Developing a New Vegetation Index Using Cyan, Orange, and Near Infrared Bands to Analyze Soybean Growth Dynamics

Abstract

Remote sensing Vegetation Indices (VIs) are simple, effective, and widely used methods for quantitative and qualitative analysis of vegetation cover, vigor, and growth dynamics. This study developed and assessed a new vegetation index (VI) using Cyan, Orange, and Near Infrared (NIR) bands to assess Soybean growth dynamics. The study was conducted at Lakehead University Agriculture Research Station, Thunder Bay, Canada, over four reproductive stages of Soybean growth (R4–R7). Spectral profiles were created for each stage, and the correlation between each spectral band at different stages was tested. There was no linear correlation between different bands except the correlation between the Cyan and Orange bands at R5 and R6 stages. Existing VIs have also been explored using approximately similar band combinations. Based on this analysis, three VIs were proposed for this new camera, and their behavior at different stages was evaluated using Leaf Area Index (LAI). Cyan and Orange spectral values were relatively high in the first and last growing seasons, while NIR values increased dramatically in the mid-growing seasons and decreased in the last stage. VI_NIR,O,C index showed the best results for mid-growing seasons (correlation with LAI = 0.39 for R5 and R6). VI_C,_O index showed a high level of details visually (leaves and background) for R4 and R7 than the other indices and correlated highly with LAI (0.48 and −0.5, respectively). Overall, the study provided new VIs that can be used to effectively analyze Soybean growth dynamics, with different VIs showing reliability over different growing stages.

1. Introduction

Several natural and anthropogenic factors, such as irrigation, nutrition, sunlight, and temperature, affect vegetation health dynamics [1]. Lack or excess of those variables leads to visible signs on vegetation or their growth dynamics, resulting in reduced yield or poor-quality harvest. According to the precision agronomy concept, good management and fewer resources can deliver better results, improving the crop’s health and yield [2]. Usually, farmers use their traditional knowledge, visual greenness observations, and empiric methods to determine plant health, ultimately making management decisions to protect their cultivations. The green color of a plant is an indicator of its level of chlorophyll which absorbs sunlight for the photosynthesis process [3]. Therefore, leaf chlorophyll can be used to estimate leaf photosynthetic capacity and, thus, vegetation health because of its close relationship with leaf nitrogen content that affects vegetation growth dynamics [4,5]. The procedures commonly used to determine leaf chlorophyll are destructive chemical analysis methods that involve the extraction of chlorophyll from leaf matter with organic solvents, followed by a spectrophotometric assay of chlorophyll [6,7,8]. However, this approach is time-consuming and expensive. Recently, several remote sensing approaches (indirect) were introduced to estimate vegetation health [9]. These remotely sensed methods are non-destructive, fast, and efficient. Vegetation health or vigor can be indirectly observed by using vegetation indices (VIs) derived from remotely sensed images [10]. Another variable used to measure health is the Leaf Area Index (LAI), the ratio of leaf area to ground area, typically reported as meters per square meter, a commonly used biophysical characteristic of vegetation [11]. It was also determined that LAI could be related to yield, another indicator of vegetation health, following the idea that healthy plants will have greater yield results than unhealthy plants [11].

1.1. Remote Sensing for Assessing Vegetation Health

Remote sensing of vegetation health monitoring provides time, cost-effective, and repeatable recordings and evaluations of status, stress, disturbances, and resource limitations over the short to long term for local and global vegetation monitoring [12,13]. The broad wavelength region of 400–700 nm is described as the most active region for leaf pigments or chlorophyll [14]. Based on the healthy vegetation spectral profile, healthy leaf pigments show higher reflectance values at the electromagnetic spectrum’s green and Near Infrared (NIR) regions [15,16]. Additionally, the Red Edge Region is the best remote sensing region descriptor of chlorophyll concentration (680–750 nm) [4].

A widely used remote sensing approach for health monitoring is using Vegetation Indices (VIs) derived from different band combinations to detect vegetation changes over time. Mathematical expressions of VIs can be selected to show how different plant components reflect the amount of electromagnetic energy. Therefore, VIs provides quantitative measures of vegetation health. This approach is time and cost-effective, reliable, and quickly covers large areas. The existing mathematical formulas for VIs were developed considering the healthy vegetation spectral profile and thus provide reasonable measures for vegetation dynamics [17]. Several studies used LAI, which is also a parameter for global and regional models of the biosphere/atmosphere exchange of carbon dioxide and water vapor, light interception, and other leaf functions and processes (photosynthesis) to predict chlorophyll content using remote sensing techniques [4,5,18,19,20]. For instance, the spatial distribution of mangrove canopy chlorophyll content can be mapped by using laboratory-measured canopy chlorophyll concentrations, LAI, and remotely sensed images [20].

