Analysis of Crop Spectral Reflectance at the Croplands in Eastern Kazakhstan Using Satellite Imagery

DOI: https://doi.org/10.52939/ijg.v19i11.2923

1. Introduction

Remote sensing technology, especially satellite imagery, has revolutionized monitoring, managing, and predicting crop performance, leading to a more sustainable and productive agricultural sector. A vital component of this techno-logical revolution is the study of crop reflectance, a non-invasive and highly informative method to assess crop health and predict yields [1] and [2]. Crop reflectance refers to how crops absorb, transmit, and reflect light across different wavelengths. By analyzing the light reflected from crops, scientists can infer valuable information about the plant’s health, growth stage, and productivity. This light can range from the visible spectrum to near and far infrared, with different crops and conditions altering the specific reflectance patterns [3] and [4].

In the visible light range, healthy vegetation typically absorbs most blue and red light due to chlorophyll, essential for photosynthesis. This light absorption causes healthy plants to reflect more green light, making them appear green to the human eye. On the other hand, in the near-infrared region, healthy plants reflect a lot of light, a characteristic known as the ’red edge.’ This high reflectance is due to the plant’s cellular structure, which scatters light in the near- infrared [5]. The Normalized Difference Vegetation Index (NDVI) is a popular measure used in remote sensing. It captures the difference between near-infrared (which vegetation strongly reflects) and red light (which vegetation absorbs) to assess plant health and vigor [6]. Tasseled Cap Transformation (TCT) is another technique developed explicitly for Landsat data, yielding three components- brightness, greenness, and wetness-representing soil brightness, vegetation cover, and moisture content, respectively [7].

In remote sensing, PCA is often used for data reduction when dealing with multi-band images. It is a statistical procedure that transforms a set of observations of possibly correlated variables into values of linearly uncorrelated variables called principal components [8]. Each principal component explains a certain percentage of the total variation in the dataset, with the first few components typically explaining the majority of the variation. PCA can help reduce the complexity of multi-dimensional data, making it easier to interpret spectral reflectance patterns across different crops and growth stages [9].

Remote sensing data can also be used to predict crop yields. By correlating historical yield data with spectral characteristics of the crops at different growth stages, machine learning algorithms can be trained to predict future yields based on remote sensing data [10]. Time-series analysis of remote sensing data can monitor the phenological stages of crops, providing crucial information about the timing of key growth stages, such as flowering or maturity [11].

By measuring and analyzing these reflectance patterns, remote sensing applications can detect subtle changes in plant health long before they become visible to the human eye [12]. For instance, a stressed plant might reflect more light in the visible spectrum and less in the near-infrared, indicating potential issues such as disease, nutrient deficiency, or drought stress [13]. This capability to ’see’ the unseeable makes crop reflectance studies using remote sensing a powerful tool in modern agriculture. By enabling early detection of plant stress, it allows for timely intervention, ultimately leading to better crop management and improved yields [14]. Furthermore, farmers and policymakers can make informed decisions about crop marketing and food supply logistics by predicting crop yields based on reflectance data, contributing to a more efficient and sustainable agricultural system.

Combined with data processing and machine learning advances, these methods hold significant potential for enhancing crop management and increasing agricultural productivity [15]. While this technology holds immense promise, it is essential to consider local crop types, growth stages, and environ- mental conditions when interpreting reflectance data. As such, studies like the one on wheat and barley at the OHMK farm in Eastern Kazakhstan are crucial in expanding our understanding of spectral reflectance across different crops and environments.

2. Methods

The research area is located in the Eastern Kazakhstan province north of Ust-Kamenogorsk city and is a farmland area with various crops (Figure 1). The relief of the study area refers to the variations in elevation across the landscape. The statistical analysis reveals essential characteristics of the relief. The maximum elevation of 612 meters indicates the highest point in the area, while the minimum elevation of 296 meters represents the lowest point. The mean elevation of 400.1 meters provides an average value, representing the typical height of the terrain. The standard deviation of 45.5 meters indicates the degree of variation or spread of the elevation values around the mean. We created the dataset using the Google Earth Engine web service [16]. It included elevation [17], soil bulk density [18], rainfall data [19], spectral reflectance data captured by Landsat-8 [20] on 21.06.2022, 08.08.2022, 24.08.2022, also Land use data with exact crop species information collected on August 9th, 23rd, and October 20th, 2022 by our group under this Research grant (Table 1). This table provides an overview of the input dataset details used in the study. It includes information on the type of data, the product/source name, and specific details such as spatial resolution and layers.

