Next Article in Journal
Rotary Tillage Plus Mechanical Transplanting Practices Increased Rice Yields with Lower CH4 Emission in a Single Cropping Rice System
Previous Article in Journal
Lime Application Reduces Methane Emissions Induced by Pig Manure Substitution from a Double-Cropped Rice Field
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Agriculture
Volume 14
Issue 7
10.3390/agriculture14071064
agriculture-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles thumb_up
...
Endorse textsms
...
Comment
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Estimating Leaf Chlorophyll Content of Winter Wheat from UAV Multispectral Images Using Machine Learning Algorithms under Different Species, Growth Stages, and Nitrogen Stress Conditions

by
Liyuan Zhang
1,
Aichen Wang
1,
Huiyue Zhang
1,
Qingzhen Zhu
1,
Huihui Zhang
2,
Weihong Sun
1 and
Yaxiao Niu
1,*
1
School of Agricultural Engineering, Jiangsu University, Zhenjiang 212013, China
2
Water Management and Systems Research Unit, USDA-ARS, 2150 Centre Avenue, Bldg. D., Fort Collins, CO 80526, USA
*
Author to whom correspondence should be addressed.
Agriculture 2024, 14(7), 1064; https://doi.org/10.3390/agriculture14071064
Submission received: 29 May 2024 / Revised: 23 June 2024 / Accepted: 29 June 2024 / Published: 1 July 2024
(This article belongs to the Section Digital Agriculture)
Browse Figures
Versions Notes

Abstract

:
The rapid and accurate estimation of leaf chlorophyll content (LCC), an important indicator of crop photosynthetic capacity and nutritional status, is of great significance for precise nitrogen fertilization management. To explore the existence of a versatile regression model that can be successfully used to estimate the LCC for different varieties under different growth stages and nitrogen stress conditions, a study was conducted in 2023 across the growing season for winter wheat with five species and five nitrogen application levels. Two machine learning regression algorithms, support vector machine (SVM) and random forest (RF), were used to establish the bridge between UAV-derived multispectral vegetation indices and ground truth LCC (relative chlorophyll content, SPAD), taking the multivariate linear regression (MLR) algorithm as a reference. The results show that the visible atmospherically resistant index, vegetative index, and normalized difference vegetation index had the highest correlation with ground truth LCC, with a Pearson’s correlation coefficient of 0.95. All three regression algorithms (MLR, RF, and SVM) performed well on the training dataset (R2: 0.932–0.944, RMSE: 3.96 to 4.37), but performed differently on validation datasets with different growth stages, species, and nitrogen application levels. Compared to winter wheat species and nitrogen application levels, the growth stages had the greatest influence on the generalization ability of LCC estimation models, especially for the dough stage. At the dough stage, compared to MLR and RF, SVM performed best, with R2 increasing by 0.27 and 0.10, respectively, and RMSE decreasing by 1.13 and 0.46, respectively. Overall, this study demonstrated that the combination of UAV-derived multispectral VIs and the SVM regression algorithm could be successfully applied to map the LCC of winter wheat for different species, growth stages, and nitrogen stress conditions. Ultimately, this research is significant as it shows the successful application of UAV data for mapping the LCC of winter wheat across diverse conditions, offering valuable insights for precision nitrogen fertilization management.

1. Introduction

Winter wheat is one of the most important staple crops in China and is of great significance to national grain security and sustainable social development [1]. Chlorophyll, the pigment used to absorb light energy during photosynthesis, plays a central role in light uptake in photosynthesis and is closely related to crop nitrogen nutrition status, photosynthetic capacity, crop yield [2]. As a key parameter reflecting crop growth status, the timely and accurate estimation of winter wheat LCC is of great significance and will be helpful for farmers in guiding precise field management, especially fertilization management.
To obtain crop LCC, traditionally, leaf samples should be firstly destructively sampled in the field and then be brought to the lab to conduct chemical analysis. During the chemical analysis, chlorophyll is susceptible to decomposition by light, leading to inaccurate LCC measurements. In addition, the above chemical analysis method also has the disadvantages of being expensive, time-consuming, and laborious [3]. As a non-destructive method, the portable handheld chlorophyll meter (SPAD-502) has been widely used to estimate LCC on the basis of crop light absorption detection [4]. The SPAD reading can be easily obtained by clipping the crop leaf using this handheld chlorophyll meter, and has been confirmed to have the ability to represent LCCs measured by physicochemical experiments [5,6]. However, when it comes to estimating LCC in large fields for precision crop management, the SPAD method is also time-consuming, laborious, and costly, and cannot represent the spatial variability of crop growth status well.
In recent years, remote sensing technology has been an advanced tool to obtain crop growth features [7,8] and production properties [9].The popular remote sensing platforms are satellite, ground-based vehicle, and unmanned aerial vehicle (UAV). Although the satellite remote sensing platform has been used in the agriculture applications, its potential ability to monitor crop growth at the field scale is limited by the coarse spatio-temporal resolution of images. The ground-based vehicle method is time-consuming and inefficient when crop monitoring in a large area is required. The UAV remote sensing platform is widely used in precision agriculture at the field scale with the advantages of flexibility, cost-effectiveness, and high spatio-temporal resolution [10,11,12], which makes it possible to obtain high-resolution LCC distributions [13]. In addition, the flight parameters of the UAV can be adjusted according to the area of the site and image resolution requirements to achieve high efficiency. Due to changes in the pigment level, dissimilar reflectance characteristics of healthy and non-healthy crop plants lay the foundation for mapping the LCC based on UAV-derived spectral information of the crop canopy [14,15]. Compared to the expensive UAV hyperspectral system, the UAV multispectral system is a low-cost choice to map LCC. For example, maize LCC was monitored in Zhuozhou City, Hebei Province, China by different vegetation indices (VIs), which were derived from UAV multispectral images with the highest coefficient of determination (R2) value of 0.8402 [16].
In previous studies, the SPAD readings were generally taken as ground truth of crop LCC, and regression algorithms were used to establish the bridge between UAV-derived multispectral VIs and SPAD readings. With the development of crop sensors and artificial intelligence [17], the machine learning algorithm has been an important tool in crop growth status monitoring [18,19]. For example, among three machine learning algorithms, random forest (RF), partial least squares regression, and artificial neural network, the model constructed using the RF regression algorithm performed best in wheat yield estimation accuracy [20]. In maize fractional vegetation estimation, RF proved to be the optimal model with the best applicability to data collected at different treatments in different growing seasons [21]. In potato LCC estimation, the models built with RF and support vector machine (SVM) regression algorithms showed a much better accuracy than partial least squares and ridge regression models [22]. Among these machine learning regression algorithms, the SVM and RF algorithms have proved to be powerful tools to monitor crop growth status based on UAV-derived parameters.
However, most of the previous LCC estimation studies focused on single crop species [23] and mostly in the flowering or early filling stages [24]. Few studies were conducted with multiple crop species to explore the ability of machine learning algorithms to map LCC based on UAV-derived multispectral VIs across multiple critical growth stages, especially the reproductive growth stage. Whether the LCC model could be used for different crop species and multiple key growth stages is one of the key constraints to determining whether it could be used to guide precise field management in practical planting scenarios, since different species may be planted, and the corresponding field management is required at each key growth stage. For example, LCC maps of winter wheat could be used to instruct precision spring nitrogen application, with a lower LCC value indicating more nitrogen fertilizer should be applied. In addition, whether different levels of abiotic stress would influence LCC estimation performance has not been clearly addressed in previous studies. It is also necessary to explore the influence of abiotic stress on the ability of machine learning algorithms to map LCC based on UAV-derived multispectral VIs, since abiotic stress has an impact on crop growth status and thus on canopy reflectance features, etc.
We considered the fact that the influences of winter wheat species, growth stage, and abiotic stress have not been comprehensively addressed in previous studies where UAV multispectral images and machine learning regression algorithms were adopted to establish an LCC estimation model. This study conducted a field experiment in 2023 across the growing season for winter wheat with five species and five nitrogen application levels and aims to explore the existence of a versatile regression model that can be successfully used to estimate the LCC for different varieties, growth stages, and nitrogen stress conditions. The specific objectives of this study were (1) to evaluate the performances of UAV-derived multispectral VIs in estimating the LCC of winter wheat; (2) to compare the performances of SVR and RF regression algorithms in estimating the LCC of winter wheat; and (3) to explore the influences of winter wheat species, growth stage, and nitrogen stress conditions on the LCC estimation performance using machine learning algorithms and UAV-derived multispectral VIs.

