Rapid Analysis of Soil Organic Carbon in Agricultural Lands: Potential of Integrated Image Processing and Infrared Spectroscopy

Abstract

The significance of soil in the agricultural industry is profound, with healthy soil representing an important role in ensuring food security. In addition, soil is the largest terrestrial carbon sink on earth. The soil carbon pool is composed of both inorganic and organic forms. The equilibrium of the soil carbon pool directly impacts the carbon cycle via all of the other processes on the planet. With the development of agricultural systems from traditional to conventional ones, and with the current era of precision agriculture, which involves making decisions based on information, the importance of understanding soil is becoming increasingly clear. The control of microenvironment conditions and soil fertility represents a key factor in achieving higher productivity in these systems. Furthermore, agriculture represents a significant contributor to carbon emissions, a topic that has become timely given the necessity for carbon neutrality. In addition to these concerns, updating soil-related data, including information on macro and micronutrient conditions, is important. Carbon represents one of the major nutrients for crops and plays a key role in the retention and release of other nutrients and the management of soil physical properties. Despite the importance of carbon, existing analytical methods are complex and expensive. This discourages frequent analyses, which results in a lack of soil carbon-related data for agricultural fields. From this perspective, in situ soil organic carbon (SOC) analysis can provide timely management information for calibrating fertilizer applications based on the soil–carbon relationship to increase soil productivity. In addition, the available data need frequent updates due to rapid changes in ecosystem services and the use of extensive fertilizers and pesticides. Despite the importance of this topic, few studies have investigated the potential of image analysis based on image processing and spectral data recording. The use of spectroscopy and visual color matching to develop SOC predictions has been considered, and the use of spectroscopic instruments has led to increased precision. Our extensive literature review shows that color models, especially Munsell color charts, are better for qualitative purposes and that Cartesian-type color models are appropriate for quantification. Even for the color model, spectroscopy data could be used, and these data have the potential to improve the precision of measurements. On the other hand, mid-infrared radiation (MIR) and near-infrared radiation (NIR) diffuse reflection has been reported to have a greater ability to predict SOC. Finally, this article reports the availability of inexpensive portable instruments that can enable the development of in situ SOC analysis from reflection and emission information with the integration of images and spectroscopy. This integration refers to machine learning algorithms with a reflection-oriented spectrophotometer and emission-based thermal images which have the potential to predict SOC without the need for expensive instruments and are easy to use in farm applications.

1. Introduction

Soil is the largest terrestrial carbon sink on earth. The total amount of carbon in the top 100 cm of soil (excluding the carbon in the top layer of dead leaves and charcoal) is between 2157 and 2293 billion tons worldwide. Specifically, the carbon from organic matter in the top 30 cm of soil is estimated to be between 684 and 724 billion metric tons [1,2]. Changes in land use, deforestation, and climate change can affect the amount of organic carbon in the top layers of soil, but this impact is less significant for soil carbonate carbon. Approximately 695 to 748 billion metric tons of carbonate carbon are in the top 100 cm of soil globally [3,4]. Humans depend on soil to produce food and fiber, and unsustainable management of this nonrenewable resource has accelerated its degradation [5].

Soil carbon is one of the key elements of the carbon cycle. This soil carbon pool is a mixture of organic and inorganic components that interact with other living and nonliving elements. This carbon plays a crucial role in defining soil properties that support soil productivity, which is also important for farming. While long-term storage of organic carbon is preferable in terms of greenhouse gas mitigation, labile fractions of soil organic carbon (e.g., with residence times of months to years) are essential in terms of soil fertility (their mineralization provides nutrients to plants), soil physical conditions (aggregate stability largely depends on labile carbon), and soil biodiversity. Hence, it is desirable to increase the stocks of both labile and stable forms of organic matter [6].

Soil organic matter refers to all decomposed, partially decomposed, and undecomposed organic matter of plant and animal origin. Soil organic matter is a primary indicator of soil health and is estimated by measuring SOC. Soil organic carbon contributes to the cation exchange capacity (CEC), and the CEC provides exchange sites for calcium (Ca), magnesium (Mg), and potassium (K). Furthermore, the CEC provides binding sites for many other organochemical compounds present in agrochemicals and reduces their ability to contaminate groundwater [7].

