical parameters (Vcmax, Jmax, TPU, and Narea) and to develop robust models for predicting Populus parameters. Therefore, our hyperspectral leaf reflectance captured variability in physiological functioning across a diverse range of Populus taxa and genotypes.

LASSO and PCA model development techniques had similar capacity in estimating Vcmax, Jmax, and TPU. However, the LASSO models used only 2 to 4 wavelengths to estimate all response variables, while the PCA models used 2 PCs, including all 512 wavelengths. Consequently, it is easier to interpret the results of the LASSO models compared to the PCA models. On the other hand, the LASSO method requires more sophisticated analysis. Since LASSO is a supervised feature extraction method, it needs to define an optimal value for the tuning parameter (λ) using cross-validation procedures to adjust the balance between sparsity (i.e., the model containing few non-zero coefficients) and model prediction accuracy [59, 71]. Furthermore, LASSO selects only one variable from the group if predictor variables within a group are highly correlated [67, 95]. Despite these setbacks, our findings suggested that sparse statistical modeling with LASSO could simplify models by selecting a small subset of predictor variables, which was especially helpful for analyzing data with high dimensionality and complexity in which small and critical differences in data could not be easily detected [72]. Therefore, we found that the LASSO method was able to develop acceptable models for the hyperspectral data in which the number of predictor variables was significantly larger than the sample size.

In contrast to the LASSO method, the PCA method was an unsupervised dimension reduction technique. Consequently, it is easier and quicker to generate results with minimum input from the analyst. The PCA method transforms the high-dimensional space of correlated variables into a new feature space of high variance, and consequently can achieve the tasks of both feature and noise reduction [96]. However, the LASSO method does not transform the data, and consequently, the data might still contain noise [97]. The disadvantage of the PCA method is that, while constructing the principal components to minimize the correlations among them, the principal components do not necessarily tie in with the interest of specific model outcome [67]. Because all the variables exist in each principal component, it is harder to interpret the direct relationship between individual spectral wavelengths and leaf physicochemical parameters. The model performance between LASSO and PCA methods was equivalent in estimating photosynthetic capacity parameters, however, the LASSO model outperformed the PCA model in estimating Narea (Table 2). Therefore, the LASSO method was able to provide a simple and robust model for all response variables even though it required more time and effort during model development.

We also analyzed previously published vegetation indices in linear regressions by using our measured data in order to evaluate whether these vegetation indices were comparable with our LASSO and PCA models in estimating Vcmax, Jmax, TPU, and Narea. We calculated R2 and RMSE of vegetation indices and compared these parameters with our best-fit LASSO and PCA models. We found that published vegetation indices did not perform better than our models because our models, particularly the LASSO models, had the highest R2 and the lowest RMSE for estimating photosynthetic capacity parameters and Narea in our Populus trees (Table 3). This was likely because these vegetation indices were developed using other species and to estimate other physiological and leaf- and canopy-level parameters than we estimated. After LASSO and PCA models, the hyperspectral vegetation indices developed for estimating canopy-level growth parameters in cotton (Gossypium hirsutum L.) and leaf-level chlorophyll content in a wide range of plants [98-102] showed the next best performance for estimating Vcmax, Jmax, and TPU (Table 3 and S2 Table). Regarding Narea, the LASSO model performed the best. However, hyperspectral red edge index (ρ738/ρ720) [103] and linear red edge index [98] previously used for estimating leaf area index in wheat and canopy N in cotton, and vegetation indices developed by Maccioni et al. [102] and Le Maire et al. [101] for estimating leaf chlorophyll content in trees outperformed the PCA model (Table 3 and S2 Table).

Table 3

Comparison of model performance among the LASSO model, the PCA model, and published vegetation indices.