2. Materials and Methods

2.1. Field Site and Experimental Design

The experiment was conducted on five different winter wheat species under five nitrogen application treatments during the winter wheat growing season of 2022–2023 in Zhenjiang, Jiangsu province, China (31°57′48.9″ N, 119°17′54.5″ E, elevation 17 m). The average elevation was 28.4 m, and the average annual temperature and precipitation over the past 30 years were 16.3 °C and 1101.4 mm, respectively. The soil of this research field is loamy sand using the United States Department of Agriculture soil taxonomy with percentages of 1.27%, 10.42%, and 88.31% for clay, powder, and sand, respectively. The soil pH and organic matter content were 6.35 and 18.8 g/Kg. The total nitrogen, phosphorus, and potassium content before the experiment began were 0.438, 0.460, and 13 g/kg, respectively. Winter wheat was planted on 10 November 2022 and was harvested on 7 June 2023. Figure 1 shows the geographical location of the experimental site and overview of the winter wheat experimental design. The experimental field had a total of 125 plots with 4.8 m2 each, including 25 treatments with 5 repetitions (R1, R2, R3, R4, and R5) for each treatment. The 25 treatments consisted of five nitrogen levels and five winter wheat species (Figure 1). Fertilizer was applied to soil on 2 December 2022 according to five nitrogen levels, which were 0 kg/ha (N0), 90 kg/ha (N1), 180 kg/ha (N2), 270 kg/ha (N3), and 360 kg/ha (N4). The five winter wheat species were Yangmai29 (S1), Zhenmai15 (S2), Ningmaizi216 (S3), Yangmai25 (S4), and Zhenmai168 (S5).

2.2. Field Data Collection

In this study, a frequently used relative chlorophyll content method of SPAD (Soil and Plant Analyzer Development) was adopted to characterize the leaf chlorophyll content (LCC) of winter wheat. Specifically, the SPAD 502PLUS chlorophyll meter (Konica Minolta, Tokyo, Japan) was used to measure the LCC values of winter wheat plants. It is worth noting that calibration was required before measuring the LCC values of winter wheat plants using the chlorophyll meter. Calibration of the SPAD 502PLUS chlorophyll meter can be achieved by closing the measuring head with nothing in the sample tank. The average of SPAD values measured on the tip, middle, and base section of the flag leaf of each winter wheat plant was taken as the LCC value of each winter wheat plant. Three winter wheat plants were selected in each plot diagonally and their LCC values were averaged for each plot. During the winter wheat growing season (Table 1), a total of 500 LCC values were collected for each plot on 20 Apr 2023 (flowering stage), 30 Apr 2023 (filling stage), 8 May 2023 (milk stage), and 23 May 2023 (dough stage), respectively.

2.3. Collection and Pre-Processing of UAV Image Data

UAV multispectral images were collected using RedEdge-MX dual camera (MicaSense, Inc., Seattle, WA, USA) mounted on DJI M300RTK UAV platform (Da-Jiang Innovations, Shenzhen, China) on sunny days between 11:00–13:00. The 10 spectral bands of RedEdge-MX dual camera are Blue-444 (28 nm), Blue-475 (32 nm), Green-531 (14 nm), Green-560 (27 nm), Red-650 (16 nm), Red-668 (14 nm), Rededge-717 (12 nm), Rededge-705 (10 nm), Rededge-740 (18 nm), and NIR-842 (57 nm), respectively. The spatial resolution of the multispectral images is 1280 × 960.
The image acquisition was conducted at the flowering stage (April, 20), filling stage (April, 30), milk stage (May, 08) and hard dough stage (May, 23). The flight was controlled using DJI GS Pro station, which directed the UAV flying at a height of 30 m and speed of 1 m/s. Both the forward and side overlaps were 85%. The ground sample distance was 2.1 cm. To perform radiometric correction for UAV multispectral images, a diffuse reflector with reflectivity of 30% and size of 1.5 m by 1.5 m was adopted and placed next to the experimental field. UAV multispectral images of the above diffuse reflector were collected simultaneously at the same height during each UAV campaign.
The image mosaic processing was performed using the Pix4DMapper software (Pix4D Inc., Prilly, Switzerland, version 4.5.6). The geographic information of images was corrected by the coordinates of ground control points, which were measured using a RTK differential GNSS device (CHCNAV, Shanghai, China) with horizontal accuracy of 8 mm + 1 ppm RMS and vertical accuracy of 15 mm + 1 ppm RMS. After generating orthophoto and geo-correcting in Pix4DMapper, radiometric correction was required to obtain the reflectance in each band. The radiometric correction of UAV multispectral images was achieved using the images of the diffuse reflector captured at the same time. Finally, shapefile data of 125 plots (Figure 1) were obtained manually, which were used to calculate the vegetation indices corresponding to each plot using R language (version 4.1.3).