Existing Vegetation Indices

Many of the existing VIs use the NIR and red spectral bands in different mathematical combinations to express quantitative and qualitative features; for example, one of the widely used VIs is the Normalized Difference Vegetation Index (NDVI); this index indicates the structure and greenness of vegetation using NIR and red spectral bands as expressed in

Table 1. Selected Vegetation Indices that are approximately equal to the spectral bands used in this study.

Vegetation Index	Mathematical Expression
Difference Vegetation Index (DVI) [22]	$\mathrm{NIR} - \mathrm{Red}$
Ratio Vegetation Index (RVI) [23]	$\mathrm{Red}/\mathrm{NIR}$
Normalized Difference Vegetation Index (NDVI) [24]	$\left(\mathrm{NIR} - \mathrm{Red}\right)/\left(\mathrm{NIR} + \mathrm{Red}\right)$
Carotenoid Reflectance Index 1 (CRI1) [21]	$\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{${\mathrm{p}}_{510}$}\right.-\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{${\mathrm{p}}_{550}$}\right.$
Soil Adjusted Vegetation Index (SAVI) [25]	$(\left(\mathrm{NIR}-\mathrm{Red}\right)/\left(\left(\mathrm{NIR}+\mathrm{Red}\right)+\mathrm{L}\right))\times \left(1+\mathrm{L}\right)$

. Carotenoid Reflectance Index 1 (CRI1), which ranges from one to eleven for green vegetation, is commonly used to assess stressed vegetation [21].

In the visible spectral range (approximately 380–700 nm), the blue, green, and red bands are the most used bands. For example, two perpendicular vegetation index (PVI) models were applied, the PVI and PVI6, that yielded significant coefficients of determination (R²) of 0.522 and 0.659, respectively, with LAI [22]. It is possible to use the visible spectral range, specifically the green region, to sense the chlorophyll concentration in order to measure the rate of photosynthesis and to monitor plant stress [26].

With the development of technology, new sensors with different spectral band combinations than the traditional blue, green, red, and NIR are being introduced to the market. For instance, Mapir Survey 3W OCN camera is sensitive to the Cyan (480–520 nm), Orange (585–620 nm), and NIR (780–870 nm) [2] (

Table 2. Bandwidth (nm) for Mapir Survey 3 Wide-angle RGN and OCN cameras.

Mapir Survey 3 (Wide Angle)
OCN Camera			RGN Camera
Cyan	Orange	NIR	Green	Red	NIR
460–520	585–645	780–870	530–570	640–680	820–880

). According to the OCN camera specifications, the OCN camera is the improved version of the RGN camera, which is sensitive to green (530–570 nm), red (640–680 nm), and NIR (820–880 nm) bands of the electromagnetic spectrum by increasing the contrast within vegetation and reducing soil noise [27,28]. The OCN camera is more sensitive to a relatively wide NIR region than the RGN camera. However, there is no in-depth analysis of spectral profiles of different bands or VIs for OCN camera band combinations as this camera is relatively new. Therefore, this study aimed to develop a new VI for analyzing Soybean plant growth dynamics using the OCN camera, which is sensitive to the electromagnetic spectrum’s Cyan, Orange, and NIR bands. The specific objectives are to (1) develop a spectral profile for Soybean’s growing stages; (2) compare the performance of different VIs at different growing seasons; (3) validate the performance using Canopy LAI; and (4) propose the best-fitted VI for Cyan, Orange and NIR bands.

2. Materials and Methods

2.1. Data and Study Area

The study was conducted at the Lakehead University Agriculture Research Station (LUARS), Thunder Bay, Ontario, Canada (48.3051°N, 89.3881°W). Several Soybean species were tested on a large area (87 m × 32 m area), and the study selected two locations to set up cameras (Figure 1).

A massive increase in Soybean production and international trade during the past 60 years has been a significant cause of the world’s average human life span increasing from 46.6 to 69.3 years during that time period [29]. An increase in dietary protein has been especially beneficial in developing countries where diets have historically been too low in protein [29]. In Canada, Soybean is a highly produced crop, with over 1,260,400 seeded hectares in Ontario [30].