Source	Formula	V cmax		J max		TPU		N area
		R 2	RMSE	R 2	RMSE	R 2	RMSE	R 2	RMSE
LASSO	This study	0.51	57.20	0.54	60.98	0.56	4.02	0.29	0.51
PCA	This study	0.52	56.99	0.53	61.92	0.50	4.30	0.15	0.56
Carter et al. [104]	ρ695/ρ420	0.02ns	81.21	0.03ns	88.67	0.03ns	5.97	0.01ns	0.60
Carter et al. [104]	ρ695/ρ760	0.00ns	81.85	0.00ns	89.88	0.00ns	6.05	0.03ns	0.60
Datt [99]	ρ672/(ρ550×ρ708)	0.08	78.62	0.09	85.74	0.11	5.72	0.03ns	0.60
Gamon (Photochemical Reflectance Index) [105]	(ρ531-ρ570)/(ρ531+ρ570)	0.00ns	81.80	0.00ns	89.70	0.01ns	6.04	0.04ns	0.60
Gitelson and Merzlyak [106]	ρ750/ρ705	0.05	79.80	0.04	88.22	0.03ns	5.97	0.15	0.56
Gitelson and Merzlyak [107]	(ρ750-ρ705)/ (ρ750+ρ705)	0.03ns	80.53	0.02ns	88.88	0.02ns	6.00	0.13	0.57
Gitelson and Merzlyak [108]	ρ750/ρ550	0.06	79.28	0.04	87.94	0.03ns	5.96	0.11	0.57
Gitelson and Merzlyak [108]	ρ750/ρ700	0.02ns	80.88	0.01ns	89.31	0.01ns	6.03	0.09	0.58
Gitelson et al. (Green Normalized Difference Vegetation Index) [109]	(ρ750-ρ550)/(ρ750+ρ550)	0.04	80.23	0.03ns	88.76	0.02ns	6.00	0.10	0.58
Gitelson et al. [100]	1/ρ700	0.16	74.85	0.20	80.34	0.21	5.40	0.00ns	0.61
Gupta et al. (Hyperspectral Red Edge Index) [103]	ρ735/ρ720	0.06	79.54	0.05	87.59	0.04	5.93	0.17	0.55
Gupta et al. (Hyperspectral Red Edge Index) [103]	ρ741/ρ717	0.01ns	81.26	0.01ns	89.48	0.01ns	6.03	0.16	0.56
Gupta et al. (Hyperspectral Red Edge Index) [103]	ρ747/ρ708	0.06	79.49	0.05	87.84	0.04ns	5.95	0.17	0.55
Gupta et al. (Hyperspectral Red Edge Index) [103]	ρ738/ρ720	0.07	78.78	0.07	86.75	0.06	5.88	0.19	0.55
Le Maire et al. [101]	(ρ749-ρ720)-(ρ701-ρ672)	0.29	69.06	0.30	75.48	0.27	5.18	0.21	0.54
Maccioni et al. [102]	(ρ780-ρ710)/(ρ780-ρ680)	0.12	76.57	0.12	84.25	0.10	5.73	0.25	0.52
Penuelas et al. (Structure Insensitive Pigment Index) [110]	(ρ800-ρ445)/(ρ800-ρ680)	0.03ns	80.47	0.04	88.09	0.04	5.94	0.00ns	0.61
Raper and Varco (Linear Red Edge Index) [98]	700+40[(ρ670+ρ780)/2-ρ700]/ (ρ740-ρ700)	0.12	76.95	0.11	84.92	0.09	5.78	0.24	0.53
Sims and Gamon [111]	(ρ750-ρ445)/(ρ705-ρ445)	0.06	79.26	0.06	87.36	0.04	5.92	0.17	0.55
Sims and Gamon [111]	(ρ800–680)/(ρ800+ρ680–2ρ445)	0.03ns	80.64	0.04ns	88.29	0.03ns	5.95	0.00ns	0.61
Sims and Gamon [111]	(ρ800-ρ445)/(ρ680-ρ445)	0.01ns	81.36	0.01ns	89.38	0.02ns	6.00	0.03ns	0.60
Stimson et al. (Hyperspectral Normalized Difference Vegetation Index) [112]	(ρ860-ρ690) /(ρ860 +ρ690)	0.00ns	81.75	0.01ns	89.56	0.01ns	6.03	0.02ns	0.60
Vogelmann et al. [113]	ρ740/ρ720	0.07	78.85	0.07	86.82	0.06	5.88	0.19	0.55
Wen et al. [114]	ρ764/ρ716	0.00ns	81.71	0.00ns	89.88	0.00ns	6.06	0.15	0.56
Zarco-Tejada et al. (Red Edge Optical Index) [115]	ρ750/ρ710	0.06	79.28	0.05	87.63	0.04	5.93	0.18	0.55