3. Methods

3.1. Vegetation Indices’ Selection and Calculation

For crop leaves, the strong absorption characteristic appears in blue and red bands due to the impact of pigment content, and the strong reflection characteristic occurs in the near-infrared band due to the internal structure of leaves [25]. Therefore, the vegetation indices derived from the multispectral images could be used to establish the winter wheat LCC estimation model. A total of 18 VIs was calculated based on the visible, NIR, and red-edge bands acquired from the multispectral images. They were divided into three categories: (1) VIs with only the visible bands (i.e., the blue, green, and red bands); (2) VIs with the NIR band but excluding the red-edge band; and (3) VIs including the red-edge band. Table 2 shows the details about each VI.
Table 2. Vegetation indices used in this study. R, G, B, RE, and NIR are the abbreviations of red, green, blue, red-edge, and near infrared bands. The subscript number is the specific central band.
Table 2. Vegetation indices used in this study. R, G, B, RE, and NIR are the abbreviations of red, green, blue, red-edge, and near infrared bands. The subscript number is the specific central band.
CategoryVegetation IndexFormula
VisibleVisible atmospherically resistant index [26] V A R I = ( G 560 − R 668 ) / ( G 560 + R 668 − B 475 )
Vegetative index [27] V E G = G 560 / ( R 668 0.667 ∗ B 475 ( 1 − 0.667 ) )
Normalized pigment chlorophyll ratio index [28] N P C I = ( R 668 − B 444 ) / ( R 668 + B 444 )
Excess green index [29] E x G = 2 G 560 − R 668 − B 475
Excess G minus excess red index [30] E x G R = 3 G 560 − 2.4 R 668 − B 475
Normalized blue-green difference index [31] N G B D I = ( G 560 − B 475 ) / ( G 560 + B 475 )
NIRNormalized difference vegetation index [32] N D V I = ( N I R 842 − R 668 ) / ( N I R 842 + R 668 )
Green normalized difference vegetation index [33] G N D V I = ( N I R 842 − G 560 ) / ( N I R 842 + G 560 )
Relative vigor index [34] R V I = N I R 842 / R 668
Structure insensitive pigment index [35] S I P I = ( N I R 842 − B 475 ) / ( N I R 842 + R 668 )
Enhanced vegetation index [36] E V I = 2.5 ( N I R 842 − R 668 ) / ( 1 + N I R 842 − 2.4 R 668 )
NDVI650 N D V I 650 = ( N I R 842 − R 650 ) / ( N I R 842 + R 650 )
RedEdgeRed-edge chlorophyll vegetation index [37] R E C I = N I R 842 / R E 717 − 1
Modified chlorophyll absorption ratio index [38] M C A R I = ( ( R E 717 − R 668 ) − 0.2 ( R E 717 − G 560 ) ) / ( R E 717 / R 668 )
Normalized difference red edge [39] N D R E = ( N I R 842 − R E 717 ) / ( N I R 842 + R E 717 )
Simplified canopy chlorophyll content index [40] S C C C I = ( N I R 842 − R E 717 ) / ( N I R 842 + R E 717 ) ( N I R 842 − R 668 ) / ( N I R 842 + R 668 )
MCARI740 M C A R I 740 = ( ( R E 740 − R 668 ) − 0.2 ( R E 740 − G 560 ) ) / ( R E 740 / R 668 )
Transformed chlorophyll absorption in reflectance index/optimized soil adjusted vegetation index [41] T O S A V I = 3 ( ( R E 717 − R 668 ) − 0.2 ( R E 717 − G 560 ) ) ∗ ( R E 717 / R 668 ) ( 1 + 0.16 ) ∗ ( N I R 842 − R 668 ) / ( N I R 842 + R 668 + 0.16 )

3.2. Prediction of Winter Wheat LCC Using Machine Learning Regression Algorithms

3.2.1. Machine Learning Algorithms

Among various machine learning regression algorithms, support vector machine (SVM) is a common approach to develop nonlinear kernel regression model [42]. It develops an ideal separating hyperplane to distinguish classes, and a huge modified feature space is produced to map the data and then separated using kernel functions. To establish the SVM regression model, the radial basis function, which has two very important parameters of penalty parameter C and kernel parameter g, was selected as the kernel function. The penalty parameter C represents the tolerance of the model to errors in order to achieve a balance between model accuracy and model complexity [43]. The kernel parameter g controls the regression error of the model and affects the complexity of the distribution of the sample data in the high-dimensional feature space [44]. In this study, the penalty parameter C and kernel parameter g were set to 1 and 0.067, respectively.
Random forest (RF), as one of the Bagging integration algorithms, combines ensemble learning methods with the decision tree framework to create multiple randomly drawn decision trees from the data, averaging the results of the decision tree model to obtain a more powerful prediction [45]. A decision tree is a tree-like structure in which each internal (non-leaf) node is labelled with a test for some attribute. To establish the RF regression model, the number of decision trees and the number of variables per node were two key parameters. In this study, the number of decision trees was set as the default 500 and the number of variables per node was set as the square root of the number of input variables.
The above two machine learning regression algorithms (SVM and RF) and multivariate linear regression (MLR) algorithm were adopted to compare their estimating performances of winter wheat LCC based on VIs derived from UAV multispectral images. SVM, RF, and MLR algorithms were implemented using “e1071” (version 1.7), “randomForest” (version 4.7), and “lm” (version 3.6.2) packages in the R language.

3.2.2. Establishment of Winter Wheat LCC Estimation Models

As shown in Figure 2, two key steps were conducted in this study to map the LCC of winter wheat based on UAV-derived VIs, namely the training and validation of LCC estimation models. During the training process, the training dataset (Table 3), collected from the experimental field in R1, R3, and R5 were applied to establish LCC estimation models using support vector machine (SVM), random forest (RF), and multivariate linear regression (MLR) algorithms. To assess the performance of the three training models in estimating the LCC of winter wheat under different nitrogen levels across the growing season, R2 and RMSE were calculated using the five-fold cross validation method implemented using the rsmp() function of the mlr3 package in R.
During the validation process, the performance of the three models was evaluated based on the validation dataset (Table 3), which was collected from the experimental field in R2 and R4. To comprehensively assess the applicability of the three models to different growth stages, species, and nitrogen levels, the validation data pairs were divided based on the growth stages, species, and nitrogen application levels, respectively. Finally, the LCC maps were obtained based on the optimal regression model.

3.3. Statistical Analysis

Pearson’s correlation coefficient was used to measure the correlation between vegetation indices and winter wheat LCC. The R2 and root mean square error (RMSE) were used to evaluate the estimation accuracy of winter wheat LCC. To reduce errors caused by the partitioning of the training set and test set, five-fold cross validation was adopted for the establishment of each model. The formulas of R2 and RMSE are shown as Equations (1) and (2):
R 2 = ∑ i = 1 n ( y ^ i − y ¯ ) ∑ i = 1 n ( y i − y ¯ )
R M S E = ∑ i = 1 n ( y ^ i − y i ) 2 n
where y ^ i is the predicted LCC, y i is the measured LCC, y ¯ is the mean value for the measured LCC, and n is the number of samples.

4. Results

4.1. Temporal Variations in LCC for Different Nitrogen Levels and Winter Wheat Species

To confirm that various LCC data were collected for the five winter wheat species, temporal variations in LCC were depicted for the five different nitrogen levels (N0, N1, N2, N3, N4, and N5). As shown in Figure 3 and Table 4, different nitrogen stress statuses were successfully applied for five winter wheat species resulting in clear gradients of LCC for all winter wheat species with the order of N4 ≥ N3 > N2 > N1 > N0. Specifically, winter wheat LCC increased significantly with increasing nitrogen levels in the range of N0 (0 kg/ha) to N3 (270 kg/ha) and then stabilized after nitrogen levels higher than N3 for most species. The lowest LCC was observed for treatment N0 (0 kg/ha), with average SPAD values of 27.0, 27.3, 24.1, 26.8, and 26.9 for winter wheat species 1–5, respectively. Compared to treatment N0, the average SPAD values of winter wheat in treatment N3 increased by 15.8, 16.0, 19.8, 20.1, and 16.7 for the five winter species, respectively. Treatment N4 (360 kg/ha) had the highest LCC, with average SPAD values of 45.6, 43.9, 44.6, 47.6, and 43.5, respectively. However, there was no significant difference in the average SPAD of winter wheat between treatment N3 and N4, which were 2.7, 0.7, 0.7, 0.7, and 0.1 for the five species, respectively. This means that the N3 level meets the nitrogen fertilizer requirement of winter wheat.
For all nitrogen application levels, the winter wheat LCC showed a decreasing trend during different growth stages. However, there was a slight difference in the decreasing trend of winter wheat LCC between nitrogen level N0 and nitrogen levels N1–N4. Specifically, the LCC of winter wheat for nitrogen application level N0 and nitrogen application levels N1–N4 decreased sharply from the filling stage (May 01) and milk stage (May 08), respectively. This is because the leaves of the winter wheat at nitrogen level N0 turned yellow earlier due to the nitrogen stress. The reason for the sharp decline in LCC after the milk stage is that most of the nitrogen is transferred from the leaves to the grains as the grains mature. The decrease in nitrogen content in the leaves led to the decrease in winter wheat LCC.