Remote Sensing DataImage acquisition started on 26 July 2022 and finished on 19 September 2022: This study followed the regional reproductive stage identification details and divided the period into four reproductive stages, hereafter denoted as “R4–R7” [31,32]; Reproductive Stage 8 (R8) was the harvesting stage.

The study used two Mapir Survey 3W aerial cameras [33]. One is sensitive to Red (660 nm), Green (550 nm), and NIR (850 nm) bands (hereafter denoted as “RGN camera”), and the other is sensitive to Cyan (490 nm), Orange (615 nm), and NIR (808 nm) bands (hereafter denoted as “OCN camera”); Table 2 explains the spectral range of these cameras. Noticeably, the NIR regions of the cameras are different (Table 2). Two Soybean plots were selected for analysis based on their plant density. Station 01 had sparsely distributed Soybean plants, and two cameras (OCN and RGN) were set up at a 90° angle (looking straight down) (Figure 2). Station 02 had dense Soybean plants and arranged two cameras at a 90° angle (OCN camera) and a 45° angle (RGN camera) (Figure 2). Images were captured every 30 min daily from 7.00 a.m. to 5.30 p.m. during the season. Images were downloaded once a month and sorted into the different reproductive stages and times of the day morning (7:00 a.m. to 11:00 a.m.), midday (11:00 a.m. to 2:00 p.m.), and afternoon (2:00 p.m. to 5:00 p.m.). Then, they were calibrated (converted DN values to spectral reflectance values) using the MAPIR Camera Control (MCC) Software version 20221111 [34] for further analysis using R Studio [35].

2.2. Spectral Signatures

To derive spectral profiles for each stage using an OCN camera, approximately 30 midday images were selected for each stage (R4–R7); a random set of points were selected inside Soybean leaves and extracted corresponding spectral reflectance values for each band. There were approximately 1000 data points for each stage. Exploratory Data Analysis (EDA) was done based on scatter plots (index plots), box plots, histograms, and Q-Q plots. Based on EDA, outliers were removed, and a set of spectral reflectance values for each stage was prepared (1000 points per each stage). The spectral values were plotted against each wavelength, and the mean spectral profiles were extracted for each stage (R4–R7).

2.3. Vegetation Indices

To date, no unified mathematical expression defines all VIs due to the complexity of different light spectra combinations, instrumentation, platforms, and spectral and spatial resolutions used [10]. Table 1 shows selected VIs that use approximately the same spectral bands as one of the cameras used for this study.

To develop the new VI, several existing band combinations (mathematical expressions) were initially tested (

Table 3. Tested Vegetation Indices based on existing Vegetation Indices.

Original Vegetation Index	Symbology	Proposed Vegetation Index
Difference Vegetation Index (DVI)	dvi_on	NIR − Orange
Ratio Vegetation Index (RVI)	sr_on	Orange/NIR
Ratio Vegetation Index (RVI)	sr_cn	Cyan/NIR
Normalized Difference Vegetation Index (NDVI)	ndvi_on	(NIR − Orange)/(NIR + Orange)
Adjusted Vegetation Index (SAVI)	savi_on	(1.5 × (NIR − Orange))/(NIR + Orange + 0.5)

). A set of random points were selected inside Soybean leaves, and corresponding VI values were extracted. This was repeated for all images and stages. The correlation between index values and spectral reflectance values was analyzed. Based on these results, three options were proposed as new VIs for OCN cameras (

Table 4. New Vegetation Indices.

Symbology	Mathematical Expression
VINIR,O,C	$\left(\frac{\mathrm{NIR} - \mathrm{Orange} }{\mathrm{NIR} + \mathrm{Orange}}\right)\times \left(\frac{\mathrm{NIR}}{\mathrm{Cyan}}\right)$
VINIR,O,O	$\left(\frac{\mathrm{NIR} - \mathrm{Orange} }{\mathrm{NIR} + \mathrm{Orange}}\right)\times \left(\frac{\mathrm{NIR}}{\mathrm{Orange}}\right)$
VIC,O	$\left(\frac{\mathrm{Cyan}}{\mathrm{Orange}}\right)$

).

2.4. Leaf Area Index

LAI was calculated using an indirect method introduced in a study conducted in the North American region (Nebraska), which indicated successful results [36]. Gaso et al. [37] also effectively used LAI in their study of Soybean yield variability. We selected Equation (1) with a Root Mean Squared Error (RMSE) of 0.48 and a coefficient of determination (R2) of 0.90 to calculate the LAI of our Soybean plants.

$$\mathrm{LAI}={\log }_{0.37}\left({\mathrm{NDVI}}^{-0.526}-1.03\right)$$

(1)

NDVI was calculated using images from the RGN camera. Once LAI was calculated, values were extracted for the same set of random points (as used earlier for VIs). EDA was performed to remove outliers and understand the data distribution.

