Sentinel-2A Image Reflectance Simulation Method for Estimating the Chlorophyll Content of Larch Needles with Pest Damage

Abstract

With the development of remote sensing technology, the estimation of the chlorophyll content (CHLC) of vegetation via satellite data has become an important means of monitoring vegetation health, and high-precision estimation has been the focus of research in this field. In this study, we used larch affected by Yarl’s larch looper (Erannis jacobsoni Djak) in the boundary region of Mongolia as the research object, simulated the multispectral reflectance, downscaled Sentinel-2A satellite data, performed mixed-pixel decomposition, analyzed the potential of Sentinel-2A satellite data for estimating the chlorophyll content by calculating the spectral indices (SIs) and spectral derivatives (SDFs) of images, and then extracted sensitive spectral features as the model training set. Spectral features sensitive to the chlorophyll content were extracted to establish the training set, and, finally, the chlorophyll content estimation model for larch was constructed on the basis of the partial least squares algorithm (PLSR). The results revealed that SI and SDF based on simulated remote sensing data were highly sensitive to the chlorophyll content under the influence of pests, with the SAVI and EVI2 spectral indices as well as the D_B2 and D_B5 spectral derivatives being the most sensitive to the chlorophyll content. The estimation models based on simulated data performed significantly better than models without simulated data in terms of accuracy, especially those based on SDF-PLSR. The simulated spectral reflectance well reflected the spectral characteristics of the larch canopy and was sensitive to damaged larch, especially in the green light, red edge, and near-infrared bands. The proposed approach improves the accuracy of chlorophyll content estimation via Sentinel-2A data and enhances the ability to monitor changes in the chlorophyll content under complex forest conditions through simulations, providing new technical means and a theoretical basis for forestry pest monitoring and vegetation health management.

1. Introduction

As some of the most important ecosystems on Earth, forest ecosystems play an important role in the global carbon cycle. The Mongolian Plateau is rich in forest resources; however, in recent years, forest ecosystems have experienced unprecedented threats, with forest pests and diseases becoming increasingly serious under the influence of global climate change, causing serious harm to forests. Erannis jacobsoni Djak is a typical pest that harms forests on the Mongolian Plateau, feeds on larch needles and leaves, and causes forests to die in patches in the event of an outbreak, seriously jeopardizing the forest ecosystems of Mongolia. Additionally, according to previous studies, outbreaks spread from the northwest to the southeast, and, if not prevented and controlled in an effective and scientific manner, pose a considerable threat to the security of forest ecosystems globally.

As chlorophyll is a key pigment in photosynthesis, accurate determination of the chlorophyll content (CHLC) is crucial for assessing the health status of plants. Traditional methods of chlorophyll content determination require onsite sampling and laboratory analysis, which are not only time-consuming and labor-intensive but also difficult to use for real-time monitoring at a large scale [1,2]. However, with the development of remote sensing technology, satellite remote sensing has become an emerging method for monitoring the health of vegetation over large areas, providing an effective tool for the early identification and assessment of vegetation health when pest damage occurs. Scholars have explored methods of estimating the vegetation chlorophyll content via multispectral or other satellite remote sensing data. Sentinel-2 data have are the first choice of many researchers because of their advantage of free access [3,4,5,6,7,8,9,10]. Some studies have utilized Sentinel-2 data combined with vegetation indices and machine learning algorithms to accurately estimate the chlorophyll content of crops such as maize and coffee. J. Clevers et al. [3] used Sentinel-2 data for the first time to estimate the LAI, chlorophyll content, and canopy chlorophyll content of potato crops through vegetation indices. Chen, ZL [10] et al. used Sentinel-2 band images combined with the random forest regression algorithm to construct a chlorophyll content estimation model, and the results revealed that the result was best when the chlorophyll content functional equation was used, with a root mean square error of 0.736.

Other remote sensing data, such as Hyperion hyperspectral data [11], MODIS data [12], Landsat data [13], GF-1 and GF-6 data [14], and other satellite data [15], have also been used for the estimation of the chlorophyll content. Chaoyang Wu et al. estimated the chlorophyll content of maize by using vegetation indices derived from Hyperion and TM images. A sensitivity analysis of the indices was conducted via the PROSPECT model [11]. Su, W et al. used the filtered reflectance of Sentinel-2 and MODIS data for the joint inversion of the maize canopy LAI and the chlorophyll content, and the results revealed that the joint inversion of the continuous LAI and the chlorophyll content can be used to monitor maize growth conditions [12].

In terms of methodology, researchers often combine various vegetation indices [4], machine learning algorithms [5,16,17,18], and physical and statistical models [5,11,15,19]. Qian, Binxiang et al. [4] constructed a triangular vegetation index (STVI) based on Sentinel-2 data and used it to estimate the chlorophyll content of vegetation leaves; the results revealed that the STVI was more sensitive to the chlorophyll content than a traditional vegetation index was. Chaoyang Wu et al. estimated the chlorophyll content of maize via vegetation indices derived from Hyperion and TM images and conducted a sensitivity analysis of the indices with the PROSPECT model [11]; specifically, the leaf chlorophyll content was determined via Sentinel-2 data in combination with an artificial neural network, which was subsequently used to estimate nitrogen uptake in an intensive winter wheat cropping system [16]. Each of these methods has unique strengths in estimating the chlorophyll content, with vegetation indices providing a fast and intuitive basis for analyses, machine learning algorithms used to handle complex nonlinear relationships, and physical and statistical models applied to provide insights into the spectral properties of vegetation.

The spatial resolution of remote sensing data is crucial for the precise representation of vegetation conditions. Low-resolution data often fall short in accurately monitoring changes in vegetation biochemical components. For instance, Sentinel-2 data with a 10 m spatial resolution has superior capabilities in detecting the early stages and severity of wood-boring pest infestations compared to MODIS data with a 250 m spatial resolution [20]. Therefore, to meet the demands for more detailed surface observations and to enhance the information content of satellite imagery, increasing the spatial resolution of satellite images has become a common approach among scholars to improve the accuracy of chlorophyll content estimations. Zhang, MZ et al. increased the resolution of the Sentinel-2 red edge band to 10 m and inverted the chlorophyll content of the summer maize canopy by using linear and physical models [21]. Sun, YH developed a series of red edge VIs insensitive to leaf chlorophyll using Sentinel-2 and GF-6 multispectral images to improve the estimation accuracy of the leaf chlorophyll content [22]; Juwon Kong et al. significantly reduced the relative bias of the NDVI by employing a dual generative adversarial network (GAN) to enhance the resolution of Landsat-8 imagery [23].

In summary, compared with traditional manual survey methods, remote sensing technology has greatly facilitated the estimation of the chlorophyll content of vegetation. Sentinel-2A data are ideal for monitoring vegetation health and chlorophyll changes because of their high spatial and temporal resolutions, multispectral band coverage, high accuracy and stability in modeling applications, free access, and broad applicability. Although satellite data provide valuable monitoring information, challenges remain. The diversity of surface conditions, limited access to ground validation data, and the limited generalizability and applicability of models may affect the estimation accuracy. In addition, the spatial resolution of imagery is a critical factor influencing the accuracy of vegetation information monitoring. Enhancing the spatial resolution of satellite data and improving pixel expressiveness present ongoing challenges in achieving high-precision estimations of chlorophyll content using remote sensing data. The above factors need to be considered to improve the accuracy and reliability of using satellite remote sensing data to estimate the chlorophyll content in forest areas. Therefore, this study was carried out to estimate the chlorophyll content of larch infested with yarrow larch looper in a typical infested area of Bindel, Kent Province, Mongolia, using Sentinel-2A data. The following issues were addressed:

1. Downscaling Sentinel-2A image data and decomposing mixed pixels in multispectral remote sensing images by simulating multispectral reflectance, and enhancing the expression of forest canopy information with Sentinel image pixels;

2. Analyzing the potential of the Sentinel-2A spectral vegetation index and its spectral derivative for estimating the chlorophyll content;

3. Improving the accuracy of chlorophyll content estimation via Sentinel-2A.

2. Materials and Methods

2.1. Study Data

2.1.1. Study Area

The experimental area was selected as the Bindel forest district (48° 26′ 14.75″–48° 26′ 34.96″ N, 110° 46′ 2.03″–110° 46′ 32.69″ E) of Kent Province, Mongolia (Figure 1), which is located in the central region of the Mongolian Plateau, with an average elevation of 1059.34 m. It has a temperate continental climate, with cold winters and warm summers, and the average annual precipitation is between 200 and 300 mm. The vegetation in this forest area is dominated by coniferous forests, which cover approximately 8% of the area of the province, and the coniferous forests are mainly composed of Siberian pine and larch, among other species. The selected study area experienced a severe outbreak of Yarl’s larch looper, with obvious signs of damage to trees.

In this study, 44 sample plots in a 10 × 10 m² area were selected as the experimental area (the sample plots were selected on the basis of the principle that the degree of forest damage within the sample plots was similar and the degree of larch dominance was greater than 80%). Within each sample plot, five sample trees were selected to obtain an even distribution, as shown in Figure 1d. Spectral measurements, relative chlorophyll content measurements, and leaf loss obtained were conducted for each sample tree, and the average values for the five trees within each sample plot were used to represent the data for the entire sample plot.

2.1.2. Multispectral Remote Sensing Data

• (1). Sentinel-2A satellite image data

Sentinel-2A is an Earth observation satellite that was launched by the European Space Agency (ESA) on 23 June 2015; Sentinel data are available through the ESA’s Copernicus Open Access Hub, among other sites. Sentinel-2A carries a high-resolution, multiband optical imaging instrument with a revisit period of 10 days, spanning 13 major bands; it is capable of capturing different surface features and information. These bands have different spatial resolutions of 10, 20, and 60 m. In this study, 2-scene image data were downloaded for the Binder test area on 26 June 2018 and 27 June 2019, and the bands used were a combination of B2(496.6), B3(560), B4(664.5), B5(703.9), B6(740.2), B7(782.5), B8(835.1), and B8A(864.8).

• (2). UAV RGB images

The test area was photographed via a DJI “Miku” Mavic2 Professional UAV (Shenzhen DJI Technology Co., Ltd., Shenzhen, China) equipped with a Hasselblad L1D-20c camera (Hasselblad, Gothenburg, Sweden), and the shooting time was consistent with that of the Sentinel-2A remote sensing image data. The flight altitude was set to 60 m, and the heading overlap and side overlap were 80% and 60%, respectively, such that spatial RGB image data with a resolution of 0.02 m were obtained.