4.2. Correlations between LCC and UAV-Derived VIs

Figure 4 shows the correlation between ground truth LCC and each individual UAV-derived VI. For all three categories of VIs, a significant correlation was found between the LCC and each VI, with the highest and lowest absolute Pearson’s correlation coefficient (r) value of 0.95 and 0.54, respectively. Specifically, as for VIs with only the visible bands, VARI and VEG had the highest absolute r value of 0.95, and the lowest absolute r value was observed for NGBDI, with a value of 0.72. When the NIR band was taken into consideration, the highest absolute r value of 0.95 was found for NDVI, and the lowest absolute r value increased from 0.72 to 0.85 (NDVI650). Regarding the red-edge-related VIs, the highest absolute r value of 0.93 was found for NDRE and the lowest absolute r value of 0.54 was found for MCARI740. Finally, the VIs with absolute r values greater than 0.80, namely VARI, VEG, NPCI, ExG, ExGR, NDVI, GNDVI, RVI, SIPI, EVI, NDVI650, RECI, MCARI, NDRE, and SCCCI, were selected to establish regression models of LCC.

4.3. Performances of Different LCC Regression Models

As shown in Figure 5, when the training dataset was used, all three regression algorithms (MLR, RF, and SVM) successfully estimated the winter wheat LCC based on UAV-derived VIs. Similar estimation performances with average R2 values of 0.944, 0.932, and 0.935 were observed for MLR, RF, and SVM, respectively, and the average RMSE values for MLR, RF, and SVM were 3.96, 4.37, and 4.27, respectively. MLR was the best regression algorithm with a slight advantage.

4.3.1. Effects of Growth Stages on the Performance of LCC Regression Models

To explore the applicability of these three LCC regression models to different growth stages, the validation data were divided into four groups according to the growth stage. As shown in Figure 6, the results showed that the applicability of the SVM regression model to different growth stages was better than that of the RF and MLR regression models. Specifically, the LCC estimation accuracy at the dough stage was the worst among the four growth stages, and the R2 (RMSE) of the MLR, RF, and SVM regression models was 0.33 (4.99), 0.49 (4.35), and 0.60 (3.86), respectively. Among the three regression models, the LCC estimation accuracy of the SVM model was higher than that of RF and SVM, with an R2 difference of 0.1 and 0.27, respectively. The second-worst LCC estimation performance was observed for the filling stage, with R2 (RMSE) values of 0.79 (3.17), 0.75 (3.41), and 0.79 (3.13) for the MLR, RF, and SVM regression models, respectively. Similar LCC estimation performances were found for the three models, with the biggest R2 difference being 0.04 and the biggest RMSE difference being 0.28. As for the flowering and milk stages, similar performances were also observed, with the biggest R2 difference being 0.05 and the biggest RMSE difference being 0.75. In a word, when all four growth stages were taken into consideration, the model established based on the SVM algorithm had the best LCC estimation performance.

4.3.2. Effects of Winter Wheat Species on the Performance of LCC Regression Models

As shown in Figure 7, when the influence of winter wheat species was taken into consideration, all three models resulted in similar R2 values, with the highest value of 0.98 and the lowest value of 0.94. However, the SVM model had a slight advantage over the other two regression models in the applicability to different species, with the lowest maximum difference of RMSE among the five winter species. Specifically, the maximum difference of RMSE among the five species was 1.46, 1.72, and 1.44 for MLR, RF, and SVM, respectively. As for the MLR model, the best LCC estimation performance was found for Zhenmai168 (S5), with an RMSE value of 2.73, and the worst performance was found for Ningmaizi216 (S3), with an RMSE value of 4.19. When it comes to the performance of the RF model, the highest and lowest RMSE values of 4.31 and 2.59 were observed for Yangmai25 (S4) and Zhenmai168 (S5), respectively. As for the SVM model, the highest RMSE was also found for Yangmai25 (S4), with the value of 4.07, and the lowest RMSE was also found for Zhenmai168 (S5), with value of 2.63. In a word, when all five species were taken into consideration, the three models had a similar LCC estimation performance, with the SVM model being slightly better.

4.3.3. Effects of Nitrogen Levels on the Performance of LCC Regression Models

Figure 8 shows the influence of nitrogen application levels on the LCC estimation performances of the MLR, RF, and SVM models. Compared to the species of winter wheat, the nitrogen levels had more influence on the performance of the LCC estimation model, resulting in a slightly higher difference for all three models, with the highest R2 value of 0.97 and the lowest R2 value of 0.89. Compared to the RF and MLR regression models, the SVM regression model had a slight advantage in the applicability to different nitrogen levels, with similar R2 values and a lower RMSE difference among the five nitrogen levels. Specifically, when it comes to each model, the lowest R2 value was observed for the same nitrogen application level of N0, with values of 0.89, 0.91, and 0.91 for MLR, RF, and SVM, respectively. When RMSE was taken into consideration, similar values were also found for all three models among the five nitrogen levels. The highest RMSE difference among the different nitrogen levels was 0.74, 0.92, and 0.61 for MLR, RF, and SVM, respectively. For all three models, treatment N3 had the highest RMSE values of 3.92, 3.96, and 3.65; the lowest RMSE values of 3.18, 3.03, and 3.04 were observed for N1. In a word, when all five nitrogen application levels were taken into consideration, the three models had a similar LCC estimation performance, with the SVM model being slightly better.

4.4. Mapping Leaf Chlorophyll Content from UAV Multispectral Images Using SVM Regression Model

When the effects of growth stage, species, and applied nitrogen levels were all taken into consideration, the model established based on the SVM regression algorithm had the best LCC estimation performance. Therefore, the SVM model was chosen to map the LCC of winter wheat. As shown in Figure 9, clear LCC differences were observed among the different nitrogen application levels and winter wheat species. In addition, due to the field heterogeneity, LCC differences could also be observed within each small plot. In a practical planting scenario, the LCC maps obtained by the combination of the above SVM model and UAV multispectral images could be used to indicate the growing status of winter wheat, and to help farmers to carry out precise field management. For example, after the winter dormancy stage of winter wheat, the LCC map could be used to help famers to carry out precision nitrogen fertilizer management.