2.5. Proposed Indices and Their Validation

The new VIs were calculated for each stage; corresponding values were extracted for the same set of random points. EDA was completed, and outliers were removed. Since data were not normally distributed at all stages, the Kruskal-Wallis test [38] was used to determine if there were significant differences between the new VIs and LAI. After that, Pearson Correlation Coefficient (r) was calculated between the new VIs and LAI to evaluate how representative they are at different stages [39].

3. Results

3.1. Spectral Profiles

According to Figure 3 (R4 stage), there were outliers in the Cyan and Orange bands. The index plots are scatterplots with the spectral value of the points (y-axis) appearing from left to right in the order they are available in the data file. So the scatter plot shows spectral values from left to right according to each stage’s ascending order of date and time. However, data were not trended, and the degree of scatter was consistent from left to right (see scatter plots). NIR values were approximately normally distributed without outliers. The theoretical quantile plots were used to determine how these points deviated from a theoretical normal distribution and visually assess whether the difference was significant. Once outliers were removed, these deviations were insignificant for this stage. The spectral profiles are shown in the upper right graph. Grey lines represent data points, and the black line represents the average spectral profile for the R4 stage. The Orange band value was slightly lower than NIR values and higher than the Cyan value. However, NIR values were seen increasing at later dates. Figure 4 shows the R5 stage. All bands had outliers, and the NIR scatter plot showed a positive trend (data ranging from 0.6 to 1). Data points significantly deviated from theoretical normality (theoretical quantile plots and histograms). Once outliers were removed, the average spectral profile indicated dramatically increased NIR and low Cyan and Orange values. Almost the same spectral behaviors were evident in the R6 stage (Figure 5). The average spectral profile of R6 was similar to R5. According to Figure 6, the data for the R7 stage had few outliers and had an average spectral profile more similar to R4 than either R5 or R6, with a NIR value range of 0.2 to 0.8.

Table 5. OCN bands Mean and Standard Deviation among Reproductive Stages.

Band	R4		R5		R6		R7
	Mean	SD	Mean	SD	Mean	SD	Mean	SD
Cyan	0.08	0.06	0.08	0.04	0.06	0.03	0.04	0.04
Orange	0.28	0.09	0.13	0.05	0.09	0.04	0.24	0.14
NIR	0.38	0.22	0.91	0.12	0.88	0.13	0.42	0.20

shows the mean and Standard Deviation (SD) of band values at each stage. The Orange band got higher mean values in R4 (0.28) and R7 (0.24) stages with lower mean values in R5 (0.13) and R6 (0.09). NIR values were changed from 0.38, 0.91, and 0.88 to 0.42 for R4–R7, respectively (Table 5).

3.2. Proposed Indices

3.2.1. Correlation between Different Spectral Bands and VIs (OCN Camera)

Figure 7, Figure 8, Figure 9 and Figure 10 show the correlation between different vegetation indices and spectral bands of the OCN camera. Within the bands themselves, the scatterplots didn’t show linear correlations. However, in R5 and R6 stages, Cyan and Orange bands showed a statistically significant positive linear correlation (0.973) (Figure 8 and Figure 9). There is a non-linear, moderate negative correlation (−0.387) in R4 and R7. In the R4 stage, there is a non-linear, negative correlation between Orange and NIR bands (−0.206) (Figure 7) and a strong positive relation between Cyan and NIR bands. Additionally, the index “ndvi_on” presents statistically significant (negative) correlations between indices “sr_cn” and “sr_on”. These patterns are almost similar at the R7 stage (Figure 10). In R5 and R6 stages, there is a high correlation between “ndvi_on” and the indices “sr_cn” and “sr_on” correlation (Figure 8 and Figure 9).

Referring to Figure 7, Figure 8, Figure 9 and Figure 10, a relationship between different bands or two indices with a high degree of correlation can be explained with the following generic equation:

VI₁ α VI₂

(2)

where VI_1&2 vegetation indices are derived from the OCN camera using existing mathematical formulas.