2.1.3. Ground Hyperspectral Data

ASD FieldSpec4 (Analytical Spectral Devices, Boulder, Colorado, USA) and SVC HR-1024 (SpaceView, Portland, OR, USA) (350 to 2500 nm) portable ground spectrometers were used for spectral measurements of the sample larch canopy. The ASD FieldSpec4 spectrometer has a spectral range of 350–2500 nm, spectral resolutions of 3 and 10 nm, and sampling intervals of 1.4 and 2 nm. The SVC HR-1024 spectrometer has a spectral range of 350–2500 nm, spectral resolutions of 3.5, 9.5, and 6.5 nm, sampling intervals of 1.5, 3.6, and 2.5 nm, and a data interval of 1 nm. Throughout the entire measurement period, a probe with a field-of-view angle of 25° and oriented vertically downward at a height of approximately 0.2 m from the object of measurement was used. All spectra were measured under clear and cloudless weather conditions between 10:30 and 14:30 Beijing time. Each sample tree was vertically divided into upper, middle, and lower levels, a typical branch was selected from each level and observed 20 times, the spectra were corrected via a reference whiteboard before and after each observation, and the average of the 20 × 3 spectral reflectance values was used to represent the canopy reflectance value of the sample tree. Twenty sample trees with different health conditions were uniformly selected in the test area to obtain spectral curves.

2.1.4. Relative Chlorophyll Content (SPAD) Data

In this study, an SPAD-502 portable chlorophyll meter (Konica Minolta, Osaka, Japan) was used to measure the SPAD values of typical branches at the upper, middle, and lower levels of sample trees. Within the spectral measurement range of each branch, three values associated with different health conditions were selected to measure the SPAD value. Then, the average of the three measurements was used as the SPAD value of the branch; finally, the SPAD values of the branches at the three levels were averaged, and the result was recorded as the SPAD value of the sample tree.

2.2. Methodology

In the experimental area, starting from the sensitivity relationship between multispectral features and the chlorophyll content of conifers, the chlorophyll content was estimated via ground-based nonimaging hyperspectral data, RGB image data from UAVs, Sentinel-2A satellite image data, and ground-based actual measurements and surveys of the chlorophyll content. The technical flowchart of the research is shown in Figure 2.

2.2.1. Data Processing

• (1). Sentinel-2A multispectral data

The Sentinel-2A remotely sensed images obtained from the European Space Agency (ESA) website are Level-1C data and require data preprocessing. Radiometric calibration processing was performed using the Sentinel Data Radiometric Calibration Extension tool, Radiance Sentinel-2L1C. On this basis, the FLAASH module of ENVI was applied for atmospheric correction, and then the 10 m B2, B3, B4, and B8 bands and the 20 m B5, B6, B7, and B8A bands of these bands were extracted as research data for subsequent use.

• (2). UAV RGB data

The UAV RGB raw image consists of several high-definition aerial films, which need to be preprocessed. Firstly, the aerial films of take-off and landing phases are excluded from the RGB raw image data, and the aerial films taken when the UAV is flying according to the route are filtered out; then the selected aerial films are stitched together by using Agisoft Photoscan software 2.1.3 to generate a dense point cloud, mesh, and texture. Finally, the DEM and orthophoto maps are generated sequentially based on the dense point cloud, mesh, and texture.

• (3). Ground hyperspectral data

When collecting hyperspectral data, anomalies and duplicate spectra are inevitably present in the data due to background and instrumental noise. Therefore, the raw hyperspectral data were first processed to remove the abnormal and repeated spectra, then averaged, and then the average spectra were smoothed using the Smooth function to obtain the smoothed spectral reflectance curves (SSR) of the larch canopy for each sample.

• (4). Chlorophyll content data

On the basis of measurements of the sample trees in the experimental area, the maximum CHLC values were 50.30 and 50.46, respectively, and the minimum value was 2.74, with a mean value of 20.06 and a standard deviation of 12.54, which conformed to a normal distribution and met the modeling requirements.

2.2.2. Sentinel-2A Multispectral Reflectance Simulation

• (1). Downscaling of individual bands of Sentinel-2A images

The B5, B6, B7, and B8A bands were downscaled by analyzing the linear correlation between the spectral bands of the Sentinel-2A images. First, 320 samples were randomly selected from the images of the test area to establish one-dimensional linear relationships among the B5, B6, B7, and B8A bands (with a spatial resolution of 20 m) and B2, B3, B4, and B8 bands (with a spatial resolution of 10 m); second, the R² values of these linear relationships were compared, and the linear regression equations with the highest R² values were selected for the downscaling process.

• (2). Sentinel-2A image geometric correction

To downscale the Sentinel-2A image data, the bands used were combined, and a UAV RGB image was used as the reference image to perform geometric correction on the 8-band Sentinel-2A images. Six points in the image were selected as control points, the spatial coordinate alignment function in ENVI was used to complete the geometric correction, and the geometric correction accuracy was controlled within half an image element in the Sentinel-2A image.

• (3). Extraction of larch infestation information

In this study, 13 vegetation indices (Table 1) related to the external morphology and internal biochemical components of the forest trees were selected as the classification indices for healthy and damaged larch. A total of 15,000 samples were selected from the test area, with 5000 samples each of healthy larch, damaged larch, and other features. The differences among the spectral vegetation indices of these features on the basis of the F value of the variance statistic were obtained through the ANOVA method; then, a continuous projection algorithm (SPA) was used to screen the sensitive indices from the 13 vegetation indices. Finally, out of 15,000 samples, 10,000 samples were randomly selected as training data for modeling, and the remaining samples were used as validation data. The RF algorithm was applied to extract the information for healthy larch, damaged larch, and other land features.

• (4). Calculation of the degree of depression for healthy and damaged larch

Depression is the ratio of the projected area of the forest canopy on the ground to the image area under direct sunlight, which reflects the density of the stand at the image scale and is calculated via the following formula:

${CD}_H={\displaystyle \frac{{\displaystyle {\sum }_{Hj=1}^{N_{Hi}}}{S}_{Hj}}{S_i}}$

${CD}_D={\displaystyle \frac{{\displaystyle {\sum }_{Dj=1}^{N_{Di}}}{S}_{Dj}}{S_i}}$

where CD_H and CD_D are the healthy and damaged forest depression values for image elements, respectively; i is the image element serial number; S_i is the area of the ith image element; N_Hi and N_Di are the total numbers of healthy and damaged forest image elements of type i in RGB images; H_j and D_j are the serial numbers of RGB images of healthy and damaged forest areas obtained with the UAV, respectively (H_j = 1, 2, 3, and ……, N_Hi;Dj = 1, 2, 3, ……, N_Di); and S_Hj and S_Dj are the areas of the jth healthy and damaged forest elements in RGB Sentinel-2A images, respectively. The CD_H and CD_D values both lie between 0 and 1, with large values indicating high stand density.

• (5). Mixed-image element decomposition

Mixed-pixel interference should be eliminated to provide accurate information regarding forest reflectance. The Sentinel-2A pixels span a combination of healthy forests, damaged forests, and other features, and pixel reflectance is a weighted summation of these three types of reflectance, which can be expressed by the following equation:

${\mathsf{\rho }}_{\mathrm{i}}={\mathrm{k}}_{\mathrm{H}\mathrm{j}}{\mathsf{\rho }}_{\mathrm{H}\mathrm{i}}+{\mathrm{k}}_{\mathrm{D}\mathrm{j}}{\mathsf{\rho }}_{\mathrm{D}\mathrm{i}}+{\mathrm{k}}_{\mathrm{O}\mathrm{j}}{\mathsf{\rho }}_{\mathrm{O}\mathrm{i}}$

where ${\rho }_i$, ${\rho }_{Hi}$, ${\rho }_{Di}$, and ${\rho }_{Oi}$ denote the reflectance of mixed-image elements, healthy forest, damaged forest, and other features in wave band i of an image, respectively; $k_{Hj}$, $k_{Dj},$ and $k_{Oj}$ are the weights of healthy forest, damaged forest, and other features among the jth mixed-image elements in wave band i of an image, respectively, and $k_{Hj}+k_{Hj}+k_{Hj}=1$ and for mixed-image elements, the reflectance values of healthy forest and other features can be obtained via the following formula:

${\mathsf{\rho }}_{\mathrm{D}\mathrm{i}}=({\mathsf{\rho }}_{\mathrm{i}}-{\mathsf{\rho }}_{\mathrm{O}\mathrm{i}}\left(1-{\mathrm{k}}_{\mathrm{H}\mathrm{j}}-{\mathrm{k}}_{\mathrm{D}\mathrm{j}}\right)-{\mathrm{k}}_{\mathrm{H}\mathrm{j}}{\mathsf{\rho }}_{\mathrm{H}\mathrm{i}})/{\mathrm{k}}_{\mathrm{D}\mathrm{j}}$

where ${\mathrm{k}}_{\mathrm{H}\mathrm{j}}$ and ${\mathrm{k}}_{\mathrm{D}\mathrm{j}}$ can be expressed in terms of ${\mathrm{C}\mathrm{D}}_{\mathrm{H}\mathrm{j}}$ for healthy and ${\mathrm{C}\mathrm{D}}_{\mathrm{D}\mathrm{j}}$ for damaged forests, and ${\mathsf{\rho }}_{\mathrm{H}\mathrm{i}}$ and ${\mathsf{\rho }}_{\mathrm{O}\mathrm{i}}$ are the reflectance values of healthy larch and other features, which are obtained by averaging the reflectance values for more than 10 healthy forest areas and other features from the test area.

• (6). Multispectral reflectance simulation

  • (i). Sample-plot-based multispectral reflectance modeling

    The Sentinel-2A image multispectral reflectance is simulated via ground-based hyperspectral measured data, and the energy received from the Sentinel-2A satellite sensor in each wavelength band can be interpreted as the sum of the energy received at each wavelength within the wavelength range of the band. Therefore, the image multispectral reflectance simulation can be regarded as a redistribution of energy by the spectral response function, which is equivalent to a weighted average of the measured hyperspectral data by wavelength. On this basis, the simulated spectral reflectance of each band can be calculated via the following algorithm:

    ${\rho }_i={\displaystyle \frac{{\int }_{{\lambda }_{min}}^{{\lambda }_{max}}{f}_i\left(\lambda \right)\rho \left(\lambda \right)d\left(\lambda \right)}{{\int }_{{\lambda }_{min}}^{{\lambda }_{max}}{f}_i\left(\lambda \right)d\left(\lambda \right)}}$

    where ${\rho }_i$ and $f_i\left(\lambda \right)$ are the spectral reflectance and spectral response function of the ith band of the image to be simulated, $\mathsf{\rho }\left(\mathsf{\lambda }\right)$ is the ground-measured spectral reflectance of the sample plots, and ${\lambda }_{max} \mathrm{a}\mathrm{n}\mathrm{d} {\lambda }_{min}$ are the upper and lower bounds of the spectral range of the ith band of the image to be simulated, respectively.