Among the different forms of carbon present in soil, soil organic carbon (SOC) has been studied for its ability to predict soil quality. This is because SOC is one of the determining factors of other soil attributes, such as the water holding capacity, nutrient holding capacity, soil resilience, and facilitation of microbial activity. The presence of carbon also has a significant effect on soil color. Understanding the SOC value can aid in determining site-specific organic applications for managing soil productivity to ensure sustainability. Land use practices play a crucial role in SOC availability. It is commonly accepted that the SOC thresholds for maintaining soil quality should be above 2%, below which deterioration may occur. Some scientists have stated that SOC can be conceptualized as a “universal keystone indicator” within the realm of soil fertility management. Land use management has a direct influence on the quantity of SOC, and a detailed description of various land use practices in the present and past is similarly crucial for a comprehensive understanding of SOC dynamics [8].

Better management practices can maintain SOC, and appropriate soil data can contribute to this decision-making [9]. Common organic materials include animal-related manure, plant residue, and compost. Soils with low carbon (C): nitrogen (N) ratios favor decomposition. Management approaches that involve the use of materials for mulching or soil erosion control disrupt SOC pools. Thus, the choice of land use management practice is a prime source of SOC loss, especially for tropical soils. Conventional tillage, biomass harvesting, burning, and using excess fertilizers, herbicides, and pesticides all threaten efforts to sustain critical SOC levels [8]. SOC data can be used for controlling tillage levels, recommending irrigation intervals, and providing site-specific fertilizer recommendations, including organic amendment applications. The mining of SOC from soil for the purpose of nutrients through organic matter decomposition influences the atmosphere in a manner similar to that of fossil fuel combustion. It is therefore crucial to link carbon and hydrological cycles through the conservation of water resources to improve agronomic yields. The low SOC stocks in rain-fed farms can be increased by increasing water harvesting. Soil C sequestration is a bridge across three global issues, namely, climate change, desertification, and biodiversity. The increase in SOC stocks increases the soil’s capacity to oxidize methane [1].

Continuous harvesting practices and a lack of organic amendments have resulted in a depletion of organic matter in cultivated soils, which reduces SOC. The rates of decline vary depending on the type of soil. For instance, sandy soils experience average annual losses as high as 4.7%, while sandy loam soils have lower losses, averaging at 2.0% [10]. SOC has been in decline due to the high intensity of decomposition and high turnover rates in tropical farmlands [8].

Research conducted on organic fertilizer has identified that organic amendments significantly increase the labile carbon fraction. However, the lower labile fraction showed no saturation behavior; therefore, it was suggested that focused attention be given to the stable carbon fraction in rice fields because of the greater potential for SOC sequestration in rice fields [11]. The application of additional organic residues can result in acidification due to nitrification. Furthermore, the application of an additional amount of carbon can also result in the formation of a crust on the soil surface, a reduction in the hydraulic conductivity, and an increase in runoff [8]. Although a portion of SOC is present in water and air, a considerable quantity can be released into the atmosphere. This can occur through two main processes: mineralization, which results in the release of CO₂, and methanogenesis, which releases CH₄. The amount of SOC that is buried is estimated to range from 0.4 to 0.6 Gt per year, while the emissions of CO₂ and CH₄ are estimated to range from 0.8 to 1.2 Gt per year [1]. Soil carbon pools have their own dynamics and are constantly changing through natural processes (Figure 1).

Globally, soils lose between 55 and 90 Pg of carbon due to the conversion from natural to agricultural soils. Furthermore, practices such as tillage and soil degradation, which are caused by erosion and other processes that emit carbon such as CO₂ or CH₄, have also contributed to this phenomenon [12]. For instance, the conversion from till to no-till farming reduces carbon loss by 30 to 35 kg per ha per season [1].

An increase in yield can be observed with increases in soil organic carbon (SOC) until no further increase above the mean optimum soil carbon is reached. The increase in yield is 43.2–43.9 g kg⁻¹ for maize, 12.7–13.4 g kg⁻¹ for wheat, and 31.2–32.4 g kg⁻¹ for rice. By increasing SOC beyond current technology to an optimum level, the global production of three staple food crops is estimated to increase by 4.3% (sufficient to feed 640 million people). However, available technologies are only capable of achieving a 0.7% increase. The relationships between SOC and improved soil health and increased yields are significant drivers of carbon sequestration adoption [13].