Figure 1: The study area

Table 1: Input dataset details

Figure 2: The research design

Table 2: The tasseled cap transformation coefficients for Landsat-8 [21]

The research workflow included data collection and preprocessing (the data clipping, NDVI calculation, and the Tasseled Cap Transformation), the PCA transformation, the data correlation, and the crop mapping (Figure 2). Tasseled Cap Transformation (TCT) and Principal Component Analysis (PCA) are widely used remote sensing techniques for image interpretation and data reduction. Tasseled Cap Transformation is a linear transformation developed for Landsat satellite data (Table 2). The transformation involves converting the original satellite bands into new components that are easier to interpret regarding physical vegetation characteristics. The three primary TCT components are brightness, greenness, and wet- ness, representing soil brightness, vegetation cover, and moisture content. In the context of crop reflectance, TCT can provide valuable insights into the crop’s health and growth stages. Principal Component Analysis (PCA) was applied [22], and it consisted of the following steps:

      1. The dataset was standardized.

      2. The covariance matrix C of the data was computed using equation 1.

Equation 1

where X is the data matrix, X^T represents the transpose of matrix X, and n is the number of data points.

      3. The eigenvalues and eigenvectors of the covariance matrix were computed.

      4. The eigenvalues and corresponding eigenvectors in decreasing order were sorted.

After the Tasseled cap transformation, correlation (Pearson’s coefficient) and statistical significance (p-values) for all data were tested. In the context of the study on wheat and barley at the OHMK farm in Eastern Kazakhstan, TCT and PCA are valuable tools. TCT helps interpret the spectral reflectance data regarding physical crop characteristics. At the same time, PCA simplifies the multi-dimensional reflectance data, making it easier to identify key patterns and differences between the crops. At the final stage, multiple linear regression equations were formulated to map the selected crops in the study area.

3. Results and Discussion

The p-values for each crop and parameter combination indicate the statistical significance of the difference between the sample mean and the null hypothesis mean (zero). A p-value less than the significance level (e.g., 0.05) indicates that the difference is statistically significant. Based on the provided p-values, here are some conclusions:

For Spring wheat, most parameters (Red, Green, Blue , NIR, SWIR1, SWIR2, Brightness, Greenness , TCT4, TCT5, TCT6, NDVI, PC1 , and PC2) have extremely low p-values close to zero, indicating a significant difference from the null hypothesis mean of zero. Wetness also has a very low p-value, suggesting a significant difference.

For Barley, like Spring wheat, most parameters have extremely low p-values close to zero, indicating a significant difference from the null hypothesis mean of zero. For Sainfoin, most parameters have very low p-values close to zero, indicating a significant difference. PC1 has a higher p-value (1.34), suggesting less statistical significance than other parameters.

For Alfalfa (2nd year), most parameters have p-values equal to zero, indicating a significant difference from the null hypothesis mean of zero. PC1 has a higher p-value (0.85), suggesting less statistical significance than other parameters. For Sunflower and Soybean, like Spring wheat and Barley, most parameters have extremely low p-values close to zero, indicating a significant difference from the null hypothesis mean of zero.

These results suggest that for most parameters, there is a significant difference in the mean values between the crops, indicating potential discriminative power in distinguishing the crops based on those parameters. However, interpreting the results should consider other factors, such as the data distribution, sample size, and specific research objectives. Due to a low statistical significance, results do not include elevation, Slope, Soil density, and Rainfall data.