    To simulate the multispectral reflectance of the sample plots, 44 sample plots were selected from the experimental area (each plot was as similar to the pure image element as possible), the canopy spectral reflectance of the sample trees in these sample plots was measured, and the spectral reflectance of the sample plots was simulated through the above method with the measured hyperspectral measured data and the spectral response function. Then, considering the relationship between the simulated and actual spectral reflectance, the Sentinel-2A image multispectral reflectance simulation model was constructed.

  • (ii). Element-based multispectral simulation of Sentinel-2A images

    On the basis of the reflectance extraction results for healthy and damaged larch, the reflectance values were simulated separately via the sample-based multispectral reflectance simulation model. Then, with the information for healthy (${CD}_{Hj}$) and damaged (${CD}_{Dj}$) larch, the reflectance of each waveband (${\rho }_{Fi}$) of image elements was calculated; notably, the reflectance of healthy forest (${\rho }_{Hi}^m$) and damaged forest (${\rho }_{Di}^m$) elements simulated with the algorithm was obtained with ${{\rho }_{Fi}=CD}_{Hj}{\rho }_{Hi}^m+{CD}_{Dj}{\rho }_{Di}^m$. ${\rho }_{Fi}$ obtained in this way can realistically reflect the reflectance of the larch canopy.

2.2.3. Estimation Models

• (1). Sensitive spectral feature extraction method

  • (i). Spectral index calculation

    The spectral index (denoted as SI) in this study encompasses a vegetation index and texture features. The vegetation index is obtained by calculating the spectral reflectance of two or more bands, which can simplify the spectral information, highlight the vegetation features, and attenuate the influence of background noise on the spectral features of the forest canopy. The spectral texture features reflect the roughness of the forest canopy for areas with different degrees of damage. In this study, the spectral vegetation index and texture features, which are closely related to the chlorophyll content of plants, were selected as the spectral features for chlorophyll estimation, and, consequently, the vegetation spectral index was calculated, as shown in Table 2. The texture features were calculated via a principal component analysis of the Sentinel-2A imagery, followed by the use of the first principal component, which was filtered on the basis of second-order probability statistics to obtain T_mean, T_variance, T_homogeneity, T_contrast, T_dissimilarity, T_entropy, T_ secondmoment and T_correlation, and eight other texture features.

  • (ii). Spectral derivative characterization

    The derivative of spectral reflectance can be used to identify spectral absorption features and reduce the interference of background noise. Therefore, in this study, the first-order derivatives of the reflectance of the spectral bands in the Sentinel-2A images (denoted D2, D3, D4, D5, D6, D7, D8, and D8A) and the first-order derivatives of the bands combined with the DN35, DR35, SD56, and SDDR indices (denoted D2, D3, D4, D5, D6, D7, D8, and D8A) were calculated with the image derivative module in ENVI as the spectral derivative features (denoted SDF) for the remote sensing monitoring of pests, as follows.

    DN35 = (D5 − D3)/(D5 + D3)

    DR35 = D5/D3

    SD56 = D5 + D6

    SDDR = (D5 + D6)/D4

    To compare the accuracy of chlorophyll estimates in cases with simulated and remotely sensed data, spectral indices and spectral derivatives were calculated separately for both cases.

  • (iii). Spectral sensitivity characterization

    To analyze the correlations between SI and SDF and CHLC, the corresponding coefficients of determination, R², were calculated. Then, an R² threshold was selected, and screening was performed to select the sensitive SI and SDF. Next, the SPA algorithm was used to downsize these sensitive spectral features and obtain the spectral features used in the final modeling process.

• (2). Monitoring models

To realize the remote sensing estimation of larch CHLC, a CHLC estimation model was constructed on the basis of a partial least squares regression (PLSR) algorithm using SI and SDF, which are sensitive to CHLC. When the model was constructed, data from 30 sample plots in the experimental area were used for model training, and data from the remaining 14 sample plots were used for model validation. Both the training and validation data were randomly selected, and the model was created in MATLAB R2023b.

2.2.4. Model Evaluation

To objectively evaluate the model performance, the coefficient of determination (R², characterizing the degree of fit between the model estimates and the true values) and the root mean square error (RMSE, characterizing the degree of dispersion between the model estimates and the true values) were used as the performance evaluation indices for the training model (denoted as ${\mathrm{R}}_{\mathrm{C}}^2$ and ${\mathrm{R}\mathrm{M}\mathrm{S}\mathrm{E}}_{\mathrm{C}}$) and the validation model (denoted as ${\mathrm{R}}_{\mathrm{V}}^2$ and ${\mathrm{R}\mathrm{M}\mathrm{S}\mathrm{E}}_{\mathrm{V}}$). The closer the value of R² is to 1 and the closer the value of RMSE is to 0, the higher the model accuracy is. To evaluate the accuracy and stability of the model, the average value of the coefficient of determination (denoted as $\mathrm{m}{\mathrm{R}}_{\mathrm{C}\mathrm{V}}^2$) and the average value of the root mean square error (denoted as ${\mathrm{m}\mathrm{R}\mathrm{M}\mathrm{S}\mathrm{E}}_{\mathrm{C}\mathrm{V}}$) for training and validation were calculated. Moreover, the relative errors of ${\mathrm{R}}_{\mathrm{C}}^2$ and ${\mathrm{R}}_{\mathrm{V}}^2$ (denoted as $\mathrm{r}{\mathrm{R}}_{\mathrm{C}\mathrm{V}}^2$) and the relative errors of ${\mathrm{R}\mathrm{M}\mathrm{S}\mathrm{E}}_{\mathrm{C}}$ and ${\mathrm{R}\mathrm{M}\mathrm{S}\mathrm{E}}_{\mathrm{V}}$ (denoted as ${\mathrm{r}\mathrm{R}\mathrm{M}\mathrm{S}\mathrm{E}}_{\mathrm{C}\mathrm{V}}$) were used as indicators of the stability of the model. The closer the $\mathrm{m}{\mathrm{R}}_{\mathrm{C}\mathrm{V}}^2$ value is to 1 and the closer the ${\mathrm{m}\mathrm{R}\mathrm{M}\mathrm{S}\mathrm{E}}_{\mathrm{C}\mathrm{V}}$ value is to 0, the higher the accuracy of the model is. Additionally, the closer the $\mathrm{r}{\mathrm{R}}_{\mathrm{C}\mathrm{V}}^2$ and ${\mathrm{r}\mathrm{R}\mathrm{M}\mathrm{S}\mathrm{E}}_{\mathrm{C}\mathrm{V}}$ values are to 0, the greater the stability of the model. Combining these indicators, we calculated the model performance index (MPI) to compare the performance of different models, and the closer the MPI value is to 1, the better the model performance is.

3. Results

3.1. Downscaling Results for Individual Bands of Sentinel-2A Images

The best fitting linear equation is shown in Figure 3, with values of 0.7943, 0.6759, 0.6564, and 0.6663, and the linear correlations were highly significant. Finally, the selected downscaling transformation equations were utilized to downscale the spectral bands B5, B6, B7, and B8A of the Sentinel-2A images in the test area.

3.2. Simulation of Sentinel-2A Multispectral Reflectance

3.2.1. UAV RGB Images-Based Health and Damage Larch Information Extraction

Variance analysis revealed significant differences in vegetation indices among healthy larch, damaged larch, and other features. represented by the variance statistical quantity F-value. Subsequently, the Successive Projection Algorithm (SPA) was utilized to select sensitive indices, with results as shown in Table 3. When F > 8.64, the corresponding p < 0.00001, indicating that the index is highly sensitive to the differences between healthy larch, damaged larch, and other features. The results demonstrated that, aside from the RGBVI and GRVI, the sensitivity of the remaining spectral indices was notably significant. To address the issue of multicollinearity among spectral indices, the SPA algorithm was employed for dimensionality reduction, and the selected sensitive spectral indices were r, g, ExR, ExGR, ExG, and ExB.

The model classification results are shown in Figure 4, and model validation revealed that the overall accuracy and kappa coefficient of the RF model were 0.8750 and 0.8193, respectively. This result indicated that the method was able to effectively discriminate between healthy and damaged larch.

3.2.2. Sentinel-2A Image Pixels-Based Healthy and Damaged Larch Depression Information Extraction

After successfully extracting the information of healthy and damaged larch, the depression degree of healthy and damaged larch based on Sentinel-2A image pixels was further calculated. The results of the depression degree calculation are shown in Figure 5. From the figure, it can be seen that the distribution of the damaged larch is large and concentrated, which indicates that the larch is more seriously attacked by Yarl’s larch.

3.2.3. Sample-Plot-Based Multispectral Reflectance Simulation for Sentinel-2A Imagery

The canopy spectral reflectance of the sample trees was tested in 44 sample plots, and the spectral reflectance of the sample plots was simulated by using the hyperspectral data obtained with the Sentinel-2A spectral response function. On this basis, a multispectral reflectance simulation model for Sentinel-2A images based on sample plots was established on the basis of the relationship between the simulated spectral reflectance of the sample plots and the Sentinel-2A spectral reflectance of the sample plots. The specific model equations are as follows:

$mB6=2.7040B6-0.1906$

$mB7=1.9761B7-0.1143$

$mB8=1.6139B8-0.0949$

$mB8A=1.719B8A-0.0833$

$(mB3+mB8)/2=0.6222mB6+0.0348$

$mB3=1.2444mB6-mB8+0.0696$

$(mB4+mB8A)/2=2.1177mB3-0.0025$

$mB4=4.2354mB3-mB8A-0.005$

$mB5=1.1541mB3+0.0385$

$mB2=0.3542mB4+0.0301$

where mB2, mB3, mB4, mB5, mB6, mB7, mB8, and mB8A represent the simulated Sentinel-2A spectral band reflectance values, and B2, B3, B4, B5, B6, B7, B8, and B8A represent the original Sentinel-2A spectral band reflectance values. The fits of the above equations are shown in Figure 6, and the linear correlations of the equations are highly significant, with R² values ranging from 0.6303 to 0.9844.