According to Figure 7, Figure 8, Figure 9 and Figure 10 and Equation (2), there were no strong linear relationships between different bands or band combinations. Hence, the mathematical expressions proposed in Table 4 were polynomial (quadratic) equations. For example, Figure 7 shows quadratic relation with NIR for the other two bands. For example, there was a statistically negative relation (r = −0.991) between ndvi_on and sr_on, and it was the basis for the proposed ${\mathrm{VI}}_{\mathrm{NIR},\mathrm{O},\mathrm{C}}$ index.

3.2.2. Correlation between LAI and New VIs

The New VIs were tested over the growing season and assessed the reliability of them based on the LAI of Soybean plants. The Kruskal Wallis test results for LAI and different indices values were not statistically significant at the 95% confidence level (p-value > 0.05), indicating no significant difference between LAI values and new VIs at each stage.

The correlation between the new VIs and LAI is summarized in

Table 6. Pearson Correlation Coefficient (r) results between the New Vegetation Indexes and LAI for each stage.

VI	Stage
	R4	R5	R6	R7
VINIR,O,C	0.44	0.39	0.39	0.2
VINIR,O,O	0.29	0.37	0.38	−0.3
VIC,O	0.48	−0.27	0.28	−0.5

. VI_NIR,O,C showed a higher correlation for middle stages (R5 and R6), slightly better than VI_NIR,O,O. The index: VI_C,O showed a moderate positive correlation for R4 and a moderate negative correlation for R7 (Table 6). Figure 11 shows examples of the new VIs for each stage. Although R5 and R6 had a linear correlation between Cyan and Orange bands, the VI_C,O didn’t differentiate Soybean leaves and the background properly (Figure 11, rows 2 and 3). This is also evident in the correlation between LAI and indices (Table 6). However, VI_C,O showed a visually high level of details (leaves and background) for R4 and R7 than the other indices. VI_NIR,O,C in R5 and R6 showed the same, where more details can be appreciated within the leaves and a clear demarcation between leaves and the background.

4. Discussion

Photographs of each growing stage in Figure 3, Figure 4, Figure 5 and Figure 6 show Soybean maturity levels. For example, R4 had pre-matured plants (early stage), and R7 was in the late season showing yellowish plants. It is also important to note that images from R4 and R7 stages, where plants were sparsely distributed, have soil in the background which must be separated from the leaf area in the analysis. In R5 and R6, plants were fully grown, and no soil was visible in the camera’s field of view.

4.1. Spectral Profiles

According to the data analysis, it can be confirmed that the generated spectral profiles followed the standard healthy vegetation spectral profile pattern. For example, the average Cyan values were 0.08 (R4 and R5), changed from 0.08 to 0.06 (R5 to R6), and further reduced to 0.04 in R7 (Table 5). In contrast, Orange spectral values were reduced from R4 to R5 and further in R6 and then increased from R6 to R7. When plants are maturing, chlorophyll and carotenoid concentration will change, and thus the absorption near the red region varies [1,2,3]. Results showed a low NIR reflectance for R4 and R7, stages where leaf pigments (leaf chlorophyll and carotenoids) are expected to be lower than in other stages. High NIR reflectance in the middle stages (R5–R6) indicates fully grown Soybean with high leaf chlorophyll concentrations [1,2,3]. Hence, these results coincided with the already developed healthy vegetation spectral profile [15,16]. Although Kokaly et al. [14] stated the wavelength region of 400–700 nm is sensitive to changes in leaf pigments and thus the leaf chlorophyll content, the NIR wavelength region of the OCN camera (780–870 nm) is outside the specified ideal range. This study also indicated meaningful variations of NIR within growing stages. The NIR spectral values in the mid and later stages showed a positive trend during EDA until the latter part of the R7 stage, which indicates a negative trend. A study comparing actual leaf chlorophyll concentration and spectral values is recommended.

4.2. Proposed Indices

No linear relations existed between individual bands except Cyan and Orange bands in the mid-seasons (R5 and R6). This can be explained in terms of chlorophyll and carotenoid concentration variations over the growing season and their effect on sunlight absorption and reflection. For example, when plants mature, chlorophyll (greenness) increases and thus absorption in the near the red region and increasing reflectance in green (near Cyan) [15,16,40]. On the other hand, as specified in Gutman et al. [41] and other studies, NIR and red didn’t show a linear relation. As noted, R5 and R6 had similar patterns as did R4 and R7. A different level of exposure to the soil background might contribute here. The existing vegetation indices derived using the OCN camera, according to Table 3, were based on linear relations of different band combinations. Hence, the new indices were derived by analyzing the correlation between those indices. For example, “ndvi_on” and “sr_cn” had statistically significant high negative relation (−0.9), and it was the basis for VI_NIR,O,C index.