There is an optimal level of soil organic carbon (SOC) below which leads to soil degradation and productivity loss. Beyond this threshold, environmental degradation can occur. The SOC threshold for sustaining soil quality is widely suggested to be approximately 2%, below which deterioration may occur [8]. Although the application of organic material to increase SOC is a good practice, the addition of excess inputs can lead to soil nutrient imbalances and a waste of resources (money, time, and labor). Therefore, it is necessary to analyze SOC before deciding on the organic fertilizer application to ensure the efficiency of the practice and soil sustainability. In addition, when we consider making decisions over a larger scale of land, it is necessary to have a consistent dataset that is able to provide reliable information on SOC to optimize the monitoring and mapping capacity [14].

SOC is also associated with interactions with fertilizers, pesticides, and waste materials. Many herbicide labels note the need to vary application rates based on SOC availability. Therefore, knowing the SOC content can aid in several practices in agriculture, particularly in precision agriculture [15].

This paper aims to develop a conceptual methodology for in situ SOC analysis that incorporates machine learning, image processing, and spectroscopy (Figure 2).

2. Widely Used Existing Methods for SOC Analysis and Their Limitations

The level of organic matter available in soil is one of the key factors used to measure soil health and is derived from SOC analysis. There are two widely used methods to analyze SOC: the Walkley and Black method and elemental analysis.

2.1. Walkley and Black Method

Here, the oxidizable organic carbon content is calculated from the amount of chromic (Cr³⁺) ions formed via titration, and this is also called the wet oxidation method. However, chloride ions in saline soils can interfere with the results. Additionally, ferrous iron and higher oxides of Mn can also undergo oxidation, resulting in inaccuracies in the analysis. Reagents, analytical instruments, and the necessary skills and knowledge, are required to conduct the analysis. These analyses are done in the laboratory by skilled technicians, and toxic byproducts are generated, including sulfuric and dry chromate [7].

2.2. Elemental Analysis

According to Tiessen and Moir (1993) mentioned in Schumacher [16], this method involves dry combustion at high temperatures in a furnace with the collection and detection of evolved CO₂. Then, acids are used to remove carbonate carbon (inorganic carbon) to improve the accuracy of the total organic carbon estimation.

2.3. Limitations

These destructive methods require the availability of laboratory facilities with skilled technicians, which narrows the scope of who can make these measurements and the number of measurements that can be made. Furthermore, these analyses take more than five days to obtain results for one sample. Therefore, these methods are not suitable for automated or in situ analysis and are difficult to use in spatiotemporal monitoring of SOC [14,17].

Elemental analysis requires sophisticated instruments available in laboratories; therefore, the cost per unit analysis is high. The Walkley and Black method, also known as the wet method, is less costly but generates toxic byproducts, including sulfuric and dry chromate [4,7]. A summary of the limitations and applications of these two methods is shown in

Table 1. Advantages, limitations, and references of widely used SOC analysis methods.

Methodology	Limitations of the Method	Advantages	Reference
Walkley Black	Analyzes only SOC; destructive method; produces toxic waste; requires analytical skills; requires access to laboratory instruments; takes 3 to 5 days to complete analysis; chloride, ferrous iron, and higher oxides of manganese can cause interference.	Analysis of all types of soil organic carbon. Higher accuracy.	[7,14,17]
Elemental analysis	Destructive method; high cost; requires analytical skills; requires access to laboratory instruments; takes 3 to 5 days to complete analysis.	Analysis of all types of soil carbon. Higher accuracy.	[14,17]

.

These limitations restrict practitioners from conducting SOC analysis, and there is a lack of updated information in soil directories. Qualitative comments are made based on visual properties, land use practices, and soil taxonomy when necessary. Furthermore, spatial data and mapping are employed to predict the availability of SOC. However, these methods may not be optimal for land use practices such as agriculture, where soil is subjected to intensive management practices [18,19]. Therefore, it is necessary to find alternative methods for monitoring SOC at the farm level that can bring results within a short time via simple equipment.