3.2.4. Image-Based Multispectral Reflectance Simulation with Sentinel-2A Images

Using a sample-based multispectral reflectance simulation model, the spectral reflectance of healthy and damaged larch trees was simulated and then combined with the depression information. The spectral reflectance of larch was calculated at the element level and compared with the nonsimulated spectral reflectance (Figure 7). As shown in the figure, the simulated spectral reflectance better reflects the spectral reflectance characteristics of the larch canopy, such as larch chlorophyll, with strong absorption in the blue (B2) and red (B4) bands, forming red valleys, than does the nonsimulated spectral reflectance. Notably, chlorophyll reflects most of the signal in the green (B3) band, forming small green peaks. The refractive indices of the porous and thin-walled cellular tissues of the larch needles and the cellular spacing vary for near-infrared light (B8 and B8A), with strong reflectance, resulting in a near-infrared high-reflection shoulder. Between the red valleys and the near-infrared high reflection shoulder, creating a steep red edge slope (B5–B6–B7), the simulated spectral reflectance of healthy larch better reflects the above spectral characteristics than does the nonsimulated reflectance. Moreover, compared with the nonsimulated spectral reflectance, the simulated spectral reflectance is more sensitive to larch damage. In the simulated spectral reflectance case, the decreases in the green light, red edge, and near-infrared bands and the increase in the red light band are significant for damaged larch, indicating that the simulated spectral reflectance is more sensitive to changes in the larch canopy than is the nonsimulated reflectance. The spectral reflectance data obtained with the Sentinel-2A satellite show excellent potential for conifer pest monitoring.

3.3. Spectral Feature Sensitivity Analysis

With increasing level of forest damage, the chlorophyll content gradually decreased, which caused the spectral reflectance values in the absorption and reflection bands to change considering the leaf pigment, resulting in spectral features such as the SI and SDF being sensitive to changes in chlorophyll content in different parts of the forest canopy. By analyzing the correlation between the spectral features (SI and SDF) and CHLC, the coefficient of determination R² of each spectral feature was calculated, and the R² distribution of each spectral feature was plotted (Figure 8). A critical and highly significant R² value (p < 0.00001, 0.3735) was used as a threshold to screen the spectral features that were highly significantly correlated with CHLC. The results revealed that the SIs with coefficients larger than the critical value were SAVI, RED_EDGE_NDVI, PSSR, PSRI, NDWI, NDVI, NDI45, MSAVI2, IRECI, GNDVI, EVI2, EVI, CRI1, CHL_RED_EDGE, BAI, ARI2, and T_mean, and the SDFs with coefficients larger than the critical value were D_B2, D_B3, D_B5, D_B6, D_B7, DN35, DR35, SD56, and SDDR, indicating that the above spectral features are highly related to forest CHLC. On this basis, the final sensitive spectral features were selected via the successive projection algorithm (SPA). The results showed that the selected sensitive SIs for forest CHLC estimation were SAVI and EVI2, and the sensitive SDFs were D_B2 and D_B5.

3.4. Conifer Chlorophyll Content Estimation Model and Accuracy Evaluation

The forest CHLC estimation model was constructed via the PLSR algorithm using SIs and SDFs. The results revealed that the CHLC estimation model based on the spectral features of simulated Sentinel-2A remote sensing data yielded the best results, and its model estimation performance was significantly better than that of the model with nonsimulated spectral features (Table 4). The performance of the SDF-PLSR model for CHLC estimation on the basis of simulated remote sensing data was best, with mR²_cv, mRMSE_cv, rR²_cv, rRMSE_cv, and MPI values of 0.8159, 0.1143, 0.1080, 0.1153, and 0.8636, respectively, which were higher than those of the SDF-PLSR model based on nonsimulated remote sensing data by at least 0.0404.

To further explore the performance of the spectral feature CHLC estimation model based on simulated Sentinel-2A remote sensing data, 1:1 straight-line fitting analyses were performed for the estimated and measured values of the modeling and validation sets with the PLSR estimation models constructed from both simulated and nonsimulated remote sensing data (SIs and SDFs; Figure 9). The results showed that the data points in the modeling and validation sets used for SI-PLSR and SDF-PLSR based on simulated remote sensing data were more concentrated and uniformly distributed along the 1:1 straight line, with a better result than that of the model with nonsimulated data. The CHLC estimation model exhibited the best fitting effect. For both SIs and SDFs, SDF-PLSR yielded a better 1:1 linear fit than did SI-PLSR, which was consistent with the results of previous comparison of model performance.

Figure 10a shows a graph of the CHLC estimates for forest trees in the test area on the basis of spectral features from simulated Sentinel-2A remote sensing data. The figure shows that there are some differences in the results on the basis of the SIs and SDFs. In Figure 9, most of the predicted points are distributed below the 1:1 line, and the SIPLSR model generally underestimates the CHLC value of yarrow larch looper-infested trees, reflecting a bias in the CHLC estimation of the model. Therefore, the estimates of the CHLC of infested stands on the basis of SDF spectral features from simulated Sentinel-2A remote sensing data are more reliable. Figure 10b shows the estimates of the CHLC of trees in the test area on the basis of spectral features from nonsimulated Sentinel-2A remote sensing data. Compared with those in the case with simulated data, the SI and SDF spectral feature-estimated CHLC values in this case are higher, and the chlorophyll contents estimated on the basis of SIs are consistent with those obtained with simulated data. Although the results based on observed values are different, the spatial distributions of the high and low values are consistent between cases, although the results differ more when SDFs are used, with nonsimulated data yielding higher estimates. According to Figure 10, the actual values are most consistent with the simulated data, and the SDF-based estimation results best reflect the true values.

4. Discussion

4.1. Multispectral Reflectance Simulation and Mixed-Image Decomposition Effects of Sentinel-2A Data

In this study, on the basis of the linear correlations between the spectral bands of Sentinel-2A images, downscaling transformations were performed for the B5, B6, B7, and B8A bands to increase the resolution of data in all eight bands to 10 m; the corresponding R² values were 0.7943, 0.6759, 0.6564, and 0.6663, respectively, and the linear correlations were highly significant. The downscaling treatment increased the spatial resolution of the image while maintaining the original spectral characteristics, thus enhancing the image details, which was beneficial for the spectral analysis of the chlorophyll content and the establishment of inversion models [38]. In addition, after downscaling, all the bands had the same resolution, and the calculation of the spectral vegetation index was consistent and accurate. Then, through the use of the hyperspectral measured data and the Sentinel-2A spectral response function, the spectral reflectance at the sample site was simulated, and through the relationship between the simulated and observed spectral reflectance at the sample site, a multispectral reflectance simulation model for Sentinel-2A imagery was established, with R² values ranging from 0.6303 to 0.9844 and highly significant linear correlations. By performing multispectral reflectance simulations, the atmospheric and topographic interference was reduced, the reflectance characteristics of the surface were accurately estimated, and the ground validation capability and accuracy of monitoring were improved [47]. On this basis, the hybrid pixel decomposition of satellite data was performed. Pixels in remote sensing images usually contain multiple feature types, and the spectral features are mixed at the pixel scale, resulting in the spectral reflectance of a single pixel not directly reflecting the real characteristics of a single feature [48]. Mixed-pixel decomposition can be used to decompose the mixed spectra of multiple feature types into spectral features and abundance ratios for a single feature, which enhances the classification accuracy and leads to an accurate estimation of the chlorophyll content. Using a sample-based multispectral reflectance simulation model, the spectral reflectance of larch for each image was calculated by combining the depression information for healthy and damaged larch trees.

The reflectance values before and after the decomposition of the mixed pixels in the three bands of B7, B8, and B8A were compared (Figure 11), and the reflectance values of the pixels in the middle part of the experimental area in the three bands were higher than those in other areas. According to the RGB images of the UAVs, the pixels with elevated reflectance values displayed a trend consistent the distribution of the forest trees. This may be because forest trees usually have high reflectance values, and after the mixed-image elements were decomposed and the spectral features of other types of features among the image elements were separated, the reflectance estimates improved. Second, multispectral simulation was performed to reduce the noise and errors in the images and best capture the true reflectance of features. In addition, the downscaling of the imagery improved the spectral resolution, making the reflectance values of the ground features more prominent, as indicated by the elevated reflectance in the spectral bands after decomposition shown in Figure 7.

The simulated spectral reflectance better reflects the spectral reflectance characteristics of the larch canopy and is more sensitive to larch damage compared to the nonsimulated spectral reflectance, indicating higher sensitivity to larch canopy changes. Chlorophyll strongly absorbs signals from the blue light (B2) and red light (B4) bands, forming a red valley; additionally, most signals from green light (B3) bands are reflected, forming a small green peak. The refractive index values of larch needles and leaves vary for porous thin-walled cell tissues (sponge tissues) and cellular interstitial tissues in terms of the reflection of near-infrared light (B8 and B8A), forming a near-infrared high-reflective shoulder. The red valleys and near-infrared high-reflective shoulder create steep slopes (B5–B6–B7). Overall, the simulated spectral reflectance of healthy larch reflects the above spectral characteristics better than the nonsimulated spectral reflectance. Compared with those obtained with nonsimulated spectral reflectance, the decreases in reflectivity in the green light, red edge, and near-infrared bands and the increase in reflectivity in the red light band obtained with simulated data for damaged larch were more significant; this finding is consistent with the results of Anushree [49]. In summary, performing hybrid pixel decomposition is crucial for monitoring vegetation health status via remote sensing imagery. This method can improve the accuracy of monitoring by revealing the composition and proportion of different feature types within a single pixel, better reflect the spectral characteristics of healthy/damaged vegetation, and provide more accurate information for vegetation monitoring under complex surface conditions.