Visual analysis of images derived from the new indices illustrates different levels of detail in the vegetation depending on the index and growth stage (Figure 11). This is compatible with the correlation coefficients obtained in Table 6. For instance, VI_NIR,O,C showed good results for R5 and R6, which is clearly shown in Figure 11. Since there was a similar correlation between Cyan and Orange bands in R4 and R7, the new index: VI_C,O showed good results. Although there was a strong linear relationship between Cyan and Orange at R5 and R6 stages, the ratio of these two bands didn’t differentiate Soybean leaves well. That indicates their low reflectance values in these stages. Once these two bands were combined with the NIR band using a polynomial equation, the results were acceptable (Table 6).

LAI is a key biophysical variable used to measure crop growth and productivity [11]. Hence, we used LAI to assess whether the details of the Soybean plot represented by the proposed indices were the same as the distribution of LAI. The statistical analysis Kruskal-Wallis test didn’t show a significant difference between index values and LAI.

Due to time limitations, we were not able to test the spectral variations over small time spans, so the data, which was collected multiple times each day were, aggregated into growing stages. It would be beneficial to analyze the same set of data as time series to see the spectral variations over time. In the future, it would recommend comparing the vegetation indices to the leaf chlorophyll content to assess the performance of the proposed indices. Additionally, testing the proposed indices over another growing season or including a lab chemical analysis, perhaps for nitrogen variations, would represent an important accuracy assessment.

4.3. Practical Applications of Results

The study developed spectral profiles and VIs for different Soybean growth stages. For example, in a practical situation, field spectral profiles can be compared with this study’s generated spectral profile and make informed decisions regarding the Soybean growth patterns, even at different reproductive stages. The proposed VIs are useful, especially in analyzing growth dynamics over a large area; for instance, this camera can be attached to a Remotely Piloted Aircraft System (RPAS or drone), capture images and calculate the proposed VIs for large Soybean fields to assess the growth patterns over time.

5. Conclusions

Using remote sensing technology, we developed new vegetation indices (VIs) from the Mapir Survey 3W OCN camera that includes the Cyan, Orange, and NIR bands from the electromagnetic spectrum on two adjustable stations attached to Raspberry Pi units and solar panels. This approach provides a cost-effective and efficient method for precise crop management. We analyzed spectral information from four Soybean reproductive stages (R4–R7) at LUARS, conducted EDA, and removed outliers. We then created and analyzed spectral profiles for each stage.

Our analysis revealed that the spectral profiles obtained for each stage were consistent with what is expected from healthy vegetation spectral profile studies. The Cyan values ranged from 0.08 to 0.09 (R4 to R5) and 0.05 to 0.09 (R5 to R6) and remained the same in R7. The Orange spectral values decreased from R4 to R5 and increased from R6 to R7, while the NIR values were higher in the middle stages and lower in R4 and R7 stages, where chlorophyll is expected to be in a lower concentration. We tested correlations between each spectral band at different stages and found a positive linear relation between Cyan and Orange reflectance in R5 and R6 stages, while all other combinations were non-linear.

We then tested various existing VIs that utilized similar band combinations and found that combining Cyan and Orange with NIR in a polynomial equation delivered strong correlations. Based on these findings, we proposed three VIs: VI_NIR,O,C, VI_NIR,O,O, and VI_C,O. We correlated these VIs against LAI values and obtained results of 0.39 for R4 and R5, 0.48 for R6, and −0.5 for R7 (Table 6). Our analysis suggests that the best-fitted VI for middle stages is VI_NIR,O,C = ((NIR − Orange)/(NIR + Orange)) × (NIR/Cyan), while the best-fitted VI for R4 and R7 is VIo,c = Cyan/Orange.

Due to time limitations, we could not test the spectral variations over small periods. Therefore, the data, which was collected multiple times each day, were aggregated into growing stages. Analyzing the same data set as time series would be beneficial to appreciate the spectral variations over time. It is also recommended to compare the vegetation indices with the leaf chlorophyll content to assess the performance of the proposed indices. For further research, it is advisable to conduct testing in another growing season and perform laboratory chemical analyses such as nitrogen content.

In conclusion, our study demonstrates the utility of remote sensing technology in developing a new VI for Soybean crop management. Our findings provide insights into the spectral profiles of Soybeans at different reproductive stages and in different wavelengths, offering new approaches for optimizing crop management.