Spectral reflectance is denoted by ρ, and non-significant correlations are indicated by the “ns” superscript. The models and vegetation indices having the lowest RMSE for estimating each parameter are bolded. The sources of the alphabetically ordered vegetation indices and their study objectives are described in S2 Table.

Interpretation of selected wavelengths

As with previous studies [32, 116], we found a strong positive correlation between Vcmax and Jmax (r = 0.97) (Table 1) likely due to their coordinated processes in CO2 assimilation. The LASSO models selected the same two wavelengths: 758 and 935 nm for estimating both Vcmax and Jmax (Table 2 and S1 Fig). PCA also indicated 935 nm as a significant wavelength in both PC1 and PC2 (Fig 4). In our leaf reflectance curves (Fig 2 and S1 Fig), peaks and troughs of spectral reflectance mainly occurred at 758 nm and 935 nm. The wavelength 758 nm is an important wavelength for photosynthesis as it corresponds with the sun-stimulated fluorescence emission spectrum in plants [117, 118]. The spike in reflectance near 758 nm (Fig 2 and S1 Fig) is the spectral signature caused by chlorophyll molecules that re-emit light to dissipate energy and return from excited to non-excited states [118]. In addition, chlorophyll fluorescence emission can be a sign of plant water stress, and steady-state chlorophyll fluorescence emission can reduce net CO2 assimilation [119-122]. According to Frankenberg et al. [123], Fraunhofer lines at 758.8 and 770.1 nm can be used for retrieving chlorophyll fluorescence. Satellite-derived solar-induced chlorophyll fluorescence data have also been suggested as a proxy for estimating and mapping Vcmax for global climate change studies [124-126]. Moreover, chlorophyll fluorescence emission can be an indicator of increased photorespiration, which can reduce the rate of photosynthetic capacity [127]. The near infrared wavelength 935 nm is commonly identified as a key wavelength in several vegetation indices. For example, this wavelength was found in the normalized difference vegetation index (NDVI = (ρ935 –ρ661)/(ρ935 + ρ661), in which ρ = spectral reflectance) used to evaluate the visual quality and the drought stress responses of turfgrasses [128-130] as well as to explore the photosynthetic responses under varying salinity stresses in seashore paspalum (Paspalum vaginatum Swartz) ecotypes [131]. Furthermore, according to Wang et al. [132], 935 nm was sensitive to leaf greenness and chlorophyll content [133]. Moreover, 935 nm was associated with thermal dissipation of absorbed energy under stress conditions. For example, Zhang et al. [134] observed that non-photochemical quenching (i.e., a mechanism used by plants to protect from adverse effects of high light intensity by dissipating the excessively absorbed radiant energy into heat in the photosystem II antenna complexes) was negatively correlated with the spectral reflectance range 935–945 nm under salinity stress. There was also a significant inverse relationship between non-photochemical quenching and light-adapted fluorescence yield [135, 136] because the fluorescence signal could originate from photosystem II [137]. In addition to its correlation with the leaf chlorophyll fluorescence signal under different abiotic stresses [66, 134], reflectance at 935 nm is related with the water content in plant leaves [138] and could provide potential for understanding leaf water status [139].