5. Discussion

LCC is an important indicator of crop photosynthetic capacity and nutritional status, and thus could be treated as a reference basis for field nitrogen fertilization management [24,46]. In this study, the winter wheat LCC first increased with the increasing nitrogen application, and tended to be stable after meeting the nitrogen requirement of winter wheat. In addition, the LCC values collected at each nitrogen treatment showed a sharp decreasing trend after the milk stage, which was caused by the nitrogen transfer from leaves to grains. Notably, due to the nitrogen stress of winter wheat at nitrogen level N0, the LCC of all winter wheat species in N0 decreased earlier than that at nitrogen levels N1–N4. Variation in LCC throughout the growing season can provide farmers with nitrogen content information. Therefore, it is of great significance to monitor the crop LCC.
Several studies have been conducted to monitor the LCC of potato [47], wheat [48], sugarcane [18], maize [16], etc. based on UAV remote sensing technology. However, whether there is a versatile regression model that could be used to accurately estimate crop LCC for different species, growth stages, and nitrogen stress conditions is uncertain. Only when it could be successfully applied to different species, growth stages, and nitrogen stress conditions, the LCC regression model could be easily adopted by farmers to guide actual crop production management. For this purpose, research was conducted in 2023 across the growing season for winter wheat, with five species and five nitrogen application levels.
Firstly, correlations between ground truth LCC and UAV multispectral VIs were explored. As shown in Figure 3, data pairs covering five winter wheat species, five nitrogen application levels, and four key winter wheat growth stages were successfully collected, and provided the fundamental basis for the whole research. Similar to previous studies [14,49,50,51], in this study, strong correlations were also found between LCC and most of the UAV-derived Vis, with the highest r value being 0.95. However, among the three types of VIs, superior LCC estimation ability was not observed for the red-edge-related VIs. The highest r value of 0.93 was observed for NDRE, which was slightly lower than that of VARI, VEG, and NDVI, all with the same r value of 0.95. In previous studies [52,53,54], the red-edge band and VIs with the red-edge band had been demonstrated that they could improve the prediction of the LCCs of crops because of their higher sensitivity than NDVI to high plant biomass. The possible reason for the different result in this study may be that the winter wheat biomass did not exceed the critical line, which leads to the low sensitivity of NDVI. Further study is still needed.
Secondly, after screening suitable Vis, which were selected based on the r value being greater than 0.80, MLR, RF, and SVM regression algorithms were adopted to establish LCC estimation models using the training dataset. A slightly higher estimation performance was found for MLR than for RF and SVM, with an average R2 difference less than 0.012 and average RMSE differences less than 0.41. However, when the established models were used to estimate LCC using the validation dataset, the worst generalization ability was observed for MLR. Similar phenomenon was found in previous studies when MLR was taken as the reference for testing the performance of machine learning algorithms [18]. For example, compared with the RF algorithm, lower generalization ability was found for MLR when it were used to estimate maize fractional vegetation cover under various water stress and growth stages conditions based on UAV-derived multispectral VIs [21]. The RF algorithm showed the best accuracy of 90% compared to MLR when estimating the chlorophyll content of citrus trees [55]. The above results demonstrated that a non-linear relationship exists between winter wheat LCC and UAV-derived multispectral VIs, and MLR, which relies on the linear relationship between multiple independent variables and dependent variables for modeling [56], could not solve complex nonlinear problems.
In addition, compared to different winter wheat species and nitrogen application levels, the growth stage had the highest influence on the LCC estimation performance. The LCC could not be accurately estimated by the established MLR model in the dough growth stage, with an R2 of 0.33 and RMSE of 4.99. The SVM model had the best generalization ability, with an R2 of 0.60 and RMSE of 3.86 in the dough growth stage. Although different crop species have differences in trait expression and different levels of abiotic stress have different impacts on crop growth status, greater changes in crop traits will be observed when the growth stage is taken into consideration, especially when the crop reaches maturity. As shown in Figure 3, sharp decreases in SPAD were observed for all winter wheat species and nitrogen application levels in the dough stage. Overall, this research shows the successful application of UAV data for mapping LCC in winter wheat across diverse conditions, offering valuable insights for precision nitrogen fertilization management.
Considering that the leaf water status would affect the vegetation indices, models built based on data collected from an experimental site with different climatic conditions would be different from those built in this study. However, since all the data in this study were from the same experimental field, the degree of influence of different climatic conditions on the performance of the winter wheat LCC estimation model was not investigated here. Further study is still needed based on data collected from different research sites with other climatic and geographic conditions across more growing seasons.

6. Conclusions

The results show that among the three types of UAV-derived multispectral VIs, VARI, VEG, NDVI, and NDRE had the highest correlation with the LCC of winter wheat, with r values of 0.93–0.95. All three regression algorithms, MLR, RF, and SVM, could be successfully used to estimate the LCC of winter wheat, with the lowest average R2 being 0.932 and the highest average RMSE being 4.37, which were obtained based on the training dataset through five-fold cross validation. However, when these models were used to estimate LCC based on the stand-alone validation dataset, the worst generalization ability was observed for MLR, which could not accurately estimate LCC in the dough stage. The reason may be that the growth stage had a greater impact on the changes in crop traits compared to the crop species and different levels of abiotic stress, especially when the crop reached maturity. The best generalization ability was found for SVM, with the lowest R2 of 0.60 and highest RMSE of 3.86 obtained in the dough stage throughout the whole validation process. Compared to winter wheat species and nitrogen application levels, the growth stage had the greatest influence on the generalization ability of the LCC estimation models. Overall, the combination of UAV-derived multispectral VIs and the SVM regression algorithm could be successfully used to map the leaf chlorophyll content of winter wheat for different species, growth stages, and nitrogen stress conditions.

Author Contributions

Conceptualization, visualization, and writing—original draft preparation, L.Z.; data curation, methodology, and software, A.W. and Q.Z.; investigation, validation, and formal analysis, Y.N. and H.Z. (Huiyue Zhang); resources and writing—review and editing, H.Z. (Huihui Zhang) and W.S.; funding acquisition, L.Z. All authors have read and agreed to the published version of the manuscript.

Funding

This study was supported by National Science and Technology Major Project (2022ZD0115801), National Natural Science Foundation of China (No. 52309051, 32001417), China Postdoctoral Science Foundation (No. 2022T150276), and Jiangsu Agricultural Science and Technology Innovation Fund (CX(21)3061).

Institutional Review Board Statement

Not applicable.

Data Availability Statement

The original contributions presented in the study are included in the article. The data presented in this study are available on request from the corresponding author.

Acknowledgments

The authors wish to thank all those who helped in this research.

Conflicts of Interest

Declare The authors declare no conflicts of interest.