After that, Pearson’s correlation tests were applied to the research dataset, and the results demonstrated specific strong values according to the Table 3. Observed correlations between PC1, PC2, and the listed parameters for each crop demonstrated positive and negative values described below:

      a) Spring wheat [23]:

            PC1 shows high positive correlations with Greenness data (TCT), Green, Blue, and SWIR2 bands, while it has a strong negative correlation with NDVI and TCT4. PC2 has no significant correlations.

      b) Barley [24]:

            PC1 exhibits significant positive correlations with Red, Green, and SWIR2 bands and strong negative with NDVI. PC2 shows negative correlations with SWIR1 and SWIR2.

      c) Sainfoin [25]:

            PC1 positively correlates with Greenness data, Green, SWIR2, Red bands, and NDVI. PC2 shows negative correlations with the Red, SWIR1, and SWIR2 bands.

      d) Alfalfa [26] (2nd year):

            PC1 is positively correlated with TCT6 and negatively with SWIR2. PC2 positively correlates with NIR, Greenness and has a moderate correlation with NDVI.

      e)Sunflower [27]:

            PC1 exhibits positive correlations with SWIR1, SWIR2, and NIR, strongly correlates negatively with NDVI. PC2 positively correlates with SWIR2, SWIR1, and NIR and has a moderate negative correlation with NDVI.

      f) Soybean [28]:

            PC1 shows a negative correlation with Greenness data and NDVI and positive correlations with SWIR1, SWIR1, NIR, and Red. PC2 does not show any significant correlations with the listed parameters.

These correlations provide insights into the relationships between the principal components (PC1, PC2) and the corresponding parameters for each crop. We plotted the strongest values to visualize some of the described correlations for crops (Figure 3). This figure demonstrates the negative correlation values of vegetation-related information (NDVI) for Barley, Sainfoin, Spring wheat, Soybean, and Sunflower. At the same time, Alfalfa showed high positive values between principal component 2 and NIR reflectance band.

Table 3: Correlation values for PC1 and PC2

Figure 3: The correlation between principal components and vegetation-related data (a) barley sainfoin and spring wheat (b) alfalfa (c) soya and sunflower

The next step of this research included formulating multiple linear equations based on the observed correlations. Therefore, the principal components (PC1, PC2) are considered dependent parameters, while others are independent as illustrates in equations 2 to 5:

Spring wheat:

PC1 =157.3Greeness − 15.4Blue + 10.6SWIR2 − 22.4Green − 9.7TCT 4 −10.9 NDVI + 7.6

Equation 2

This regression equation represents the relationship between PC1 and the Spring wheat crop’s input features (Greeness, Green, Blue, SWIR2, TCT4, and NDVI). The coefficients indicate the magnitude and direction of the impact of each input feature on PC1.

Barley:

PC1 = −206.2Red + 177.73Green + 12.85WIR2 − 103.37NDVI + 86.12

Equation 3

This equation represents the relationship between PC1 and the Barley crop’s input features (Red, Green, SWIR2, and NDVI).

The coefficients indicate how changes in each input feature influence PC1.

Sainfoin:

PC1 = 198.33Red + 62.31Green + 51.11SWIR2 − 39.33Greenness + 35.53NDVI − 32.64

Equation 4

This equation represents the relationship between PC1 and the input features (Red, Green, SWIR2, Greenness, and NDVI) for the Sainfoin crop. The coefficients indicate the strength and direction of the influence of each input feature on PC1.

Alfalfa:

PC2 = 105.02NIR − 82.63Greenness + 19.48NDVI − 30.6

Equation 5

This equation represents the relationship between PC2 and the Alfalfa crop’s input features (NIR, Greenness, and NDVI). The coefficients indicate the impact of each input feature on PC2are presented in equation 6 and 7.

Sunflower:

PC1 = −15.27NIR + 57.16SWIR1 − 11.87SWIR2 − 15.25NDVI + 7.0

Equation 6

This equation represents the relationship between PC1 and the Sunflower crop’s input features (NIR, SWIR1, SWIR2, and NDVI). The coefficients indicate the strength and direction of the influence of each input feature on PC1.