3. Potential Replacement Methods for Existing Methods

Few studies have been conducted in the field of identifying nondestructive, environmentally friendly, and reliable SOC analysis methods. These can be explained by two main principles: color-based models and spectroscopy-based models. However, there are color-based models supported by spectroscopic findings as well.

3.1. Color-Based Models

In the context of image processing in soil analysis, analyses based on soil color are of particular importance. Color is a property of a material or object that represents its unique characteristics. This color can be represented formally and numerically within a mathematical formula, which can then be used in device storage and analysis. These mathematical systems are referred to as models [19]. Soil color-based SOC determination can be a cost-effective and time-saving method [20]. Soil color is a continuous variable that varies in the X, Y, and Z spatial dimensions, varies with depth and throughout the landscape, and is expressed qualitatively using the Munsell soil color chart (). The abbreviation for the prominent color (Y = yellow, YR = yellow red, etc.) is used as a hue; this can be one or two letters. Within each range, the color becomes more prominent as the chroma number increases. The value ranges from absolute black (0) to absolute white (10) and consists of 443 color chips [20,21]. Unfortunately, soil color assessment using the is subjective or varies with the user. Soil scientists use soil color as an important property in soil classification. Additionally, vertical changes in soil color are used to identify different horizons and provide indirect measures of important soil characteristics, including drainage, aeration, organic matter content, and general fertility. In general, dark colors indicate a high level of decomposed organic matter known as humas [21].

In addition to the MSCC, there are several other color space models, namely, RGB, decorrelated RGB (DRGB), Commission International de l’Eclairage (CIE) XYZ, CIE Yxy, CIEL*a*b*, CIEL*u*v*, CIEHLC, Helmoltz chromaticity coordinates, and CMY (K), which overcome some of the restrictions of the MSCC. In the RGB system, color is produced by summing or subtracting the spectra of three primary colors: red (R), green (G), and blue (B). According to Wyszecki and Stiles (as cited in [22]), in the CIE XYZ standardized color space, by specifying the light source, and in the XYZ color space, Y represents the brightness (or luminance) of the color, while X and Z are virtual components of the primary spectra. In the CIE Yxy color space, the chromaticity coordinates x and y are independent of luminance (Y), where x specifies the color variation from blue to red and y specifies the color variation from blue to green. Both the XYZ and Yxy systems are perceptually nonlinear. To overcome the difficulties of determining xy chromaticity coordinates, Helmholtz chromaticity coordinates were introduced, which is the first uniform color space derived from CIE XYZ and describes the dominant wavelength, purity of extraction, and luminance. CIEL*a*b* and CIE L*u*v* are approximately uniform color systems. In both systems, L, the metric lightness function, ranges from 0 (black) to 100 (white). In CIEL*a*b*, a positive a* represents red, a negative a* represents green, and b* ranges from blue to yellow as the value increases from negative. In the CIE L*u*v* model, u* and v* are the chromaticity coordinates. CMY (K) is a subtractive model based on complementary colors (cyan, magenta, yellow, and black) generally used for output devices [20,22]. If we consider one sample and the MSCC value of that sample, based on that, we can find the dimensions of other color models (Figure 3).

In color-based soil analysis, color-based models are used to provide a value that can be used as a proxy to develop a relationship with the SOC content. Another method for recording emissions or absorption based on soil properties is spectral analysis. Correlating these spectral data or color data requires multivariate statistical methods, also known as chemometrics. When the relationships are linear, methods such as partial least squares can be used, but when they are nonlinear, machine learning algorithms are increasingly used [14]. According to research on color-based SOC detection and comparisons (

Table 2. Advantages, limitations, and references of different color models used to predict SOC.

Methodology	Limitations of the Method	Advantages	Reference
Munsell soil color chart (MCC)	Requires technical know-how of MCC use. Requires knowledge on the soil color and SOC relationship (principles of soil science). Readings are subjective or vary with the user. Results are qualitative. Scaling of the Munsell value and chroma coordinates within the color space is not perceptually the same.	Useful to compare different soils with significant SOC changes. Useful in rapid in situ assessments.	[20,22]
Munsell soil color (Munsell HVC) model	Restricted to descriptive purposes. Produces a weak relationship not suitable for qualitative measurements.	Useful to compare SOC changes over time. Compatible freely available software can be used with smart devices. Have potential to avoid human error.	[20,22]
CIE color model	The CIExy system gives a greater bias. Accuracy is low when Munsell HVC data and algorithms are used for transformations between these color space models.	Useful to predict SOC in soils after calibration. When reflectance spectra are used to color designate the system, the accuracy is high.	[22]
RGB color model	Each model is developed using Munsell HVC data and algorithms for transformations between these color space models.	Qualitative measurements with chronometer are possible.	[4]

), we can find the best model to use, considering the application and resources.