References

  1. Lv, Z.; Liu, X.; Cao, W.; Zhu, Y. Climate change impacts on regional winter wheat production in main wheat production regions of China. Agric. For. Meteorol. 2013, 171, 234–248. [Google Scholar] [CrossRef]
  2. Gitelson, A.A.; Buschmann, C.; Lichtenthaler, H.K. The chlorophyll fluorescence ratio F735/F700 as an accurate measure of the chlorophyll content in plants. Remote Sens. Environ. 1999, 69, 296–302. [Google Scholar] [CrossRef]
  3. Verma, B.; Prasad, R.; Srivastava, P.K.; Singh, P.; Badola, A.; Sharma, J. Evaluation of Simulated AVIRIS-NG Imagery Using a Spectral Reconstruction Method for the Retrieval of Leaf Chlorophyll Content. Remote Sens. 2022, 14, 3560. [Google Scholar] [CrossRef]
  4. Martínez, D.E.; Guiamet, J.J. Distortion of the SPAD 502 chlorophyll meter readings by changes in irradiance and leaf water status. Agronomie 2004, 24, 41–46. [Google Scholar] [CrossRef]
  5. Rigon, J.P.G.; Capuani, S.; Fernandes, D.M.; Guimarães, T.M. A novel method for the estimation of soybean chlorophyll content using a smartphone and image analysis. Photosynthetica 2016, 54, 559–566. [Google Scholar] [CrossRef]
  6. Fernandes, F.M.; Soratto, R.P.; Fernandes, A.M.; Souza, E.F.C. Chlorophyll meter-based leaf nitrogen status to manage nitrogen in tropical potato production. Agron. J. 2021, 113, 1733–1746. [Google Scholar] [CrossRef]
  7. Zhang, Q.; Chen, Q.; Xu, L.; Xu, X.; Liang, Z. Wheat Lodging Direction Detection for Combine Harvesters Based on Improved K-Means and Bag of Visual Words. Agronomy 2023, 13, 2227. [Google Scholar] [CrossRef]
  8. Sun, Y.; Luo, Y.; Zhang, Q.; Xu, L.; Wang, L.; Zhang, P. Estimation of crop height distribution for mature rice based on a moving surface and 3D point cloud elevation. Agronomy 2022, 12, 836. [Google Scholar] [CrossRef]
  9. Li, W.; Zhang, C.; Ma, T.; Li, W. Estimation of summer maize biomass based on a crop growth model. Emir. J. Food Agric. 2021, 33, 742. [Google Scholar] [CrossRef]
  10. Zhang, L.; Niu, Y.; Zhang, H.; Han, W.; Li, G.; Tang, J.; Peng, X. Maize canopy temperature extracted from UAV thermal and RGB imagery and its application in water stress monitoring. Front. Plant Sci. 2019, 10, 1270. [Google Scholar] [CrossRef] [PubMed]
  11. Zhu, W.; Feng, Z.; Dai, S.; Zhang, P.; Wei, X. Using UAV multispectral remote sensing with appropriate spatial resolution and machine learning to monitor wheat scab. Agriculture 2022, 12, 1785. [Google Scholar] [CrossRef]
  12. Cong, C.; Guangqiao, C.; Yibai, L.; Dong, L.; Bin, M.; Jinlong, Z.; Liang, L.; Jianping, H. Research on Monitoring Methods for the Appropriate Rice Harvest Period Based on Multispectral Remote Sensing. Discret. Dyn. Nat. Soc. 2022, 2022, 1519667. [Google Scholar] [CrossRef]
  13. Zhou, L.; Nie, C.; Su, T.; Xu, X.; Song, Y.; Yin, D.; Liu, S.; Liu, Y.; Bai, Y.; Jia, X.; et al. Evaluating the Canopy Chlorophyll Density of Maize at the Whole Growth Stage Based on Multi-Scale UAV Image Feature Fusion and Machine Learning Methods. Agriculture 2023, 13, 895. [Google Scholar] [CrossRef]
  14. Sun, Q.; Gu, X.; Chen, L.; Xu, X.; Wei, Z.; Pan, Y.; Gao, Y. Monitoring maize canopy chlorophyll density under lodging stress based on UAV hyperspectral imagery. Comput. Electron. Agric. 2022, 193, 106671. [Google Scholar] [CrossRef]
  15. Zhu, X.; Yang, Q.; Chen, X.; Ding, Z. An Approach for Joint Estimation of Grassland Leaf Area Index and Leaf Chlorophyll Content from UAV Hyperspectral Data. Remote Sens. 2023, 15, 2525. [Google Scholar] [CrossRef]
  16. Mao, Z.-H.; Deng, L.; Sun, J.; Zhang, A.-W.; Chen, X.-Y.; Zhao, Y. Research on the Application of UAV Multispectral Remote Sensing in the Maize Chlorophyll Prediction. Spectrosc. Spectr. Anal. 2018, 38, 2923–2931. [Google Scholar]
  17. Fuentes-Peñailillo, F.; Acevedo-Opazo, C.; Ortega-Farías, S.; Rivera, M.; Verdugo-Vásquez, N. Spatialized system to monitor vine flowering: Towards a methodology based on a low-cost wireless sensor network. Comput. Electron. Agric. 2021, 187, 106233. [Google Scholar] [CrossRef]
  18. Narmilan, A.; Gonzalez, F.; Salgadoe, A.S.A.; Kumarasiri, U.W.L.M.; Weerasinghe, H.A.S.; Kulasekara, B.R. Predicting Canopy Chlorophyll Content in Sugarcane Crops Using Machine Learning Algorithms and Spectral Vegetation Indices Derived from UAV Multispectral Imagery. Remote Sens. 2022, 14, 1140. [Google Scholar] [CrossRef]
  19. Guo, Y.; Xiao, Y.; Hao, F.; Zhang, X.; Chen, J.; de Beurs, K.; He, Y.; Fu, Y.H. Comparison of different machine learning algorithms for predicting maize grain yield using UAV-based hyperspectral images. Int. J. Appl. Earth Observ. Geoinf. 2023, 124, 103528. [Google Scholar] [CrossRef]
  20. Fu, Z.; Jiang, J.; Gao, Y.; Krienke, B.; Wang, M.; Zhong, K.; Cao, Q.; Tian, Y.; Zhu, Y.; Cao, W.; et al. Wheat Growth Monitoring and Yield Estimation based on Multi-Rotor Unmanned Aerial Vehicle. Remote Sens. 2020, 12, 508. [Google Scholar] [CrossRef]
  21. Niu, Y.; Han, W.; Zhang, H.; Zhang, L.; Chen, H. Estimating fractional vegetation cover of maize under water stress from UAV multispectral imagery using machine learning algorithms. Comput. Electron. Agric. 2021, 189, 106414. [Google Scholar] [CrossRef]
  22. Yin, H.; Huang, W.; Li, F.; Yang, H.; Li, Y.; Hu, Y.; Yu, K. Multi-temporal UAV Imaging-Based Mapping of Chlorophyll Content in Potato Crop. PFG-J. Photogramm. Remote Sens. Geoinf. Sci. 2023, 91, 91–106. [Google Scholar] [CrossRef]
  23. Wei, Q.; Zhang, B.; Wei, Z.; Han, X.; Duan, C. Estimation of Canopy Chlorophyll Content in Winter Wheat by UAV Multispectral Remote Sensing. J. Triticeae Crop. 2020, 40, 365–372. [Google Scholar]
  24. Wu, Q.; Zhang, Y.; Zhao, Z.; Xie, M.; Hou, D. Estimation of Relative Chlorophyll Content in Spring Wheat Based on Multi-Temporal UAV Remote Sensing. Agronomy 2023, 13, 211. [Google Scholar] [CrossRef]
  25. Qiao, L.; Tang, W.; Gao, D.; Zhao, R.; An, L.; Li, M.; Sun, H.; Song, D. UAV-based chlorophyll content estimation by evaluating vegetation index responses under different crop coverages. Comput. Electron. Agric. 2022, 196, 106775. [Google Scholar] [CrossRef]
  26. Gitelson, A.A.; Kaufman, Y.J.; Stark, R.; Rundquist, D. Novel algorithms for remote estimation of vegetation fraction. Remote Sens. Environ. 2002, 80, 76–87. [Google Scholar] [CrossRef]
  27. Hague, T.; Tillett, N.D.; Wheeler, H. Automated crop and weed monitoring in widely spaced cereals. Precis. Agric. 2006, 7, 21–32. [Google Scholar] [CrossRef]
  28. Merzlyak, M.N.; Gitelson, A.A.; Chivkunova, O.B.; Rakitin, V.Y.U. Non-destructive optical detection of pigment changes during leaf senescence and fruit ripening. Physiol. Plant. 1999, 106, 135–141. [Google Scholar] [CrossRef]
  29. Woebbecke, D.M.; Meyer, G.E.; Von Bargen, K.; Mortensen, D.A. Color indices for weed identification under various soil, residue, and lighting conditions. Trans. ASAE 1995, 38, 259–269. [Google Scholar] [CrossRef]