Soybean:

PC1 = 97.38 Red − 17.74NIR + 12.24SWIR1 + 20.61SWIR2 + 2.24NDVI − 5.25

Equation 7

This equation represents the relationship between PC1 and the input features (Red, NIR, SWIR1, SWIR2, and NDVI) for the Soybean crop. The coefficients indicate the impact of each input feature on PC1. These regression equations provide a mathematical representation of the relationship between the principal component PC1 and the input features for each crop. Using the coefficients, we estimated the PC1 and PC2 based on the given input features using input raster data. We derived the principal components map that mainly exposes vegetation growth in the study area croplands (Figure 4). The description of the PCA values for each crop is provided in Table 4, which includes the maximum, mean, minimum, and standard deviation of the PCA values for each crop. It overviews the distribution and variation in PCA values across the crops. These values provide insights into each crop’s distribution and variation of PCA values. The maximum PCA value represents the highest level of variability, while the mean PCA value indicates the average level of variability. The standard deviation shows the spread of PCA values around the mean. Comparing the crops, Sunflower exhibits the highest maximum PCA value, indicating more significant variability compared to other crops. At the same time, Barley shows the lowest mean and standard deviation, suggesting relatively lower variability.

Figure 4: The map of the PC values distribution for crops in the study area

Table 4: PCA statistics for each crop

We can gain insights into spectral characteristics and variability by comparing these statistical measures across different crops. For example, crops with higher maximum and mean PCA values exhibit more significant spectral variability, indicating diverse and distinct spectral signatures. Conversely, crops with lower maximum and mean PCA values have less spectral variability and may exhibit more uniform spectral characteristics. The standard deviation provides a measure of the variation within each crop, reflecting the degree of heterogeneity or similarity in the spectral properties of the crop pixels. The given PCA analysis, including the NDVI band, suggests that the analysis considered the spectral information related to vegetation health and density. The statistics provided for each crop, such as the maximum, mean, minimum, and standard deviation of PCA values, indicate the variability of NDVI values within each crop.

By examining the statistics of NDVI within each crop, we can infer information about the vegetation health and density characteristics specific to that crop. For example, a higher maximum NDVI value suggests dense and healthy vegetation areas. In contrast, a lower mean or standard deviation of NDVI values may indicate less variability in vegetation health within the crop area. It is important to note that NDVI alone cannot expose crop yield potential [29]. Other factors, such as soil conditions, weather, management practices, and pest/disease pressures, influence crop productivity. However, NDVI is a valuable tool for monitoring vegetation health and can be used to indicate crop performance in conjunction with other agronomic data.

4. Conclusions

The main findings of the study can be summarized as follows:

      A. The principal component analysis (PCA) revealed strong correlations between specific spectral bands and each crop’s principal components ( PC1 and PC2). These correlations indicate the importance of specific bands in capturing the variability and characteristics of the crops.

      B. Each crop showed distinct correlations between the spectral bands and the principal components. These correlations provide insights into each crop’s unique spectral signatures and vegetation characteristics, such as greenness, near-infrared reflectance, and other spectral properties.

      C. The equations derived from the PCA analysis provide a mathematical relationship between the spectral bands and the principal components for each crop. These equations can be used to estimate the values of the principal compo- nents based on spectral information, allowing for a better understanding of the crop’s characteristics and variability.

      D. The analysis of PCA values for each crop revealed variations in the principal components’ maximum, mean, min- imum, and standard deviation. These variations indicate differences in the spectral response and variability of the crops, which can be related to vegetation health, density, and productivity.

The implications of these findings are:

      1. The study highlights the importance of spectral bands and their correlations with principal components in under- standing the characteristics and variability of different crops. This information can be used to develop crop-specific remote sensing models and monitoring techniques for accurate crop assessment and management.

      2. The derived equations provide a means to estimate the principal components based on spectral information. This can be useful for crop monitoring and assessment, allowing for a more comprehensive understanding of crop growth, health, and productivity.

      3. The variations in PCA values among crops suggest that each crop has unique spectral signatures and responses. This information can differentiate between crops, monitor their growth stages, identify stress conditions, and assess crop yield potential.

      4. The study underscores the value of remote sensing and spectral analysis in crop monitoring and precision agriculture. By integrating spectral information with other agronomic data, such as soil conditions and weather, more informed decisions can be made regarding crop management practices, resource allocation, and yield optimization. Overall, the findings of this study contribute to understanding crop characteristics, variability, and monitoring using re- mote sensing data. They have implications for improving crop management practices, optimizing resource allocation, and enhancing crop productivity in agriculture.