In addition to color model-based SOC detection systems, few studies have been conducted on spectroscopy-based SOC detection. Here, spectroscopy was used to record reflection or absorption from the sample, and these measurements were subsequently used to develop models to predict SOC. Some of these models are CIE color models, and some of them are spectral data-based models.

3.2. Spectroscopy-Based Models

Since spectroscopy data are immensely important, researchers have attempted to develop spectral libraries. The objective of these libraries is to use quantification models of soil properties and to support the classification and characterization of soil. The main requirements for developing a spectral library are that a sufficient number of samples represent the variability of soils, that the samples must be subsampled and handled with special care, and that the analytical and reference data used in calibration must be developed through standardized procedures [23].

Spectral analysis of soil samples to provide data has been used for several decades, but the application of these methods in SOC analysis is still at the experimental level. Spectroradiometers measure the amount of radiation using narrow wavelength bands throughout the spectrum, and they depend on ambient light, which is used to measure the reflectance at some distance or color of the radiant source. Tristimulus colorimeters use three filtered photodetectors and are usually built with a light source. Additionally, spectrophotometers have a built-in light source, and they can record measurements throughout the spectrum and provide the ratio between the light reflected from the sample and that from a calibrated working standard. Even though spectrophotometers are flexible in their analysis, they are expensive, and laboratory conditions and skilled technicians are required to conduct the analysis. However, inexpensive portable colorimeters and spectrophotometers are currently available, and these are feasible options for the development of in situ SOC analysis [24].

The use of spectral analysis for SOC has advantages compared to other methods. The use of mid-infrared (MIR) spectra provides better information related to soil C, and this is a nondestructive method that consumes zero reagent and is highly adaptable for automation and in situ analysis. Spectral analysis provides space to utilize simultaneous measurements of organic and inorganic soil C, which simplifies the analysis of widely used rational methods [17].

Considering the seven methods mentioned in

Table 3. Comparison of spectroscopic methods used for SOC analysis.

Sample Type	Instrument Used	Wavelength Used	Key Findings	Reference
1. Direct measurement from sandy soil	Minolta CM508d spectrometer Xenon light source	400–700 nm with 10 nm increments (visible spectroscopy)	Better results compared to visual matching methods. Cannot isolate small soil features.	[24]
2. Dried, ground, and sieved soil	FieldSpec® Pro visible and near-infrared spectrometer	350 nm to 1100 nm at 1.5 nm intervals (UV–visible spectroscopy)	CIE model performance is better than spectroscopy-based PLS regression model performance.	[25]
3. Dried, ground, and sieved soil	Konica Minolta chroma meter CR410 camera with a light conducting tube Bio-Rad FTS spectrometer	2500–25,000 nm at 1 cm−1 resolution (4000–400 cm−1)	L* values and SOC are negatively correlated. Ground soils have higher correlation. Visible NIR region is better for soil with high SOC.	[26]
4. Air-dried, ground, and sieved soil	FOSS XDS Rapid Content Analyzer	400–2500 nm with 0.5 nm increments	Reflectance and SOC content are negatively corelated.	[27]
5. Air-dried, ground (roller mill), mixed, and sieved (180 µm) samples	MIR: DigiLab FTS-60 Fourier transform spectrometer. NIR: Fosss-NIR System Model 6500	2500–25,000 nm at 4 cm−1 resolution (4000–400 cm−1); 1100–2498 nm at 2 nm intervals.	Carbonate carbon influences the MIR and NIR spectra. MIR has a higher model fit. Acid-treated soils respond better than untreated ones.	[17]
6. Artificially made soil samples with commercially available organic compounds	Analytical Spectral Device (ASD Boulder)	500–2500 nm	Significant accuracy. Cross-validated model. Requires powerful hardware and software.	[28]
7. Air-dried and sieved sample	Tristimulus colorimeter. Minolta CR310 chroma meter fitted with CR A33e glass light projection tube	400 to 700 nm (visible spectroscopy)	Strong correlation with the Munsell value/chroma and reflectance. Soils with similar properties provide better correlation.	[15]