4.2. Sensitivity of Spectral Characteristics to the Chlorophyll Content

Spectral indices and derivatives are often used to enhance spectral features through the use of image bands to improve the sensitivity to changes in the chlorophyll content, which can reflect the reflection and absorption characteristics of plant leaves at different wavelengths [41]. In this study, we used the continuous projection algorithm to screen the sensitive vegetation indices and derivatives and selected the SAVI and EVI2 from among the vegetation indices and D_B2 and D_B5 from among the derivatives as the input variables of the spectral indices and derivatives, respectively, in the model for estimating the chlorophyll content. Both the SAVI and EVI2 were calculated by combining the reflectance information in the B4 and B8 bands. Band 4 of the Sentinel data is a band within the visible spectrum, specifically at 665 nm, and is referred to as the red edge band. Chlorophyll strongly absorbs red light, so the red edge band is one of the important spectral bands reflecting the changes in chlorophyll content. Jan C et al. reported that red edge band data can be effectively used to monitor the chlorophyll content of crops, and changes in the red edge position are positively correlated with the chlorophyll content [3,17,50,51]. The reflectance decreased significantly with increasing chlorophyll content, and the red edge position moved to longer wavelengths. This relation can be used to estimate the chlorophyll content. Compared with other red light bands, band B4 is closer to the red edge region, so using information from this band can reduce the effect of the soil background on the vegetation index to a certain extent and, thus, more accurately reflect the growth of vegetation [52]. The specific wavelength of band 8 is 842.5 nm, which is located in the near-infrared region, and the reflectance in the near-infrared band is closely related to the chlorophyll content of the vegetation. In this band, when the chlorophyll content is relatively high, the reflectance of the vegetation in B8 increases accordingly, whereas the absorption is relatively low. This phenomenon is due mainly to the scattering and reflecting effects of the internal structure of vegetation leaf cells, especially considering the absorption properties of chlorophyll in this band [53]. Therefore, the reflectance information from the near-infrared and red light bands, such as the dual-band enhanced vegetation index (EVI2) and soil-adjusted vegetation index (SAVI) selected in this paper, can be combined to effectively estimate the chlorophyll content of vegetation. The EVI2 is an improved vegetation index that is selected to reduce the influence of the atmospheric and soil background noise through parameter adjustment [54] and improve the monitoring of changes in the chlorophyll content [55]. The SAVI is an improved vegetation index that can be used to reduce the effect of soil background radiation on vegetation index calculations through parameter adjustment [54], improve the sensitivity of monitoring changes in the chlorophyll content, and enhance estimates of the chlorophyll content of vegetation under atmospheric conditions. Specifically, the SAVI reduces the effect of soil background radiation on the vegetation index by introducing a soil brightness correction factor and accurately reflects the biophysical and biochemical characteristics of vegetation, such as the chlorophyll content [55]. Gao, DH et al. reported that the best estimate of the chlorophyll content in wheat was obtained via the SAVI [52]. Yang, HB et al. reported that among all the selected vegetation indices, the SAVI displayed the highest accuracy in estimating the chlorophyll content of potatoes [56]. Qiao, L et al. reported that there was a significant linear correlation between the vegetation indices and the chlorophyll content of the maize canopy and that the use of the EVI and SAVI enhances estimates [57]. These indices are based on combinations of spectral information from different bands and can be used to estimate the chlorophyll content of maize.

The spectral derivative is the derivative of spectral reflectance with respect to wavelength and can be used to highlight the subtle changes in spectral curves. By analyzing the trend and characteristics of the changes, it is possible to assess the chlorophyll content of plants and enhance the spectral characteristics to effectively identify and quantify the changes in the chlorophyll content of plant leaves. The D_B2 and D_B5 values selected in this paper were obtained via second-order derivations of Band 2 and Band 5, respectively. The Band 2 band of Sentinel-2A, also known as the blue band, with a wavelength range of approximately 490–530 nm, is sensitive to the reflective properties of vegetation in chlorophyll content monitoring; however, it is not the most sensitive band in comparison with other bands, such as the red light band (B4) and the near-infrared band (B8). The derivative of the B2 band encompasses the local characteristics of the spectral curve, making subtle changes in vegetation properties more obvious, which significantly enhances the sensitivity to changes in the chlorophyll content; thus, D_B2 was used as an input variable for modeling the chlorophyll content. Band 5, with a wavelength range of approximately 705–730 nm, is also located at the red edge of the spectral range of vegetation, which is the region with the most obvious chlorophyll absorption characteristics, such as absorption valleys. In addition, through the derivatization process, noise due to atmospheric scattering, the soil background, and other factors was reduced, thus improving the accuracy of chlorophyll content estimation [58]. Horler et al. investigated the relationship between vegetation spectra and chlorophyll concentrations, indicating the role of conductive light in the estimation of the chlorophyll concentration of vegetation [59]. Zhang, AW et al. utilized a fractional-order derivatization method to improve the sensitivity of spectral data to chlorophyll [60]. Spectral index and spectral derivative analyses are important for assessing the sensitivity of different factors to the chlorophyll content; the corresponding spectral features can be emphasized to improving the accuracy of monitoring and chlorophyll content estimates.

4.3. Accuracy of Chlorophyll Content (CHLC) Estimation with Sentinel-2A Data

In this study, 44 sample plots were selected in the experimental forest area, the hyperspectral and relative chlorophyll content data for the sample trees in the sample plots were collected, and the multispectral reflectance of the Sentinel-2A images was decomposed and simulated for the mixed pixels of the Sentinel-2A images. On this basis, the spectral features of Sentinel-2A images before and after simulation were separately obtained, and the sensitive spectral features for chlorophyll content estimation were selected to establish a model for estimating the chlorophyll content of damaged larch trees via the partial least squares method; the results revealed the influence of pests on the chlorophyll content of larch needles. Moreover, the results with and without simulated Sentinel-2A image data were compared to obtain an accurate method of estimating the chlorophyll content. The results show that the $\mathrm{m}{\mathrm{R}}_{\mathrm{C}\mathrm{V}}^2$ values of the PLSR-based chlorophyll content estimation models using spectral features were greater than 0.75, suggesting that combining spectral features with machine learning, neural network, or multivariate statistical methods can provide accurate estimates of the chlorophyll content of the vegetation canopy. On the basis of the CHLC estimation results and the model fitting effect for the two kinds of data, the SDF-PLSR model based on simulated data was the most effective, whereas the SI-PLSR model underestimated the CHLC value of yellow larch looper-infested trees, potentially because the spectral derivatives selected in this study are more sensitive to the CHLC content of trees than spectral indices are. Notably, most scholars use vegetation indices in similar research. In some previous studies, derivatives were shown to provide better monitoring results; for example, Aiwu Zhang et al. reported that the fractional order derivative (FOD) was superior to the vegetation index in grassland chlorophyll estimation, with improved monitoring accuracy [60]. Similarly, in the estimation of the chlorophyll content in maize canopies, Xuehong Zhang et al. found that the first-order derivative of the O2–A band was superior to the vegetation index, yielding higher prediction accuracy [61]. In addition, the features of the simulated data were best predicted in the derivative range of −0.2 to 0.5, whereas errors occurred from 0.5 to 1.0, with general underestimation potentially related to the saturation phenomenon of the spectral features. The saturation phenomenon usually occurs at high chlorophyll contents, and the spectral feature values cannot continue to increase with increasing chlorophyll content, thus limiting the accuracy of model estimates; notably, the position of the red edge is correlated with the chlorophyll content, and with increasing chlorophyll content, the position of the red edge shifts in the longwave direction (redshift phenomenon) [62]. However, when the chlorophyll content is very high, the movement of the red edge position may slow, resulting in highly concentrated chlorophyll content estimates; thus, chlorophyll changes can be best reflected at low to moderate chlorophyll contents, which is consistent with the results reported by Richardson [63]. The overall effect of the estimation model based on nonsimulated data is poor, with overestimated values, as equally reflected in the monitoring results. Notably, mixed-pixel interference, atmospheric effects, and the influence of the altitude angle of the sun cannot be removed, which directly affects the accuracy of the data.

5. Conclusions

In this study, we estimated the chlorophyll content of healthy and looper-infested yarrow larch by simulating the spectral reflectance of Sentinel-2A data, decomposing the mixed-image elements, selecting chlorophyll-sensitive vegetation indices and derivatives in combination with field measurement data, and constructing a model for estimating the chlorophyll content on the basis of a partial least squares method. Then, the results were analyzed and discussed, with the following findings:

• (1). SI and SDF spectral features based on simulated remote sensing data were significantly sensitive to the chlorophyll content of trees with yarrow larch looper infestations, and the highly sensitive spectral features for th chlorophyll content were the SAVI and EVI2 SIs and the D_B2 and D_B5 SDFs.

• (2). The estimation model based on simulated data yielded significantly higher accuracy than the model based on nonsimulated data, with better CHLC estimates. Notably, the $\mathrm{m}{\mathrm{R}}_{\mathrm{C}\mathrm{V}}^2$, $\mathrm{m}{\mathrm{R}\mathrm{M}\mathrm{S}\mathrm{E}}_{\mathrm{C}\mathrm{V}}$, $\mathrm{r}{\mathrm{R}}_{\mathrm{C}\mathrm{V}}^2$, $\mathrm{r}{\mathrm{R}\mathrm{M}\mathrm{S}\mathrm{E}}_{\mathrm{C}\mathrm{V}}$, and MPI of the SDF-PLSR-based model for pest CHLC estimation were 0.8159, 0.1143, 0.1080, 0.1153, and 0.8636, respectively, which was an improvement in the performance over the SDF-PLSR model on the basis of nonsimulated remote sensing data of at least 0.0404 in each case.

• (3). When the simulated and nonsimulated spectral reflectance values of image elements were compared, the simulated spectral reflectance better reflected the spectral reflectance characteristics of the larch canopy, and the simulated spectral reflectance was more sensitive to larch damage than the nonsimulated spectral reflectance was. Compared with those in the case of nonsimulated spectral reflectance, the decreases in damaged larch in the green light, red edge, and near-infrared bands and the increase in the red light band in the case of simulated data were more significant, indicating that the simulated values were more sensitive to changes in the larch canopy.

This study provides highly accurate estimation results for the chlorophyll content of larch damaged by pests, and a reliable technical tool for the rapid assessment, prevention, and control of forest pests is printed. Despite the methodological success of this study, its limitations still need to be recognized. For example, Sentinel-2 imagery is affected by atmospheric conditions and cloud cover when estimating chlorophyll content. Atmospheric aerosols and water vapor, as well as cloud cover, can degrade the image quality and affect the spectral signal, thus interfering with the accurate estimation of chlorophyll. In the future, multisource data fusion can be used to reduce the effects of cloud cover and atmospheric conditions by combining data from different sensors and different times of the day, e.g., Sentinel-1 is a radar imaging system that is able to penetrate cloud cover, regardless of weather conditions, and enables all-weather, all-day monitoring to improve the accuracy and reliability of chlorophyll content estimation.

Author Contributions

L.Y. and X.H. analyzed the data and wrote the paper; X.H. conceived and designed the experiments. D.Z. and J.Z. provided test site and measurement data; G.B., S.T., Y.B., D.G., D.A., D.E. and M.A. revised this manuscript. All authors have read and agreed to the published version of the manuscript.