The TPU model selected remote sensing reflectance at 687, 746, and 757 nm (Table 2 and S1 Fig). Due to its capacity to detect chlorophyll fluorescence signal, the wavelength 757 nm has been widely used for modeling terrestrial gross primary productivity from space [140-145]. In addition, 687 nm is also used for sun-induced fluorescence retrieval for estimating leaf- and canopy-level net photosynthesis of vegetation [117, 146–149]. The absorption maximum positions of the excitation spectra of fluorescence also occurred at 746 nm in pea plants [150]. The wavelength 746 nm is also associated with photosynthetic processes. Traditionally, the wavelength range of 400–700 nm is considered photosynthetically active radiation (PAR). However, recent studies suggested wavelengths 711 nm, 723 nm, and 746 nm as extended PAR wavelengths as well [151, 152] because these far-red wavelengths also manifested comparable efficiency at driving canopy photosynthesis and yield of both C3 and C4 species and their cultivars [152]. The reflectance at 746 nm may distinguish levels of accessary pigments with Sonobe et al. [153] finding that reflectance at 746 nm had a strong negative correlation with carotenoid concentration in shade-grown tea leaves. Therefore, the predictors included in our TPU model (687, 746, and 757 nm) are in agreement with formerly proposed wavelengths for estimating photosynthetic capacity.

Our hyperspectral leaf reflectance model for estimating Narea had lower R2 (0.29 for the LASSO model and 0.15 for the PCA model) than models for estimating Vcmax, Jmax, and TPU (Table 2). Nevertheless, all four wavelengths selected in our Narea LASSO model have been reported elsewhere as critical wavelengths, which were largely associated with, and also useful for, estimating leaf N content in different plant species. Our best model for estimating Narea contained four wavelengths– 304, 712, 921, and 1021 nm (Table 2 and S1 Fig). The first wavelength, 304 nm, is in the ultraviolet-B (UV-B) region, which extends from 280 to 320 nm [154-156]. Leaf protective chemicals have been shown to absorb UV-B radiation at 305 nm [157]. Leaves with more protective chemicals to absorb UV-B radiation might have less N content because protective compounds (e.g., flavonoid content) in plants were found to have a negative relationship with leaf N content [158]. The wavelength near 712 nm was also highly correlated with total N content in cotton plants (Gossypium hirsutum L.) [159]. In Chinese cabbage (Brassica campestris L. ssp. Pekinensis ‘Norgangbom’) leaves, 710 nm was correlated with chlorophyll content and was also a significant wavelength for predicting leaf N [160]. According to Zhao et al. [41], changes in reflectance at 710 nm could detect levels of N deficiency because of its negative relationship with leaf N in corn (Zea mays L.). Besides agricultural crops, 711 nm was one of the correlated spectral bands with leaf N in balsam fir (Abies balsamea L. Mill) [161]. The wavelength near 921 nm can be a leaf N predictor as well. Zhao et al. [162] found that the reflectance at 920 nm was associated with leaf N concentration and canopy N density in winter wheat (Triticum aestivum L.), and the reflectance at this wavelength was controlled by leaf structural features. Moreover, Tarpley et al. [163] used 920 nm for estimating leaf N content in cotton, and Yu et al. [164] used 921 nm for estimating whole-plant total (including leaves, stems, and roots) N content in pepper (Capsicum annuum L.) plants. The wavelength near 1021 nm also showed a relationship with leaf N content. Due to the N-H stretch from protein (i.e., the stretching vibrations of the strong molecular bonds between hydrogen atoms and nitrogen atoms), the N absorption feature centers at 1020 nm in leaves [165]. Moreover, 1020 nm was used as a N absorption feature for estimating canopy N in sagebrush (Artemisia spp.) [166] and as a protein absorption feature for estimating leaf N concentrations in grass [167]. Therefore, previously documented studies support the wavelengths selected in our LASSO models for estimating Vcmax, Jmax, TPU, and Narea.