, the use of expensive laboratory instruments and complex sampling procedures limits the ability to use them in an in situ context. Therefore, a matrix can be made from these methods to develop a process with minimum sample preparation and the use of portable inexpensive instruments (Figure 4). Here, the zero-sample process is given as 0 and increases up to 3. The complexity and instrument use also increase from 1 to 3. Since all of the methods use an instrument, zero is not used. The accuracy is categorized as remarkable, significant, and high among the compared methodologies. These methods show similarities as well as variations in the processes used and their key findings.

The relationship between SOC and interested variables to predict SOC in the different methods have been achieved using different statistical models: multiple linear regression (MLR) and partial least square (PLS) regression [4,15,17,22]. MLR is suitable to use with non-multicollinear data and data with more samples than variable. The ability of PLS to handle nonlinear influences within spectral data is important [17] and a PLS fit usually requires a small number of latent variables.

3.3. NIR Image-Based Model to Analyze SOC

Compared to digital images, IR images and NIR spectroscopy can capture more details of a material. Using soil samples and a spectrometer, we can record reflectance or absorbance data and co-relate this to the SOC values. Figure 5 shows the reflectance graph for a soil sample related to the 2000 to 2450 nm wavelength range recorded using the spectral engine (NIRONE sensor Spectral Engine 2.5). This portable spectrometer has an inbuilt illumination source which consists of two tungsten vacuum lamps and compatible software. The detailed and specific information recorded from the device can be used for various analyses and can replace the high-cost laboratory analysis. There is a significant relationship between SOC concentration and reflectance.

The color model dimensions also have a logarithmic relationship with SOC, such as the Munsell value [15]. Therefore, portable spectroscopic instruments are useful to perform in situ SOC analysis.

The values of the MCC and the lightness in the CIE images show clear relationships with SOC, which indicates that when the soil is lighter in color, it has a lower SOC and higher reflectivity (R). Therefore, for L (lightness),

$$\mathrm{L}\propto {\displaystyle \frac{1}{\mathrm{S}\mathrm{O}\mathrm{C}}},$$

$$\mathrm{R}\propto \mathrm{L},$$

$$\mathrm{R}\propto {\displaystyle \frac{1}{\mathrm{S}\mathrm{O}\mathrm{C}}}.$$

IR images are developed to capture images based on their unique emissivity. According to Kirchhoff’s law of thermal radiation [29],

$$\mathrm{T}+\mathrm{R}+\mathrm{E}=1,$$

where R is the reflectivity, T represents the transmissivity, and E is the emissivity. Therefore, an increase in reflectivity can reduce E or T, if not both. Since dark soils reflect less, they have a higher emissivity that can be captured by IR technology and can be used to compare soils and quantify the SOC based on emissivity.

Figure 6 shows IR images captured from nine soil samples with different SOC amounts (the SOC content is mentioned as a percentage) using an FLIR C3 camera (80 × 40 infrared resolution, 4800 temperature pixels, 0.01 °C thermal sensitivity, and temperature ranging from −10 °C to +150 °C). These samples were air-dried for 72 h under shade, ground, and sieved before being placed in a Petri dish to capture images. These images were enhanced using the compatible software available for the camera. The images are on a scale from −1.11 °C to 21.67 °C (30 °F to 80 °F), where the yellow color represents higher temperatures. Using FLIR Ignite software (https://ignite.flir.com), the IR images were standardized, and the images were further processed through Open CV to develop a region of interest (ROI) considering the middle of the Petri dish with the soil to improve the accuracy of the comparison. This is to reduce the impact of the Petri dish edges’ emission and to provide similar image sizes for comparison.

Furthermore, using a 2 mega pixel digital camera aligned with the same thermal camera, we were able to record RGB images. The images with smaller SOC differences and representing similar soil color ranges (the SOC content is given as a percentage), along with their corresponding IR images, are shown in Figure 5. Here, the IR images show the advantage of their potential to avoid interactions between light and atmospheric distortions, which impact the corresponding digital images shown. Figure 6 shows a gradual decrease in yellowness with increasing SOC.