Data Availability Statement

Data are contained within the article.

Conflicts of Interest

The authors declare no conflicts of interest.

Footnotes

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.

Figures and Tables

Figure 1. Experimental area: (a) Topography of Mongolia, (b) Sentinel-2A image of the test area, (c) UAV RGB map of the test area, (d) schematic diagram of the sample tree used for sample plot selection, (e) health sample tree, (f) damage sample tree.

Figure 2. Sentinel-2A image processing and chlorophyll content estimation methodological framework.

Figure 3. Linear fits to downscaled reflectance in the B5, B6, B7, and B8A bands.

Figure 4. RF model classification results.

Figure 5. Healthy and damaged larch depression: (a) health, (b) damage.

Figure 6. Effectiveness of fitting multispectral simulation models: (a–h) represent the fitting effects of the above equations, respectively.

Figure 7. Simulated and nonsimulated reflectance curves for each spectral band.

Figure 8. Spectral characteristics and CHLC correlation.

Figure 9. CHLC estimation model 1:1 linear fit: (a,b) show the fitted plots of the model results for the simulated remote sensing data, (c,d) show the fitted plots for the nonsimulated data.

Figure 10. Estimation of CHLC in insect-infested stands based on spectral features from simulated (a) and nonsimulated (b) remote sensing data.

Figure 11. Comparison of images before and after Sentinel-2A hybrid pixel decomposition.

Table 1

Vegetation index based on UAV RGB images and calculation formulas.

Vegetation Index	Formulation	References
VARI (visible atmospherically resistant index)	$(\mathrm{G}-\mathrm{R})/(\mathrm{G}+\mathrm{R}-\mathrm{B})$	[24]
ExG (excess green index)	$2\mathrm{G}-\mathrm{R}-\mathrm{B}$	[25]
ExR (excess red index)	$(1.4\mathrm{R}-\mathrm{G})/(\mathrm{G}+\mathrm{R}+\mathrm{B})$	[26]
ExB (excess blue index)	$(1.4\mathrm{B}-\mathrm{G})/(\mathrm{G}+\mathrm{R}+\mathrm{B})$	[26]
ExGR (extra green minus extra red)	$\mathrm{E}\mathrm{x}\mathrm{G}–\mathrm{E}\mathrm{x}\mathrm{R}$	[26]
GRVI (green red vegetation index)	$(\mathrm{G}-\mathrm{R})/(\mathrm{G}+\mathrm{R})$	[27]
MGRVI (modified green red vegetation index)	$({\mathrm{G}}^2-{\mathrm{R}}^2)/({\mathrm{G}}^2+{\mathrm{R}}^2)$	[28]
GLI (green leaf index)	$(2\mathrm{G}-\mathrm{R}-\mathrm{B})/(-\mathrm{R}-\mathrm{B})$	[29]
RGBVI (red, green and blue vegetation indices)	$({\mathrm{G}}^2-\mathrm{B}\times \mathrm{R})/({\mathrm{G}}^2+\mathrm{B}\times \mathrm{R})$	[28]
IKAW(Kawashima Index)	$(\mathrm{R}-\mathrm{B})/(\mathrm{R}+\mathrm{B})$	[30]
r (red index)	$\mathrm{R}/(\mathrm{R}+\mathrm{G}+\mathrm{B})$	[31]
g (green index)	$\mathrm{G}/(\mathrm{R}+\mathrm{G}+\mathrm{B})$	[31]
b (blue index)	$\mathrm{B}/(\mathrm{R}+\mathrm{G}+\mathrm{B})$	[31]

Note: R, G, and B in the table are the red, green, and blue band DN values of the UAV images, respectively.

Table 2

Vegetation index based on Sentinel-2 images and calculation formulas.

Vegetation Index (VI)	Formulation	References
ARI1 (Anthocyanin Reflectance Index 1)	$(1/\mathrm{B}3)-(1/\mathrm{B}5)$	[32]
ARI2 (Anthocyanin Reflectance Index 2)	$(\mathrm{B}8/\mathrm{B}3)-(\mathrm{B}8/\mathrm{B}5)$	[32]
BAI (Biochemical Affinity Index)	$1/\left(\right(0.1-\mathrm{B}4)2+(0.06-\mathrm{B}8\left)2\right)$	[33]
CRI1 (Carotenoid Reflectance Index 1)	$(1/\mathrm{B}2)-(1/\mathrm{B}3)$	[24]
CRI2 (Carotenoid Reflectance Index 2)	$(1/\mathrm{B}2)-(1/\mathrm{B}5)$	[24]
CHL RED EDGE	$\mathrm{B}5/\mathrm{B}8$	[24]
EVI (Enhanced Vegetation Index)	$2.5\times (\mathrm{B}8-\mathrm{B}4)/(\mathrm{B}8+6\times \mathrm{B}4-7.5\times \mathrm{B}2+1)$	[34]
EVI2 (Enhanced Vegetation Index 2)	$2.5\times (\mathrm{B}8-\mathrm{B}4)/(\mathrm{B}8+2.4\times \mathrm{B}4+1)$	[34]
GNDVI (Green Normalized Difference Vegetation Index)	$(\mathrm{B}8-\mathrm{B}3)/(\mathrm{B}8+\mathrm{B}3)$	[35]
IRECI (Inverted Red-Edge Chlorophyll Index)	$(\mathrm{B}7-\mathrm{B}4)\times \mathrm{B}6/\mathrm{B}5$	[36]
MCARI (Modified Chlorophyll Absorption in Reflectance Index)	$1-0.2\times (\mathrm{B}5-\mathrm{B}3)/(\mathrm{B}5-\mathrm{B}4)$	[37]
MSAVI2 (Modified Soil Adjusted Vegetation Index 2)	$(\mathrm{B}8+1)-0.5\times \mathrm{s}\mathrm{q}\mathrm{r}\mathrm{t}\left(\right(2\times \mathrm{B}8-1)2+8\times \mathrm{B}4))$	[38]
MTCI (MERIS Terrestrial Chlorophyll Index)	$(\mathrm{B}6-\mathrm{B}5)/(\mathrm{B}5-\mathrm{B}4)$	[39]
NDI45 (Normalized Difference Index 45)	$(\mathrm{B}5-\mathrm{B}4)/(\mathrm{B}5+\mathrm{B}4)$	[7]
NDVI (Normalized Difference Vegetation Index)	$(\mathrm{B}8-\mathrm{B}4)/(\mathrm{B}8+\mathrm{B}4)$	[40]
NDWI (Normalized Difference Water Index)	$(\mathrm{B}3-\mathrm{B}8)/(\mathrm{B}3+\mathrm{B}8)$	[41]
PSRI (Plant Senescence Reflectance Index)	$(\mathrm{B}4-\mathrm{B}2)/\mathrm{B}6$	[42]
PSSR (Plant Senescence Reflectance)	$\mathrm{B}8/\mathrm{B}4$	[43]
RED EDGE NDVI	$(\mathrm{B}8-\mathrm{B}6)/(\mathrm{B}8+\mathrm{B}6)$	[44]
SAVI (Soil Adjusted Vegetation Index)	$1.5\times (\mathrm{B}8-\mathrm{B}4)/(\mathrm{B}8+\mathrm{B}4+0.5)$	[45]
S2REP (Sentinel-2 Red Edge Position)	$705+35\times (0.5\times (\mathrm{B}7+\mathrm{B}4)-\mathrm{B}5)/(\mathrm{B}6-\mathrm{B}5)$	[46]

Table 3

Vegetation index variance and sensitivity index selection results.

Vegetation Index	F	Selection of Indicators
R	2254.10	●
G	3439.70	●
B	326.48	○
VARI	100.46	○
RGBVI	1.11	○
MGRVI	1.10	○
IKAW	257.52	○
GRVI	1.05	○
GLI	2988.70	○
ExR	3981.50	●
ExGR	4113.40	●
ExG	3439.70	●
ExB	1241.40	●

Note: Indicators marked ● in the table are selected indicators and those marked ○ are nonselected indicators.

Table 4

Comparison of CHLC estimation model performance.

Spectral Characteristics	Simulated Data					Nonsimulated Data
	mR2cv	mRMSEcv	rR2cv	rRMSEcv	MPI	mR2cv	mRMSEcv	rR2cv	rRMSEcv	MPI
SI	0.8150	0.1194	0.1796	0.3888	0.8029	0.7570	0.1245	0.1849	0.3984	0.7796
SDF	0.8159	0.1143	0.1080	0.1153	0.8636	0.7722	0.1648	0.1251	0.1528	0.8232

References

1. Bannari, A.; Khurshid, K.S.; Staenz, K.; Schwarz, J.W. A comparison of hyperspectral chlorophyll indices for wheat crop chlorophyll content estimation using laboratory reflectance measurements. IEEE Trans. Geosci. Remote Sens.; 2007; 45, pp. 3063-3074. [DOI: https://dx.doi.org/10.1109/TGRS.2007.897429]

2. Kandpal, K.C.; Kumar, A. Migrating from Invasive to Noninvasive Techniques for Enhanced Leaf Chlorophyll Content Estimations Efficiency. Crit. Rev. Anal. Chem.; 2023; 54, pp. 2583-2598. [DOI: https://dx.doi.org/10.1080/10408347.2023.2188425] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/36995248]

3. Clevers, J.; Kooistra, L.; van den Brande, M.M.M. Using Sentinel-2 Data for Retrieving LAI and Leaf and Canopy Chlorophyll Content of a Potato Crop. Remote Sens.; 2017; 9, 405. [DOI: https://dx.doi.org/10.3390/rs9050405]

4. Qian, B.; Ye, H.; Huang, W.; Xie, Q.; Pan, Y.; Xing, N.; Ren, Y.; Guo, A.; Jiao, Q.; Lan, Y. A sentinel-2-based triangular vegetation index for chlorophyll content estimation. Agric. For. Meteorol.; 2022; 322, 109000. [DOI: https://dx.doi.org/10.1016/j.agrformet.2022.109000]

5. Gara, T.W.; Skidmore, A.K.; Darvishzadeh, R.; Wang, T.J. Leaf to canopy upscaling approach affects the estimation of canopy traits. Gisci. Remote Sens.; 2019; 56, pp. 554-575. [DOI: https://dx.doi.org/10.1080/15481603.2018.1540170]