  30. Meyer, G.E.; Neto, J.C. Verification of color vegetation indices for automated crop imaging applications. Comput. Electron. Agric. 2008, 63, 282–293. [Google Scholar] [CrossRef]
  31. Hunt, E.R.; Cavigelli, M.; Daughtry, C.S.T.; McMurtrey, J.E.; Walthall, C.L. Evaluation of digital photography from model aircraft for remote sensing of crop biomass and nitrogen status. Precis. Agric. 2005, 6, 359–378. [Google Scholar] [CrossRef]
  32. Rouse, J.W., Jr.; Haas, R.H.; Deering, D.; Schell, J.; Harlan, J.C. Monitoring the Vernal Advancement and Retrogradation (Green Wave Effect) of Natural Vegetation; NASA: Washington, DC, USA, 1974.
  33. Gitelson, A.A.; Merzlyak, M.N. Remote sensing of chlorophyll concentration in higher plant leaves. Adv. Space Res. 1998, 22, 689–692. [Google Scholar] [CrossRef]
  34. Jordan, C.F. Derivation of leaf-area index from quality of light on the forest floor. Ecology 1969, 50, 663–666. [Google Scholar] [CrossRef]
  35. Penuelas, J.; Baret, F.; Filella, I. Semi-empirical indices to assess carotenoids/chlorophyll a ratio from leaf spectral reflectance. Photosynthetica 1995, 31, 221–230. [Google Scholar]
  36. Huete, A.; Didan, K.; Miura, T.; Rodriguez, E.P.; Gao, X.; Ferreira, L.G. Overview of the radiometric and biophysical performance of the MODIS vegetation indices. Remote Sens. Environ. 2002, 83, 195–213. [Google Scholar] [CrossRef]
  37. Gitelson, A.A.; Viña, A.; Ciganda, V.; Rundquist, D.C.; Arkebauer, T.J. Remote estimation of canopy chlorophyll content in crops. Geophys. Res. Lett. 2005, 32, L08403. [Google Scholar] [CrossRef]
  38. Daughtry, C.S.T.; Walthall, C.L.; Kim, M.S.; De Colstoun, E.B.; McMurtrey, J.E., III. Estimating corn leaf chlorophyll concentration from leaf and canopy reflectance. Remote Sens. Environ. 2000, 74, 229–239. [Google Scholar] [CrossRef]
  39. Li, F.; Miao, Y.; Feng, G.; Yuan, F.; Yue, S.; Gao, X.; Liu, Y.; Liu, B.; Ustin, S.L.; Chen, X. Improving estimation of summer maize nitrogen status with red edge-based spectral vegetation indices. Field Crop. Res. 2014, 157, 111–123. [Google Scholar] [CrossRef]
  40. Barnes, E.M.; Clarke, T.R.; Richards, S.E.; Colaizzi, P.D.; Haberland, J.; Kostrzewski, M.; Waller, P.; Choi, C.; Riley, E.; Thompson, T. Coincident detection of crop water stress, nitrogen status and canopy density using ground based multispectral data. In Proceedings of the Fifth International Conference on Precision Agriculture, Bloomington, MN, USA, 16–19 July 2000; p. 6. [Google Scholar]
  41. Wu, C.; Niu, Z.; Tang, Q.; Huang, W. Estimating chlorophyll content from hyperspectral vegetation indices: Modeling and validation. Agric. For. Meteorol. 2008, 148, 1230–1241. [Google Scholar] [CrossRef]
  42. Ta, N.; Chang, Q.; Zhang, Y. Estimation of apple tree leaf chlorophyll content based on machine learning methods. Remote Sens. 2021, 13, 3902. [Google Scholar] [CrossRef]
  43. Han, H.; Wan, R.; Li, B. Estimating forest aboveground biomass using Gaofen-1 images, Sentinel-1 images, and machine learning algorithms: A case study of the Dabie Mountain Region, China. Remote Sens. 2021, 14, 176. [Google Scholar] [CrossRef]
  44. Zhang, L.; Shao, Z.; Liu, J.; Cheng, Q. Deep learning based retrieval of forest aboveground biomass from combined LiDAR and landsat 8 data. Remote Sens. 2019, 11, 1459. [Google Scholar] [CrossRef]
  45. Wang, L.A.; Zhou, X.; Zhu, X.; Dong, Z.; Guo, W. Estimation of biomass in wheat using random forest regression algorithm and remote sensing data. Crop J. 2016, 4, 212–219. [Google Scholar] [CrossRef]
  46. Fu, Z.; Yu, S.; Zhang, J.; Xi, H.; Gao, Y.; Lu, R.; Zheng, H.; Zhu, Y.; Cao, W.; Liu, X. Combining UAV multispectral imagery and ecological factors to estimate leaf nitrogen and grain protein content of wheat. Eur. J. Agron. 2022, 132, 126405. [Google Scholar] [CrossRef]
  47. Bian, M.; Ma, Y.; Fan, Y.; Chen, Z.; Yang, G.; Feng, H. Estimation of Potato Chlorophyll Content Based on UAV Multi-source Sensor. Trans. Chin. Soc. Agric. Mach. 2023, 54, 240–248. [Google Scholar]
  48. Wang, W.; Gao, X.; Cheng, Y.; Ren, Y.; Zhang, Z.; Wang, R.; Cao, J.; Geng, H. QTL Mapping of Leaf Area Index and Chlorophyll Content Based on UAV Remote Sensing in Wheat. Agriculture 2022, 12, 595. [Google Scholar] [CrossRef]
  49. Liu, H.; Lei, X.; Liang, H.; Wang, X. Multi-Model Rice Canopy Chlorophyll Content Inversion Based on UAV Hyperspectral Images. Sustainability 2023, 15, 7038. [Google Scholar] [CrossRef]
  50. Yin, H.; Li, F.; Yang, H.; Li, Y. Estimation of canopy chlorophyll in potato based on UAV hyperspectral images. J. Plant Nutr. Fertil. 2021, 27, 2184–2195. [Google Scholar]
  51. Kanning, M.; Kuehling, I.; Trautz, D.; Jarmer, T. High-Resolution UAV-Based Hyperspectral Imagery for LAI and Chlorophyll Estimations from Wheat for Yield Prediction. Remote Sens. 2018, 10, 2000. [Google Scholar] [CrossRef]
  52. Evri, M.; Akiyama, T.; Kawamura, K. Spectrum analysis of hyperspectral red edge position to predict rice biophysical parameters and grain weight. J. Jpn. Soc. Photogramm. Remote Sens. 2008, 47, 4–15. [Google Scholar] [CrossRef]
  53. Kanke, Y.; Raun, W.; Solie, J.; Stone, M.; Taylor, R. Red edge as a potential index for detecting differences in plant nitrogen status in winter wheat. J. Plant Nutr. 2012, 35, 1526–1541. [Google Scholar] [CrossRef]
  54. Deng, L.; Mao, Z.; Li, X.; Hu, Z.; Duan, F.; Yan, Y. UAV-based multispectral remote sensing for precision agriculture: A comparison between different cameras. ISPRS J. Photogramm. Remote Sens. 2018, 146, 124–136. [Google Scholar] [CrossRef]
  55. Prado Osco, L.; Marques Ramos, A.P.; Roberto Pereira, D.; Akemi Saito Moriya, É.; Nobuhiro Imai, N.; Takashi Matsubara, E.; Estrabis, N.; de Souza, M.; Marcato Junior, J.; Gonçalves, W.N. Predicting canopy nitrogen content in citrus-trees using random forest algorithm associated to spectral vegetation indices from UAV-imagery. Remote Sens. 2019, 11, 2925. [Google Scholar] [CrossRef]
  56. Patonai, Z.; Kicsiny, R.; Géczi, G. Multiple linear regression based model for the indoor temperature of mobile containers. Heliyon 2022, 8, e12098. [Google Scholar] [CrossRef]
Agriculture 14 01064 g001
Figure 1. Geographical location of the experimental site and overview of winter wheat experimental design. R1–R5 present repetition areas of 1 to 5, N0–N4 represent nitrogen levels of 0 to 4, and S1–S5 represent winter wheat species of 1 to 5. GCP is the abbreviation of ground control point.
Figure 1. Geographical location of the experimental site and overview of winter wheat experimental design. R1–R5 present repetition areas of 1 to 5, N0–N4 represent nitrogen levels of 0 to 4, and S1–S5 represent winter wheat species of 1 to 5. GCP is the abbreviation of ground control point.