3.4. Integrated Image Processing-Based SOC Analysis

There are identical relationships that have been derived through research conducted on SOC analysis based on color-based models and spectroscopy-based models. The following equations are useful for developing rapid SOC analysis methods.

According to Rossel and Walter [25], when considering lightness as L and SOC, the following relationship holds:

$$\mathrm{S}\mathrm{O}\mathrm{C}=-0.883+68.65\left({\displaystyle \frac{1}{\mathrm{L}}}\right)$$

Additionally, Rossel [22] provided another equation for SOC when lightness is considered to be L:

$$\mathrm{S}\mathrm{O}\mathrm{C}=4.843-1.139\mathrm{l}\mathrm{o}\mathrm{g}\left(\mathrm{L}\right)$$

In 2021, Jorge [21] identified a correlation between lightness (L) and SOC (X), but the model’s fit (R²) was 0.58.

$$\mathrm{L}=-0.44\left(\mathrm{S}\mathrm{O}\mathrm{C}\right)+40.08$$

Furthermore, the relationships between SOC and the dry soil Munsell value (V) have been identified by different researchers. Rossel and Walter [25], Konen et al. [15], and Jorge [21] mentioned this phenomenon as follows:

$$\mathrm{S}\mathrm{O}\mathrm{C}=9.934+3.074\mathrm{l}\mathrm{o}\mathrm{g}\left(\mathrm{V}\right)$$

$$\mathrm{V}=-0.7128\log \left(\mathrm{S}\mathrm{O}\mathrm{C}\right)+5.8669$$

$$\mathrm{V}=-0.04\mathrm{S}\mathrm{O}\mathrm{C}+3.92$$

To develop an SOC rapid analysis system integrated with machine learning, the relationship between SOC and lightness factors can be considered.

4. Discussion

Soil properties are either inherent or dynamic. For natural soils, soil color is directly associated with soil-forming factors, and therefore, it is inherent. However, with SOC levels, these inherent color properties can change in different dimensions with increasing or decreasing darkness. According to previous studies, SOC is also strongly correlated with the lightness parameters of the different color models. However, their prediction models have adjusted R square (adjusted coefficient of determination) values ranging from 0.68 to 0.8. In addition, the accuracy of prediction using larger, similar soil type libraries has improved [17]. Furthermore, even though the color models are different, they were developed using the Munsell HVC system, unless spectrometers were used to record reflection and values were derived using software to calculate dimensions in the CIE or RGB models.

During the research considered for comparison, several instruments with different wavelengths and light sources have been used. Sample preparation and selection are important for prediction accuracy. Apart from one study, all of the other methods have used processed samples. One study has used commercially available organic compounds [28] to create artificial soils and some other studies have modified the samples by adding acids or water. Therefore, it is necessary to consider all of the processes to find a new solution.

4.1. Color-Based Models and Machine Learning

Several color models can be used for SOC detection and these methods use Munsell color charts, which are better for qualitative purposes [4,20,21,22] and can be used simply by anyone with guidance. However, here, the estimation is conducted based on the assumption that the soil color (Munsell value) is due to a mixture of dark-colored organic substances and light-colored minerals. But this does not work well in strongly colored natural soils. On the other hand, Cartesian-type color models are appropriate for quantification [22]. Furthermore, lightness measurements in each model, for instance, the value (V) in the Munsell model and the lightness (L) in the CIE L*a*b* model, show clear correlations with SOC levels. Even for the color model, spectroscopy data could be used, and these data have the potential to improve the precision of measurements. For example, Jorge et al. [21] have used reflectance data from spectroscopy measurements to develop RGB coordinates. Research that included recording images from the soil samples also has used pixel data to develop color models [4,25].