6. Chemura, A.; Mutanga, O.; Odindi, J. Empirical Modeling of Leaf Chlorophyll Content in Coffee (Coffea arabica) Plantations with Sentinel-2 MSI Data: Effects of Spectral Settings, Spatial Resolution, and Crop Canopy Cover. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens.; 2017; 10, pp. 5541-5550. [DOI: https://dx.doi.org/10.1109/JSTARS.2017.2750325]

7. Frampton, W.J.; Dash, J.; Watmough, G.; Milton, E.J. Evaluating the capabilities of Sentinel-2 for quantitative estimation of biophysical variables in vegetation. Isprs J. Photogramm. Remote Sens.; 2013; 82, pp. 83-92. [DOI: https://dx.doi.org/10.1016/j.isprsjprs.2013.04.007]

8. Revill, A.; Florence, A.; MacArthur, A.; Hoad, S.P.; Rees, R.M.; Williams, M. The Value of Sentinel-2 Spectral Bands for the Assessment of Winter Wheat Growth and Development. Remote Sens.; 2019; 11, 2050. [DOI: https://dx.doi.org/10.3390/rs11172050]

9. Ali, A.M.; Darvishzadeh, R.; Skidmore, A.; Gara, T.W.; O’Connor, B.; Roeoesli, C.; Heurich, M.; Paganini, M. Comparing methods for mapping canopy chlorophyll content in a mixed mountain forest using Sentinel-2 data. Int. J. Appl. Earth Obs. Geoinf.; 2020; 87, 102037. [DOI: https://dx.doi.org/10.1016/j.jag.2019.102037]

10. Chen, Z.L.; Jia, K.; Wei, X.Q.; Liu, Y.; Zhan, Y.L.; Xia, M.; Yao, Y.J.; Zhang, X.T. Improving leaf area index estimation accuracy of wheat by involving leaf chlorophyll content information. Comput. Electron. Agric.; 2022; 196, 106902. [DOI: https://dx.doi.org/10.1016/j.compag.2022.106902]

11. Wu, C.Y.; Wang, L.; Niu, Z.; Gao, S.A.; Wu, M.Q. Nondestructive estimation of canopy chlorophyll content using Hyperion and Landsat/TM images. Int. J. Remote Sens.; 2010; 31, pp. 2159-2167. [DOI: https://dx.doi.org/10.1080/01431161003614382]

12. Su, W.; Sun, Z.P.; Chen, W.H.; Zhang, X.D.; Yao, C.; Wu, J.Y.; Huang, J.X.; Zhu, D.H. Joint Retrieval of Growing Season Corn Canopy LAI and Leaf Chlorophyll Content by Fusing Sentinel-2 and MODIS Images. Remote Sens.; 2019; 11, 2409. [DOI: https://dx.doi.org/10.3390/rs11202409]

13. Croft, H.; Chen, J.M.; Zhang, Y.; Simic, A.; Noland, T.L.; Nesbitt, N.; Arabian, J. Evaluating leaf chlorophyll content prediction from multispectral remote sensing data within a physically-based modelling framework. Isprs J. Photogramm. Remote Sens.; 2015; 102, pp. 85-95. [DOI: https://dx.doi.org/10.1016/j.isprsjprs.2015.01.008]

14. Gu, C.P.; Li, J.; Liu, Q.H.; Zhang, H.; Liu, L.Y.; Mumtaz, F.; Dong, Y.D.; Zhao, J.; Wang, X.H.; Liu, C. Retrieving decametric-resolution leaf chlorophyll content from GF-6 WFV by assessing the applicability of red-edge vegetation indices. Comput. Electron. Agric.; 2023; 215, 108455. [DOI: https://dx.doi.org/10.1016/j.compag.2023.108455]

15. Darvishzadeh, R.; Skidmore, A.; Abdullah, H.; Cherenet, E.; Ali, A.; Wang, T.J.; Nieuwenhuis, W.; Heurich, M.; Vrieling, A.; O’Connor, B. et al. Mapping leaf chlorophyll content from Sentinel-2 and RapidEye data in spruce stands using the invertible forest reflectance model. Int. J. Appl. Earth Obs. Geoinf.; 2019; 79, pp. 58-70. [DOI: https://dx.doi.org/10.1016/j.jag.2019.03.003]

16. Delloye, C.; Weiss, M.; Defourny, P. Retrieval of the canopy chlorophyll content from Sentinel-2 spectral bands to estimate nitrogen uptake in intensive winter wheat cropping systems. Remote Sens. Environ.; 2018; 216, pp. 245-261. [DOI: https://dx.doi.org/10.1016/j.rse.2018.06.037]

17. Clevers, J.; Gitelson, A.A. Remote estimation of crop and grass chlorophyll and nitrogen content using red-edge bands on Sentinel-2 and-3. Int. J. Appl. Earth Obs. Geoinf.; 2013; 23, pp. 344-351. [DOI: https://dx.doi.org/10.1016/j.jag.2012.10.008]

18. Xu, J.; Quackenbush, L.J.; Volk, T.A.; Im, J. Estimation of shrub willow biophysical parameters across time and space from Sentinel-2 and unmanned aerial system (UAS) data. Field Crops Res.; 2022; 287, 108655. [DOI: https://dx.doi.org/10.1016/j.fcr.2022.108655]

19. Dong, T.F.; Meng, J.H.; Shang, J.L.; Liu, J.G.; Wu, B.F. Evaluation of Chlorophyll-Related Vegetation Indices Using Simulated Sentinel-2 Data for Estimation of Crop Fraction of Absorbed Photosynthetically Active Radiation. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens.; 2015; 8, pp. 4049-4059. [DOI: https://dx.doi.org/10.1109/JSTARS.2015.2400134]

20. Mahanta, D.K.; Bhoi, T.K.; Komal, J.; Samal, I.; Mastinu, A. Spatial, spectral and temporal insights: Harnessing high-resolution satellite remote sensing and artificial intelligence for early monitoring of wood boring pests in forests. Plant Stress; 2024; 11, 100381. [DOI: https://dx.doi.org/10.1016/j.stress.2024.100381]

21. Zhang, M.Z.; Su, W.; Fu, Y.T.; Zhu, D.H.; Xue, J.H.; Huang, J.X.; Wang, W.; Wu, J.Y.; Yao, C. Super-resolution enhancement of Sentinel-2 image for retrieving LAI and chlorophyll content of summer corn. Eur. J. Agron.; 2019; 111, 125938. [DOI: https://dx.doi.org/10.1016/j.eja.2019.125938]

22. Sun, Y.H.; Wang, B.Y.; Zhang, Z.X. Improving Leaf Area Index Estimation With Chlorophyll Insensitive Multispectral Red-Edge Vegetation Indices. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens.; 2023; 16, pp. 3568-3582. [DOI: https://dx.doi.org/10.1109/JSTARS.2023.3262643]

23. Kong, J.; Ryu, Y.; Jeong, S.; Zhong, Z.; Choi, W.; Kim, J.; Lee, K.; Lim, J.; Jang, K.; Chun, J. et al. Super resolution of historic Landsat imagery using a dual generative adversarial network (GAN) model with CubeSat constellation imagery for spatially enhanced long-term vegetation monitoring. Isprs J. Photogramm. Remote Sens.; 2023; 200, pp. 1-23. [DOI: https://dx.doi.org/10.1016/j.isprsjprs.2023.04.013]

24. Gitelson, A.A.; Stark, R.; Grits, U.; Rundquist, D.; Kaufman, Y.; Derry, D. Vegetation and soil lines in visible spectral space: A concept and technique for remote estimation of vegetation fraction. Int. J. Remote Sens.; 2002; 23, pp. 2537-2562. [DOI: https://dx.doi.org/10.1080/01431160110107806]

25. Woebbecke, D.M.; Meyer, G.E.; Von Bargen, K.; Mortensen, D.A. Color Indices for Weed Identifcation under Various Soil, Residue, and Lighting Conditions. Am. Soc. Agric. Eng.; 1995; 38, pp. 259-269. [DOI: https://dx.doi.org/10.13031/2013.27838]

26. Meyer, G.E.; Neto, J.C. Verification of Color Vegetation Indices for Automated Crop Imaging Applications. Comput. Electron. Agric.; 2008; 63, pp. 282-293. [DOI: https://dx.doi.org/10.1016/j.compag.2008.03.009]

27. Tucker, C.J. Red and Photographic Infrared Linear Combinations for Monitoring Vegetation. Remote Sens. Environ.; 1979; 8, pp. 127-150. [DOI: https://dx.doi.org/10.1016/0034-4257(79)90013-0]

28. Bendig, J.; Yu, K.; Aasen, H.; Bolten, A.; Bennertz, S.; Broscheit, J.; Gnyp, M.L.; Bareth, G. Combining Uav-Based Plant Height from Crop Surface Models, Visible, and near Infrared Vegetation Indices for Biomass Monitoring in Barley. Int. J. Appl. Earth Obs. Geoinf.; 2015; 39, pp. 79-87. [DOI: https://dx.doi.org/10.1016/j.jag.2015.02.012]

29. Louhaichi, M.; Borman, M.M.; Johnson, D.E. Spatially Located Platform and Aerial Photography for Documentation of Grazing Impacts on Wheat. Geocarto Int.; 2001; 16, pp. 65-70. [DOI: https://dx.doi.org/10.1080/10106040108542184]

30. Kawashima, S.; Nakatani, M. An Algorithm for Estimating Chlorophyll Content in Leaves Using a Video Camera. Ann. Bot.; 1998; 81, pp. 49-54. [DOI: https://dx.doi.org/10.1006/anbo.1997.0544]

31. Lu, N.; Zhou, J.; Han, Z.; Li, D.; Cao, Q.; Yao, X.; Tian, Y.; Zhu, Y.; Cao, W.; Cheng, T. Improved Estimation of Aboveground Biomass in Wheat from Rgb Imagery and Point Cloud Data Acquired with a Low-Cost Unmanned Aerial Vehicle System. Plant Methods; 2019; 15, 17. [DOI: https://dx.doi.org/10.1186/s13007-019-0402-3] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/30828356]

32. Gitelson, A.A.; Merzlyak, M.N.; Chivkunova, O.B. Optical Properties and Nondestructive Estimation of Anthocyanin Content in Plant Leaves. Photochem. Photobiol.; 2001; 74, pp. 38-45. [DOI: https://dx.doi.org/10.1562/0031-8655(2001)074<0038:OPANEO>2.0.CO;2] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/11460535]