Conclusion

LASSO and PCA model development methods showed similar capacity for estimating photosynthetic capacity parameters, but the LASSO model outperformed the PCA model in estimating Narea. For analyzing hyperspectral data, the LASSO models resulted in superior model interpretability compared to the PCA models. With only a few spectral wavelengths, narrowband hyperspectral reflectance can estimate photosynthetic capacity and leaf N content of highly variable eastern cottonwood and hybrid poplar genotypes. This is particularly critical when the most important spectral bands are adjacent to each other (within about 10 nm in TPU model; Table 2) suggesting that high spectral resolution is needed for spectral precision and modeling biophysical traits. Previous physiological studies also found that wavelengths selected by our LASSO models were strongly correlated with photosynthetic capacity and leaf N content. The results of our single-leaf studies corroborated that sun-induced chlorophyll fluorescence retrieval bands (687 and 758 nm) [117, 146–149] could predict Vcmax, Jmax, and TPU. This suggests that sun-induced fluorescence techniques can be successful in estimating Vcmax and Jmax of Populus plantations at the ecosystem level. In total, hyperspectral leaf reflectance can be used as a cost-effective means for high-throughput phenotyping and rapid clonal screening due to its ability to model photosynthetic capacity and leaf N content of a wide variety of Populus genotypes.

Locations of wavelengths selected by the LASSO models for estimating Vcmax, Jmax, TPU, and Narea.

D×T and T×D taxa were combined and shown as D×T, and (D×N)×M taxa was included in D×N (S1 Table). D×D = P. deltoides × P. deltoides, D×M = P. deltoides × P. maximowiczii, D×N = P. deltoides × P. nigra, (D×N)×M = P. deltoides × P. nigra × P. maximowiczii, D×T = P. deltoides × P. trichocarpa, T×D = P. trichocarpa × P. deltoides, and T×M = P. trichocarpa × P. maximowiczii.

(DOCX)

Click here for additional data file.

List of taxa and genotypes measured at the Monroe and Pontotoc sites.

Monroe and Pontotoc columns describe the number of leaves on which measurements were made and measurement dates. There were a total of seven taxa and 62 genotypes measured at both study sites.

(DOCX)

Click here for additional data file.

Study objectives of selected vegetation indices.

(DOCX)

Click here for additional data file.

23 Dec 2021

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.

A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.

An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols.

We look forward to receiving your revised manuscript.

Kind regards,

Janusz J. Zwiazek

Academic Editor

PLOS ONE

Journal Requirements:

1. When submitting your revision, we need you to address these additional requirements.

Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at

https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and

https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf

2.  Please include your tables as part of your main manuscript and remove the individual files. Please note that supplementary tables (should remain/ be uploaded) as separate "supporting information" files.

3. Thank you for stating in your Funding Statement:

( This material is based upon work that is supported by the National Institute of Food and Agriculture, U.S. Department of Agriculture under award numbers 2018-67020-27934 to HJR and CMS, and 2018-68005-27636 to HJR and CMS, as well as U.S. Department of Agriculture McIntire Stennis Program under accession numbers: MISZ-067050 to HJR, MISZ-032100 to CMS, and MISZ-0621210 to KPP.

The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.)

Please provide an amended statement that declares *all* the funding or sources of support (whether external or internal to your organization) received during this study, as detailed online in our guide for authors at http://journals.plos.org/plosone/s/submit-now.  Please also include the statement “There was no additional external funding received for this study.” in your updated Funding Statement.

Please include your amended Funding Statement within your cover letter. We will change the online submission form on your behalf.

4. We note that you have indicated that data from this study are available upon request. PLOS only allows data to be available upon request if there are legal or ethical restrictions on sharing data publicly. For more information on unacceptable data access restrictions, please see http://journals.plos.org/plosone/s/data-availability#loc-unacceptable-data-access-restrictions.