Machine learning (ML) is a rapidly growing branch of artificial intelligence (AI) which allows computer systems to learn from data and improve their performance over time without being explicitly programmed with specific rules or algorithms. Image processing is a subfield of machine learning that involves manipulating digital images using computer algorithms, serving as a crucial preprocessing step in various applications, including computer vision and deep learning (DL) [30]. In recent days, the experimental and computational data together create a complicated scenario for interpretation, leading to the demand for ML, which promises to accelerate research in the future through automatic data interpretation using different models such as classification, regression, clustering, etc. [31]. Soil color prediction using the Munsell chart can create human errors [21,22], which can be avoided by using a trained model supported with machine learning [22]. Further digital images from soil samples can be used to develop color models where pixel information from the images has been considered in the analysis [4,25]. In the suggested model analysis, data from samples will be incorporated with thermal images to train deep learning algorithms which can be then used for SOC prediction based on the samples.

4.2. Spectroscopy

Considering spectroscopic analysis, most of the research was able to find a good fit for the data taken from laboratory-established expensive spectrophotometers. There is a remarkable improvement in spectroscopy-related analysis compared to visual color matching with the MCC, and the correlation shows higher accuracy in recording diffuse reflectance data in the 2000–2500 nm range [17,26,28]. However, the practiced methodologies have limitations; most of the instruments are laboratory-based and cannot be applied in the field, are sophisticated and complex to use, and require skilled professionals to conduct the analysis since they are expensive and therefore difficult to access and purchase [24,28]. However, the use of an instrument to record color or reflection-related data provides greater precision than visual color matching.

4.3. NIR-Based Image Analysis

Compared to digital images, IR images and NIR spectroscopy can capture more details of a material. The detailed information recorded based on emission and reflectance can be used for analysis in a border context [32]. Thermal cameras capture the details of soil based on emission, and unlike digital images, they are independent from the impact of light and atmospheric distortion. Furthermore, compatible software with these instruments can be used for standardization and simple analysis. However, to train a model, we require a significantly larger number of known samples, and this can be considered a limitation.

4.4. Integration with Multiple Sensors

When we consider rapid, simple sample processing at the farmer’s field level, inexpensive instruments for measuring reflectance from soil samples at suitable wavelengths while achieving remarkable accuracy are a must. Since reflection and absorption data are sensitive to background noise and impurities, direct reading from soil can impact the accuracy of the results [24]. Therefore, it is necessary to process the sample before testing, and here, we design a simple sample processing tool kit with easy-to-follow guidelines. Furthermore, avoiding the use of ovens and desiccators for the drying process and using simple air drying is enough. Grinding and sieving are inexpensive, so it is necessary to include these simple tools within the system to ensure uniformity in the sample data and improve the accuracy of the prediction.

If we consider the abstract findings of the conducted research, potential, rapid, and in situ quantification of SOC can be used to make precision agriculture decisions (Figure 7). From the above discussion, we can conclude that the development of spectral libraries can be achieved via the collaboration of several sensors with low-cost and compatible in situ analysis.

5. Conclusions

There is increasing concern over the necessity of reliable and updated soil-related data for the management of nutrients and water in precision agriculture; therefore, the identification of an affordable in situ SOC analysis method is crucial.

In this study, several approaches, including color-based models, spectroscopy, and NIR integration, were reviewed critically to determine the best solution for increasing soil sustainability in agricultural practices with low-cost SOC analysis. From this report, we can conclude the following outcomes to be reached in the decision to develop an in-situ analysis method:

• With similar types of soil samples, analysis using color-based models is more accurate, and there are no universal equations for all types of soils.
• The information from spectroscopy data can be used in the process of SOC analysis, and the accuracy is higher around the 2000 to 2500 nm wavelength range.
• The accuracy of the color models can be increased using spectroscopic data. Inexpensive spectrophotometers can be used to record diffuse reflection data to develop an in situ SOC analysis method.

One of the main limitations in the spectroscopy-based method is that we need known soil sample data to calibrate the model, considering all soil types. Therefore, there is a necessity to establish local spectral libraries with SOC data to support SOC monitoring. Integrating machine learning models and spectrophotometer devices with a simple toolkit with guidelines to process soil samples can motivate farmers to perform their own soil analysis. This help will help in making precision agricultural decisions related to water management and fertilizer management, as well as in reducing CO2 and CH4 emissions. They can achieve soil sustainability while enhancing crop productivity in the long term. This will be a viable solution for farmers and the environment and can open doors in the field of achieving carbon neutrality in agriculture.