33. Chuvieco, E.; Martín, M.P.; Palacios, A. Assessment of different spectral indices in the red-near-infrared spectral domain for burned land discrimination. Int. J. Remote Sens.; 2002; 23, pp. 5103-5110. [DOI: https://dx.doi.org/10.1080/01431160210153129]

34. Jiang, Z.Y.; Huete, A.R.; Didan, K.; Miura, T. Development of a two-band enhanced vegetation index without a blue band. Remote Sens. Environ.; 2008; 112, pp. 3833-3845. [DOI: https://dx.doi.org/10.1016/j.rse.2008.06.006]

35. Gitelson, A.A.; Kaufman, Y.J.; Merzlyak, M.N. Use of a green channel in remote sensing of global vegetation from EOS-MODIS. Remote Sens. Environ.; 1996; 58, pp. 289-298. [DOI: https://dx.doi.org/10.1016/S0034-4257(96)00072-7]

36. Clevers, J.G.P.W.; de Jong, S.M.; Epema, G.F.; van der Meer, F.; Bakker, W.H.; Skidmore, A.K.; Addink, E.A. MERIS and the red-edge position. Int. J. Appl. Earth Obs. Geoinf.; 2001; 3, pp. 313-320. [DOI: https://dx.doi.org/10.1016/S0303-2434(01)85038-8]

37. Daughtry, C.S.T.; Walthall, C.L.; Kim, M.S.; de Colstoun, E.B.; McMurtrey, J.E. Estimating corn leaf chlorophyll concentration from leaf and canopy reflectance. Remote Sens. Environ.; 2000; 74, pp. 229-239. [DOI: https://dx.doi.org/10.1016/S0034-4257(00)00113-9]

38. Qi, J.G.; Chehbouni, A.R.; Huete, A.R.; Kerr, Y.H.; Sorooshian, S. A Modified Soil Adjusted Vegetation Index. Remote Sens. Environ.; 1994; 48, pp. 119-126. [DOI: https://dx.doi.org/10.1016/0034-4257(94)90134-1]

39. Dash, J.; Curran, P.J. Evaluation of the MERIS terrestrial chlorophyll index (MTCI). Adv. Space Res.; 2007; 39, pp. 100-104. [DOI: https://dx.doi.org/10.1016/j.asr.2006.02.034]

40. Rouse, J.W.; Haas, R.H.; Schell, J.A.; Deering, D.W. Monitoring Vegetation Systems in the Great Plains with Erts. Proceedings of the Third ETRS-1 Symposium; Washington, DC, USA, 10–14 December 1974; Fraden, S.C.; Marcanti, E.P.; Bechker, M.A. Scientific and Technical Information Office, National Aeronautics and Space Administration: Hampton, VA, USA, 1974; Volume 10–14, pp. 309-317.

41. Gao, B.C. NDWI—A normalized difference water index for remote sensing of vegetation liquid water from space. Remote Sens. Environ.; 1996; 58, pp. 257-266. [DOI: https://dx.doi.org/10.1016/S0034-4257(96)00067-3]

42. Merzlyak, M.N.; Gitelson, A.A.; Chivkunova, O.B.; Rakitin, V.Y. Non-destructive optical detection of pigment changes during leaf senescence and fruit ripening. Physiol. Plant.; 1999; 106, pp. 135-141. [DOI: https://dx.doi.org/10.1034/j.1399-3054.1999.106119.x]

43. Blackburn, G.A. Quantifying chlorophylls and caroteniods at leaf and canopy scales: An evaluation of some hyperspectral approaches. Remote Sens. Environ.; 1998; 66, pp. 273-285. [DOI: https://dx.doi.org/10.1016/S0034-4257(98)00059-5]

44. Fernández-Manso, A.; Fernández-Manso, O.; Quintano, C. SENTINEL-2A red-edge spectral indices suitability for discriminating burn severity. Int. J. Appl. Earth Obs. Geoinf.; 2016; 50, pp. 170-175. [DOI: https://dx.doi.org/10.1016/j.jag.2016.03.005]

45. Huete, A.R. A soil-adjusted vegetation index (SAVI). Remote Sens. Environ.; 1988; 25, pp. 295-309. [DOI: https://dx.doi.org/10.1016/0034-4257(88)90106-X]

46. Guyot, G.; Baret, F. Utilisation De La Haute Resolution Spectrale Pour Suivre Letat Des Couverts Vegetaux. Spectr. Signat. Objects Remote Sens.; 1988; 287, pp. 279-286.

47. Wang, Z.J.; Sui, L.C.; Zhang, S.Q. Generating Daily Land Surface Temperature Downscaling Data Based on Sentinel-3 Images. Remote Sens.; 2022; 14, 5752. [DOI: https://dx.doi.org/10.3390/rs14225752]

48. Du, M.M.; Li, M.Z.; Noguchi, N.; Ji, J.T.; Ye, M.C. Retrieval of Fractional Vegetation Cover from Remote Sensing Image of Unmanned Aerial Vehicle Based on Mixed Pixel Decomposition Method. Drones; 2023; 7, 43. [DOI: https://dx.doi.org/10.3390/drones7010043]

49. Badola, A.; Panda, S.K.; Roberts, D.A.; Waigl, C.F.; Bhatt, U.S.; Smith, C.W.; Jandt, R.R. Hyperspectral Data Simulation (Sentinel-2 to AVIRIS-NG) for Improved Wildfire Fuel Mapping, Boreal Alaska. Remote Sens.; 2021; 13, 1693. [DOI: https://dx.doi.org/10.3390/rs13091693]

50. Filella, I.; Penuelas, J. The Red Edge Position and Shape as Indicators of Plant Chlorophyll Content, Biomass and Hydric Status. Int. J. Remote Sens.; 1994; 15, pp. 1459-1470. [DOI: https://dx.doi.org/10.1080/01431169408954177]

51. Pinar, A.; Curran, P.J. Technical Note Grass chlorophyll and the reflectance red edge. Int. J. Remote Sens.; 1996; 17, pp. 351-357. [DOI: https://dx.doi.org/10.1080/01431169608949010]

52. Gao, D.H.; Qiao, L.; Song, D.; Li, M.Z.; Sun, H.; An, L.L.; Zhao, R.M.; Tang, W.J.; Qiao, J.B. In-field chlorophyll estimation based on hyperspectral images segmentation and pixel-wise spectra clustering of wheat canopy. Biosyst. Eng.; 2022; 217, pp. 41-55. [DOI: https://dx.doi.org/10.1016/j.biosystemseng.2022.03.003]

53. Zhang, H.; Li, J.; Liu, Q.H.; Lin, S.R.; Huete, A.; Liu, L.Y.; Croft, H.; Clevers, J.; Zeng, Y.L.; Wang, X.H. et al. A novel red-edge spectral index for retrieving the leaf chlorophyll content. Methods Ecol. Evol.; 2022; 13, pp. 2771-2787. [DOI: https://dx.doi.org/10.1111/2041-210X.13994]

54. Biparva, M.; Tsotsos, J. Contextual Interference Reduction by Selective Fine-Tuning of Neural Networks. arXiv; 2020; arXiv: 2011.10857

55. Chen, X.H.; Guo, Z.F.; Chen, J.; Yang, W.; Yao, Y.M.; Zhang, C.S.; Cui, X.H.; Cao, X. Replacing the Red Band with the Red-SWIR Band (0.74ρ_red + 0.26ρ_swir) Can Reduce the Sensitivity of Vegetation Indices to Soil Background. Remote Sens.; 2019; 11, 851. [DOI: https://dx.doi.org/10.3390/rs11070851]

56. Yang, H.B.; Hu, Y.H.; Zheng, Z.Z.; Qiao, Y.C.; Zhang, K.L.; Guo, T.F.; Chen, J. Estimation of Potato Chlorophyll Content from UAV Multispectral Images with Stacking Ensemble Algorithm. Agronomy; 2022; 12, 2318. [DOI: https://dx.doi.org/10.3390/agronomy12102318]

57. Qiao, L.; Tang, W.J.; Gao, D.H.; Zhao, R.M.; An, L.L.; Li, M.Z.; Sun, H.; Song, D. UAV-based chlorophyll content estimation by evaluating vegetation index responses under different crop coverages. Comput. Electron. Agric.; 2022; 196, 106775. [DOI: https://dx.doi.org/10.1016/j.compag.2022.106775]

58. Liu, B.; Yue, Y.M.; Li, R.; Shen, W.J.; Wang, K.L. Plant Leaf Chlorophyll Content Retrieval Based on a Field Imaging Spectroscopy System. Sensors; 2014; 14, pp. 19910-19925. [DOI: https://dx.doi.org/10.3390/s141019910]

59. Horler, D.N.H.; Dockray, M.; Barber, J.; Barringer, A.R. Red edge measurements for remotely sensing plant chlorophyll content. Adv. Space Res.; 1983; 3, pp. 273-277. [DOI: https://dx.doi.org/10.1016/0273-1177(83)90130-8]

60. Zhang, A.W.; Yin, S.N.; Wang, J.; He, N.P.; Chai, S.T.; Pang, H.Y. Grassland Chlorophyll Content Estimation from Drone Hyperspectral Images Combined with Fractional-Order Derivative. Remote Sens.; 2023; 15, 5623. [DOI: https://dx.doi.org/10.3390/rs15235623]

61. Zhang, X.H.; He, Y.; Wang, C.; Xu, F.; Li, X.H.; Tan, C.W.; Chen, D.M.; Wang, G.J.; Shi, L.X. Estimation of Corn Canopy Chlorophyll Content Using Derivative Spectra in the O₂—A Absorption Band. Front. Plant Sci.; 2019; 10, 1047. [DOI: https://dx.doi.org/10.3389/fpls.2019.01047]

62. Gao, S.; Zhong, R.; Yan, K.; Ma, X.L.; Chen, X.K.; Pu, J.B.; Gao, S.C.; Qi, J.B.; Yin, G.F.; Myneni, R.B. Evaluating the saturation effect of vegetation indices in forests using 3D radiative transfer simulations and satellite observations. Remote Sens. Environ.; 2023; 295, 113665. [DOI: https://dx.doi.org/10.1016/j.rse.2023.113665]

63. Song, D.; Gao, D.H.; Sun, H.; Qiao, L.; Zhao, R.M.; Tang, W.J.; Li, M.Z. Chlorophyll content estimation based on cascade spectral optimizations of interval and wavelength characteristics. Comput. Electron. Agric.; 2021; 189, 106413. [DOI: https://dx.doi.org/10.1016/j.compag.2021.106413]