In your revised cover letter, please address the following prompts:

a) If there are ethical or legal restrictions on sharing a de-identified data set, please explain them in detail (e.g., data contain potentially sensitive information, data are owned by a third-party organization, etc.) and who has imposed them (e.g., an ethics committee). Please also provide contact information for a data access committee, ethics committee, or other institutional body to which data requests may be sent.

b) If there are no restrictions, please upload the minimal anonymized data set necessary to replicate your study findings as either Supporting Information files or to a stable, public repository and provide us with the relevant URLs, DOIs, or accession numbers. For a list of acceptable repositories, please see http://journals.plos.org/plosone/s/data-availability#loc-recommended-repositories.

We will update your Data Availability statement on your behalf to reflect the information you provide.

5. Please include your tables as part of your main manuscript and remove the individual files. Please note that supplementary tables (should remain/ be uploaded) as separate "supporting information" files.

Additional Editor Comments:

Both reviewers agree that this is a potentially a valuable contribution. However, they raise a number of important points that need to be addressed by the authors before the manuscript can be recommended for publication.

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Yes

Reviewer #2: Yes

**********

2. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: Yes

Reviewer #2: Yes

**********

3. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

Reviewer #2: Yes

**********

4. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #2: Yes

**********

5. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: The authors proposed using hyperspectral leaf reflectance to estimate the photosynthetic capacity and nitrogen content, and modeled it by LASSO and PCA, so as to verify that hyperspectral reflectance modeling can be used as a cost-effective means to conduct extensive field phenotyping to quickly screen Populus genotypes. The language is fluent and there are few small mistakes. The methods are innovative, with strong innovation, reliable experimental design and detailed data, however, there are some problems, such as the lack of organization of each part and disorderly writing. The manuscript should be substantial revisions.

Introduction

The second paragraph (Line 65), is necessary to strengthen the importance of "photosynthetic parameters" and "leaf nitrogen content".

The third paragraph (Line 87) needs to introduce the relationship between spectral data and "photosynthetic parameters", "leaf nitrogen content per unit area". For example, how does the spectral data behave when the net photosynthetic rate is high or leaf nitrogen is high? And clarify the advantages and disadvantages of the field experiments and hyperspectral leaf reflectance.

The fourth paragraph (Line 104) puts forward many methods for analyzing hyperspectral data. Why did the author choose LASSO and PCA? To the best of my knowledge, the stepwise regression of multiple linear regression and ridge regression are also good methods to deal with multicollinearity. What are the advantages and disadvantages of the two models selected? These should be described in the article.

The last paragraph (Line 123) of Introduction should add some introduction concerning the importance of poplar in Mississippi, such as planting area, age of forest, and annual yield and so on.

Materials and Methods

Line 164, photosynthetic traits measurement should be described more detailed, measurement time and conditions should be clarified. See Liu et al. (2021).

Liu X, Zhang Q, Song M, Wang N, Fan P, Wu P, Cui K, Zheng P, Du N, Wang H, and Wang R. Physiological response of Robinia pseudoacacia and Quercus acutissima seedlings to repeated drought-rewatering under different planting methods. Frontiers in Plant Science; 2021; 12:760510.

Line 224, why this study used leaf N content on a per unit area instead of leaf N content on a per unit mass?

Discussion

There are too many paragraphs in Discussion, and the meaning of some paragraphs is unclear, for example, what is the significance of the interpretation of the wavelength at 935 nm (Line 423)? It is necessary to sort out the structure again. The paragraphs on Line 452 and 466 are the discussion of leaf nitrogen content, which can be combined.

It is suggested to separate the Discussion in some parts including, (1) the relationship between traits and spectral data, (2) the discussion of LASSO model, (3) the discussion of PCA model, (4) a comparison among the two models and any other models. The parts above are for your reference only, but they are not appropriate as subtitles. The authors may either divide the discussion into several parts according to the content; or merges paragraphs with similar content.

Reviewer #2: The manuscript reports a study that estimated the photosynthetic capacity and nitrogen content of eastern cottonwood and hybrid Poplar genotypes using hyperspectral leaf reflectance. The results help quickly screen Poplar genotypes with maximum yield potential. However, the manuscript has some significant flaws with the method description. A major revision should be conducted before considering for publication in PLOS ONE. Key concerns with the manuscript are summarised below.

1, What are the soil physicochemical properties (i.e., total N content, organic matter) in those study areas, i.e., Monroe County, Pontotoc County；

2, The authors need to provide more detailed information about the gas exchange measurement, i.e., leaf position in the branches, were the selected leaves fully expanded?

3, How many times did the authors determine the photosynthetic parameters in mid-July and early September in 2019 at the Monroe site and mid-July 2020 at the Pontotoc site? As we know, every time the photosynthetic parameters of those 62 genotypes were determined, the measurement should be conducted at a same time or during the same period, in order to make the comparison between different genotypes reasonable, however, the mid-July and early September in 2019 were a long period.

4, How many repetitions of the photosynthetic parameters measurement did the authors conducted for each genotypes? According to the description in the manuscript, the repetitions of the photosynthetic parameters measurement for each genotypes seemed less than 3, which was not sufficient for the statistical analysis.

5, In the Fig.1, Fig.3, Table 1 and Table, how many values for each parameter of each Populus genotype were used for the statistical analyses, i.e., n=?

6, The leaf spectra has been successfully used for representing the leaf N content, but in this research, why the R2 of the LASSO or PCA model for Narea was the lowest compared with that of Vcmax, Jmax and TPU (Table 3)? The value was only 0.29 or 0.15, it seemed that the leaf spectral reflectance showed weak correlation with N conten.

7, The hyperspectral leaf reflectance would be easily influenced by the leaf growth status and its surroundings, were all the parameters determined at the same time and in the same conditions, if not, how did the authors ensure that the leaf reflectance varied just as the photosynthetic parameters changed? And was the establishment of the relationship between these parameters reasonable?

8, The authors intended to establish methods for rapidly estimating those physicochemical parameters of different Populus genotypes, and quickly screening genotypes with maximum yield potential. In this research, there were 62 Populus genotypes, maybe there should be 62 group data of photosynthetic parameters, leaf N content and hyperspectral leaf reflectance values, respectively. However, only 7 group data were showed in Figure 1. Did the authors establish the correlations between these two parameters by using these 7 group data or 62 group data?

**********

6. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: No

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org. Please note that Supporting Information files do not need this step.

6 Feb 2022

We thank two anonymous reviewers for their helpful comments and constructive suggestions for improving our manuscript for publication. We addressed the questions of each reviewer and made a substantial revision in this revised manuscript. We also thank the academic editor for giving us the opportunity to revise our manuscript.

Submitted filename: Response to Reviewers.docx

Click here for additional data file.

17 Feb 2022

Using hyperspectral leaf reflectance to estimate photosynthetic capacity and nitrogen content across eastern cottonwood and hybrid poplar taxa

PONE-D-21-35240R1

Dear Dr. Kyaw,

We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements.

Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication.

An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at http://www.editorialmanager.com/pone/, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to-date. If you have any billing related questions, please contact our Author Billing department directly at authorbilling@plos.org.

If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org.

Kind regards,

Janusz J. Zwiazek

Academic Editor

PLOS ONE

Additional Editor Comments (optional):

Reviewers' comments:

1 Mar 2022

PONE-D-21-35240R1

Using hyperspectral leaf reflectance to estimate photosynthetic capacity and nitrogen content across eastern cottonwood and hybrid poplar taxa

Dear Dr. Kyaw:

I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department.

If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact onepress@plos.org.

If we can help with anything else, please email us at plosone@plos.org.

Thank you for submitting your work to PLOS ONE and supporting open access.

Kind regards,

PLOS ONE Editorial Office Staff

on behalf of

Professor Janusz J. Zwiazek

Academic Editor

PLOS ONE