Agriculture 14 01064 g001
Agriculture 14 01064 g002
Figure 2. The flowsheet for mapping LCC of winter wheat based on UAV-derived vegetation indices (VIs). (a) The establishment of LCC estimation models; (b) the evaluation of three LCC estimation models; (c) LCC maps derived based on the optimal model and UAV-derived VIs. MLR, RF, and SVM represent multivariate linear, random forest, and support vector machine regression algorithm, respectively.
Figure 2. The flowsheet for mapping LCC of winter wheat based on UAV-derived vegetation indices (VIs). (a) The establishment of LCC estimation models; (b) the evaluation of three LCC estimation models; (c) LCC maps derived based on the optimal model and UAV-derived VIs. MLR, RF, and SVM represent multivariate linear, random forest, and support vector machine regression algorithm, respectively.
Agriculture 14 01064 g002
Agriculture 14 01064 g003
Figure 3. Temporal variations in LCC of winter wheat for different nitrogen application levels and winter wheat species. “N_levels” represents the five nitrogen application levels at 0, 90, 180, 270, and 360 kg/ha.
Figure 3. Temporal variations in LCC of winter wheat for different nitrogen application levels and winter wheat species. “N_levels” represents the five nitrogen application levels at 0, 90, 180, 270, and 360 kg/ha.
Agriculture 14 01064 g003
Agriculture 14 01064 g004
Figure 4. Pearson’s correlation coefficient (r) between LCC and each individual UAV-derived VI. (a) VIs with only the visible bands; (b) VIs with the NIR band but excluding the red-edge band; (c) VIs including the red-edge band.
Figure 4. Pearson’s correlation coefficient (r) between LCC and each individual UAV-derived VI. (a) VIs with only the visible bands; (b) VIs with the NIR band but excluding the red-edge band; (c) VIs including the red-edge band.
Agriculture 14 01064 g004
Agriculture 14 01064 g005
Figure 5. The distributions of (a) coefficient of determination (R2) and (b) root mean square error (RMSE) derived using five-fold cross validation method for three machine learning algorithms.
Figure 5. The distributions of (a) coefficient of determination (R2) and (b) root mean square error (RMSE) derived using five-fold cross validation method for three machine learning algorithms.
Agriculture 14 01064 g005
Agriculture 14 01064 g006
Figure 6. R2 and RMSE values obtained for three regression models based on validation datasets, which were divided based on the growth stages of winter wheat. The dashed line is the corresponding estimation accuracy observed in the training process.
Figure 6. R2 and RMSE values obtained for three regression models based on validation datasets, which were divided based on the growth stages of winter wheat. The dashed line is the corresponding estimation accuracy observed in the training process.
Agriculture 14 01064 g006
Agriculture 14 01064 g007
Figure 7. R2 and RMSE values obtained for three regression models based on validation datasets, which were divided based on the species of winter wheat. The dashed line is the corresponding estimation accuracy observed in the training process.
Figure 7. R2 and RMSE values obtained for three regression models based on validation datasets, which were divided based on the species of winter wheat. The dashed line is the corresponding estimation accuracy observed in the training process.
Agriculture 14 01064 g007
Agriculture 14 01064 g008
Figure 8. R2 and RMSE values obtained for three regression models based on validation datasets, which were divided based on the nitrogen application levels of winter wheat. The dashed line is the corresponding estimation accuracy observed in the training process.
Figure 8. R2 and RMSE values obtained for three regression models based on validation datasets, which were divided based on the nitrogen application levels of winter wheat. The dashed line is the corresponding estimation accuracy observed in the training process.
Agriculture 14 01064 g008
Agriculture 14 01064 g009
Figure 9. LCC (SPAD value) maps of winter wheat obtained based on UAV-derived VIs and SVM regression algorithm.
Figure 9. LCC (SPAD value) maps of winter wheat obtained based on UAV-derived VIs and SVM regression algorithm.
Agriculture 14 01064 g009
Table 1. Measurements of leaf chlorophyll content (LCC) of winter wheat and UAV multispectral images with respect to growing stages. √ indicates that the corresponding data were collected.
Table 1. Measurements of leaf chlorophyll content (LCC) of winter wheat and UAV multispectral images with respect to growing stages. √ indicates that the corresponding data were collected.
DataGrowing Stages in 2023
Flowering
(20 April)		Filling
(30 April)		Milk
(8 May)		Hard Dough
(23 May)
LCC
(SPAD)		√√√√
UAV multispectral images
(10 bands)		√√√√
Table 3. Detailed information about the training and validation datasets.
Table 3. Detailed information about the training and validation datasets.
DatasetSourceNumber of Pairs
Training datasetR1, R3, and R5300
Validation datasetR2 and R4200
Table 4. The statistics of SPAD values collected from five experimental field repetitions.
Table 4. The statistics of SPAD values collected from five experimental field repetitions.
Nitrogen LevelsSpeciesNumber of PairsMeanSD
N0S12027.012.3
S22027.310.6
S32024.111.2
S42026.813.7
S52026.912.7
N1S12035.916.6
S22036.417.1
S32035.417.4
S42037.520.0
S52036.118.0
N2S12041.218.4
S22039.618.7
S32039.619.2
S42042.419.0
S52039.819.2
N3S12042.920.1
S22043.317.2
S32043.917.6
S42046.916.4
S52043.618.6
N4S12045.615.9
S22043.917.1
S32044.616.5
S42047.617.5
S52043.519.4
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Zhang, L.; Wang, A.; Zhang, H.; Zhu, Q.; Zhang, H.; Sun, W.; Niu, Y. Estimating Leaf Chlorophyll Content of Winter Wheat from UAV Multispectral Images Using Machine Learning Algorithms under Different Species, Growth Stages, and Nitrogen Stress Conditions. Agriculture 2024, 14, 1064. https://doi.org/10.3390/agriculture14071064

AMA Style

Zhang L, Wang A, Zhang H, Zhu Q, Zhang H, Sun W, Niu Y. Estimating Leaf Chlorophyll Content of Winter Wheat from UAV Multispectral Images Using Machine Learning Algorithms under Different Species, Growth Stages, and Nitrogen Stress Conditions. Agriculture. 2024; 14(7):1064. https://doi.org/10.3390/agriculture14071064

Chicago/Turabian Style

Zhang, Liyuan, Aichen Wang, Huiyue Zhang, Qingzhen Zhu, Huihui Zhang, Weihong Sun, and Yaxiao Niu. 2024. "Estimating Leaf Chlorophyll Content of Winter Wheat from UAV Multispectral Images Using Machine Learning Algorithms under Different Species, Growth Stages, and Nitrogen Stress Conditions" Agriculture 14, no. 7: 1064. https://doi.org/10.3390/agriculture14071064

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop