Handheld In Situ Methods for Soil Organic Carbon Assessment

Abstract

Soil organic carbon (SOC) assessment is crucial for evaluating soil health and supporting carbon sequestration efforts. Traditional methods like wet digestion and dry combustion are time-consuming and labor-intensive, necessitating the development of non-destructive, cost-efficient, and real-time in situ measurements. This review focuses on handheld in situ methodologies for SOC estimation, underscoring their practicality and reasonable accuracy. Spectroscopic techniques, like visible and near-infrared, mid-infrared, laser-induced breakdown spectroscopy, and inelastic neutron scattering each offer unique advantages. Preprocessing techniques, such as external parameter orthogonalization and standard normal variate, are employed to eliminate soil moisture content and particle size effects on SOC estimation. Calibration methods, like partial least squares regression and support vector machine, establish relationships between spectral reflectance, soil properties, and SOC. Among the 32 studies selected in this review, 14 exhibited a coefficient of determination (R²) of 0.80 or higher, indicating the potential for accurate SOC content estimation using in situ approaches. Each study meticulously adjusted factors such as spectral range, pretreatment method, and calibration model to improve the accuracy of SOC content, highlighting both the methodological diversity and a continuous pursuit of precision in direct field measurements. Continued research and validation are imperative to ensure accurate in situ SOC assessment across diverse environments. Thus, this review underscores the potential of handheld devices for in situ SOC estimation with good accuracy and leveraging factors that influence its precision. Crucial for optimizing carbon farming, these devices offer real-time soil measurements, empowering land managers to enhance carbon sequestration and promote sustainable land management across diverse agricultural landscapes.

1. Introduction

The accurate estimation of soil organic carbon (SOC) is crucial to addressing global issues such as climate change, food and water security, and biodiversity conservation, thus promoting environmental sustainability and ecosystem health [1]. SOC refers to the quantity of carbon (C) stored in soils, resulting from the incorporation of organic matter derived from plant residues such as leaves, stems, and roots, which undergo microbial decomposition [2]. It remains a central and indispensable component in any research aimed at reducing greenhouse gas (GHG) emissions from land management [3]. Sustainable agricultural practices (SAPs), like conservation tillage, cover cropping, agroforestry, crop rotation, and integrated nutrient management, aim at protecting and enhancing SOC content and consequently reducing GHG emissions from agriculture while also addressing overarching global concerns [4]. SOC is intricately linked with soil health, fertility, and ecosystem resilience through its influence on soil structure, nutrient cycling, water regulation, microbial activity, and C sequestration [5]. Its accurate estimation is essential for effective land-use planning and natural resource management as SOC data support decisions regarding the practices that can help to mitigate soil erosion, improve water quality, and conserve biodiversity [6,7]. SOC estimation provides critical data for monitoring and reporting on climate actions [8]. It helps nations to track changes in SOC stocks over time, which is essential for assessing the effectiveness of SAPs in achieving national and international climate goals outlined in agreements like the Paris Agreement [9]. SOC data support C farming and participation in C offset programs by verifying C-negative practices [10,11] and enabling Payments for Ecosystem Services (PES) initiatives that compensate landowners for enhancing C sequestration [12]. Governments use SOC data to develop effective policies, set targets for SOC storage, and monitor progress towards climate goals, ensuring the sustainable management of agricultural landscapes to mitigate climate change impacts [8]. To achieve Sustainable Development Goals (SDGs) focused on agriculture and environmental sustainability, in situ SOC assessments are crucial for monitoring soil health, combating degradation, and promoting sustainable land management practices [9,13]. These assessments directly contribute to SDG targets for food security (SDG 2) and the prevention of land degradation (SDG 15) [14].

SOC assessment has traditionally relied on strenuous and time-intensive laboratory methods, which can be costly and impractical for real-time monitoring [15]. Doetterl et al. [16] compared traditional methods with spectroscopic analysis, revealing significant cost and time reductions with spectroscopy, approximately 83% and 85%, respectively, while maintaining a comparable accuracy, especially in fine-textured soils [16]. Additionally, traditional methods often involve destructive sampling, compromising soil integrity and hindering long-term monitoring [17]. Hidden erosion, as highlighted by the research of Gholami et al. [18], may result from traditional soil sampling, contributing to global soil loss when soil removal is repeated on a larger scale. Another limitation is the potential oversight of spatial variability in SOC levels due to limited sampling points and interpolation assumptions [19]. In contrast, diffuse reflectance Fourier transform infrared (DRIFT) spectroscopy offers a rapid assessment across variable landscapes as a cost-effective alternative [20]. Exploring portable, in situ alternatives like spectroscopy is crucial for SOC assessment due to these limitations. Researchers are increasingly favoring the integration of different SOC estimation methods to leverage the strength of different methods and overcome the limitations of traditional approaches. This integration of SOC measurement technologies offers scalable, resource-efficient solutions for assessing SOC across spatial and temporal scales [21].

SOC estimation in the field can be conducted using a variety of portable instruments that employ spectroscopy [22], reflectometry [23], electrochemical sensors [24], and C dioxide (CO₂) sensors [25]. Portable instruments, such as infrared spectrometers, handheld meters, soil penetrometers, and SOC analyzers, offer convenient and rapid on-site measurements of SOC content without the need for laboratory analysis [15,26,27,28,29]. Different combinations of these sensor technologies may be employed in specific portable devices [30]. Remote sensing methods, including satellite imagery, airborne LiDAR (light detection and ranging), satellite-born EnMAP sensor [31], and proximal or drone-based remote sensing, allow for the assessment of SOC with a large area coverage and provide insights into spatial patterns and variations [32,33,34]. However, it typically provides indirect measurements and may lack the fine-scale details necessary for precise SOC quantification [35]. In contrast, in situ methods offer direct measurements and provide more detailed information about SOC content at specific locations, albeit over smaller areas. Combining remote sensing with in situ measurements allows for the integration of both datasets, leveraging the strengths of each approach. Remote sensing data can be used to identify region of interests (RoIs) and provide context for in situ measurements, while in situ data can validate and calibrate remote sensing models and improve the accuracy of SOC estimates [36,37]. This integration enhances the scientific understanding of SOC dynamics at both the regional and local scales supports more effective soil management and C sequestration efforts [32]. Geographic Information Systems (GIS) further integrate spatial data to create detailed maps and models of soil C distribution [38,39]. These approaches enable real-time and spatially explicit SOC estimation and aid informed decision-making processes for soil management practices.

Within climate change mitigation and C management goals, SOC assessment has become increasingly important. To meet the demand for on-site and real-time measurements, the utilization of portable and handheld devices has gained prominence. Recognizing the significance of these devices and the challenges associated with their use, this review paper aims to:

• Explore the methods used for in situ SOC estimation, particularly focusing on the sensor technologies employed by portable and handheld devices for assessing SOC content in field settings.
• Examine the strengths, limitations, and practical applications of in situ approaches to offer insights into SOC assessment across diverse environments.
• Enhance the understanding of researchers, land managers, and policymakers regarding the factors influencing the accuracy of SOC estimation using handheld devices and offer practical strategies to mitigate challenges associated with their utilization.
• Highlight the role of in situ SOC estimation using handheld devices in the context of C farming.

Future directions for in situ SOC assessment methods involve advancements in spectroscopic techniques [20], sensor networks [24], nanotechnology [40], and remote sensing [35]. Advances in spectroscopic instrumentation and data analysis algorithms enhance the accuracy and efficiency of SOC estimation, enabling the high-throughput analysis of soil samples [41]. Sensor networks comprise arrays of in situ sensors deployed across landscapes to continuously monitor soil properties, including SOC dynamics. These sensors measure various parameters, such as soil moisture, temperature, and electrical conductivity (EC), which influence SOC decomposition and stabilization processes [24]. The integration of sensor networks with wireless communication technologies and data analytics enables the real-time monitoring of SOC changes at multiple spatial and temporal scales [42]. Nanomaterial-based sensors and probes can detect subtle changes in soil properties like soil structure, which is associated with SOC content, providing enhanced sensitivity compared to conventional methods [43]. Additionally, nanomaterials can be functionalized to target specific SOC compounds or microbial metabolites, enabling the targeted analysis of SOC fractions and microbial activity in soils [44]. Advanced remote sensing platforms, coupled with hyperspectral and LiDAR sensors, offer opportunities for mapping SOC stocks and monitoring changes in SOC dynamics at the regional and global scales [45]. Continuous monitoring aids in the early detection of soil degradation, supporting timely intervention and restoration efforts. Accurate SOC measurement empowers farmers to optimize C sequestration strategies through SAPs, crucial for enhancing SOC storage and reducing GHG emissions [46,47].

2. Methodology

Literature Search: A comprehensive literature search was conducted during the period of June 2023–April 2024 using a range of scientific databases, including Web of Science, Google Scholar, Research Gate, PubMed, and Science Direct. The search was aimed at identifying studies relevant to in situ SOC estimation using portable or handheld devices. The keyword term combinations used in the search included: SOC + portable; SOC + in situ + portable; SOC+ handheld; SOC + in situ measurement; SOC + in situ + spectroscopy; SOC + in situ + reflectometer; SOC + wireless sensing technologies.

Study Selection: The initial search generated 1646 results on handheld in situ methods for SOC estimation. Relevant studies were identified by a rigorous screening process involving the review of titles, abstracts, and full text. These studies were required to provide an overview of an in situ approach used for the estimation of SOC content across diverse land use and environmental conditions, along with corresponding accuracy metrics and performance evaluations. Consequently, 32 studies meeting these criteria were carefully selected (Figure 1), and a table was compiled to include detailed information such as study location, in situ sensing technology utilized, spectral range, number of soil samples, sampling depth, reference laboratory method for SOC estimation, preprocessing techniques, calibration method, and accuracy/error attained using the most effective model among those tested (

Table 1. Selected studies employing portable in situ approaches for soil organic carbon (SOC).

Study	Country	Land Use	In Situ Approach	R (nm)	N	D (cm)	Reference Method	Preprocessing Technique	Calibration Model	Validation	R2	RMSE (%)
Murad et al. (2023) [48]	USA	Ag	FT-NIR	1350–2500	270	0–5	DC	EPO	SVM	10-fold CV	0.17–0.70	9.2–5 *
Singha et al. (2023) [49]	India	Ag	VNIR	350–2500	200	0–30	WB	FD-SG and SD-SG	SVM	LOOCV	0.84	0.12
Jia et al. (2023) [50]	Tibet	S.M and F	VNIR	350–2500	85	0–30	DC	FD + SG	LWR	Independent	0.68	16.15 *
Karyotis et al. (2023) [51]	Cyprus, Greece, Lithuania	Na	SWIR	1750–2150	280	0–20	WB	FD-SG and SNV	RF	Independent	0.43	0.36
Murad et al. (2022) [52]	Australia	Ag	VNIR	350–2500	33	0–90	WB	EPO	Cubist	Independent	0.86	0.36
Priori et al. (2022) [53]	Italy	Ag	FT-NIR	1350–2500	182	-	WB	-	PLSR	10-fold CV	0.82	0.34
			VNIR	350–2500							0.88	0.27
			VNIR	1350–2500							0.78	0.37
Semella et al. (2022) [54]	Germany	Ag	VNIR	350–2500	75	0–25	DC	-	PLSR	Monte Carlo CV (k = 100)	0.89	2.57 *
			MIR	2500–15,000							0.98	1.12 *
Vohland et al. (2022) [55]	Germany	Ag; F and semi-Na	VNIR and MIR	400–2500 and 2500–15,000	186	0–5	DC	PCHIP	PLSR	10-fold CV	0.78	0.24
Liu et al. (2021) [56]	China	Ag	VNIR	350–2500	117	0–20	WB	SG	PLSR	LOOCV	0.77	0.09
Nawar et al. (2020) [57]	Belgium	Ag	VNIR	305–1700	381	15–25	DC	EPO	Cubist	Independent	0.76	0.12
Chen et al. (2020) [58]	China	F, S.M.	VNIR	350–2500	547	0–100	DC	SG	MLP	10-fold-CV	0.92	6.22 *
Hutengs et al. (2019) [22]	Germany	Ag	MIR	2500 –15,000	90	0–5	DC	MSC	PLSR	10-fold CV	0.63	0.17
Kusumo et al. (2011) [59]	New Zealand	Pa	VNIR	350–2500	210	3.75–11.25	DC	SG	PLSR	Independent	0.75	1.21 *
Sharififar et al. (2019) [27]	Australia	Ag and Pa	VNIR	350–2500	232	100	WB	SG& SNV	Cubist	10-fold CV	0.89	0.12
			VNIR (Neo Spectra)	1300–2500							0.78	0.16
Nawar and Mouazen (2019) [60]	Yorkshire, UK	Ag	VNIR	305–2200	139	0–15	DC	SG	RF (Spiked)	-	0.75	0.17
Crucil et al. (2019) [61]	UK	Ag	VNIR	350–2500	96	0–2	DC	SG	PLSR	LOOCV	0.95	2.5 *
Bricklemyer et al. (2011) [62]	Montana, USA	Ag	LIBS	200–300	306	0–50	DC	-	PLSR	Independent	0.22	3.2 *
Hutengs et al. (2018) [63]	Germany	Ag	MIR	2500–15,000	40	0–5	DC	DHR PCHIP	PLSR	10-fold CV	0.85	0.16
Gomez et al. (2008) [32]	Australia	Ag and Pa	VNIR	350–2500	146	0–10	MIR [64]	-	PLSR	LOOCV	0.71	0.53
			Hyperion hyperspectral RS	400–2500	72						0.51	0.73
Soriano-Disla et al. (2017) [65]	Australia	Ag	VNIR	350–2500	458	0–180	WB	SG	PLSR	LOOCV	0.65	0.33
			MIR	2500–22,000 and 2500–13,000							0.67	0.31
Wijewardane et al. (2016) [66]	USA	Ag, F, Pa, R, and W	VNIR	350–2500	19,891	-	DC	-	ANN	Independent	0.96	3.61 *
O’Rourke et al. (2016) [67]	Ireland	Pa, Ag, F, Pe	VNIR	350–2500	322	0–10	WB	SG + SNV	GRA	Independent	0.79	3.17 *
Jiang et al. (2016) [68]	China	Ag	VNIR	350–2500	95	0–10	DC	SG + OSC + GLSW	GLSW-PLSR	LOOCV	0.77	0.08
Li et al. (2015) [69]	China	Ag, F	VNIR	350–2500	413	0–100	DC	SG + FD	LS-SVM	Independent	0.81	8.40 *
Wang et al. (2015) [70]	USA	Ag	VNIR	350–2500	675	0–45	DC	SG + FD	PSR	Independent	0.88	0.2
Ji et al. (2016) [71]	China	Ag	VNIR	350–2500	184	0–20	DC	SG	LS-SVM	Independent	0.79	2.95 *
Cozzolino et al. (2013) [39]	Australia	Ag	VNIR	350–1850	140	0–50	DC	SG + SNV	PLSR	Independent	0.74	0.03
Nocita et al. (2011) [31]	South Africa	Shrubs and trees (2–5 m)	VNIR	400–2400	113	0–20	WB	MSC + FD-SG	PLSR	Independent	0.93	2.87 *
Muñoz & Kravchenko (2011) [33]	Michigan, USA	Ag	NIR	1100–2225	100	0–10	DC	None	PLS-O	Independent	0.46 (Adjusted)	1.58 *
Kuang et al. (2015) [72]	Denmark	Ag	VNIR	305–2200	212	15–20	DC	SG	ANN	Independent	0.90	1.50 *
Biney et al. (2022) [73]	Czech Republic	Ag	VNIR	350–2500	57	0–15	WB	SG + SNV + MSC	SVM	5-fold CV	0.72	0.21
Pei et al. (2019) [74]	Missouri, USA	Ag	VNIR	343–2222	148	0–120	DC	GS + SNV	PLSR	Random CV	0.80	0.19

Note: Numbers marked with an asterisk (*) indicate values in gC/kg. R-Spectral Range; N—No. of Soil Samples; D—Sampling Depth; R²—Coefficient of Determination; RMSE—Root Mean Squared Error; Ag—Agriculture; Pa—Pasture; F-Forest; S.M—Shrub Meadows; Pe—Peatland; W—Wetland; Na-Natural Areas; VNIR—Visible Near Infrared Spectroscopy; NIR—Near Infrared Spectroscopy; MIR—Mid Infrared Spectroscopy; FT-NIR—Fourier-transform Near Infrared spectroscopy; INS—Inelastic Neutron Spectroscopy; LIBS—Laser-Induced Breakdown Spectroscopy; RS—Remote Sensing; WB—Walkley & Black Method; DC—Dry Combustion Method; OSC—Orthogonal Signal Correction; GLSW—Generalized Least Squares Weighting; MSC—Multiplicative Spectral Correction; EPO—External Parameter Orthogonalization; SNV—Standard Normal Variate; FD—First Derivative; SD-Second Derivative; SG—Savitzky–Golay Smoothing; DHR—Directional Hemispherical Reflectance; PCHIP—Piecewise Cubic Hermite Interpolating Polynomial Approach; GS—Gaussian smoothing; SVM—Support Vector Machine; LWR—Locally Weighted Regression; ANN—Artificial Neural Network; PLSR—Partial Least Squares Regression, PSR—Penalized Spline Regression; PLS-O—Partial Least Squares Leaving-One-Outlier-Out; RF—Random Forest; LS-SVM—Least-Square Support Vector Machine; GRA—Granger-Ramanathan Model Averaging; CV—Cross-validation; LOOCV—Leave-one-out cross validation.

). This table forms the basis for discussing various aspects of in situ SOC assessment using handheld devices in the subsequent sections.

3. In Situ SOC Methods

3.1. Spectroscopic Methods

For several decades, spectroscopic methods have been extensively utilized for in situ SOC estimation using handheld devices. Among these methods, visible near-infrared (VNIR) and mid-infrared (MIR), inelastic neutron scattering (INS), and laser-induced breakdown spectroscopy (LIBS) have emerged as the most prevalent techniques for soil C estimation [69,75,76]. Each method offers unique benefits and presents its own set of challenges (

Table 2. In situ approaches for SOC estimation.

Method	Spectral Range (nm)	Benefits	Challenges	References
Visible and Near-Infrared Spectroscopy	350–2500	Rapid; low-cost; require little sample preparation; account for moisture and texture variations	Limited depth of measurement; sample size and variability	[80,81,82,83]
Mid-Infrared Spectroscopy	2500–25,000	Suitability across different organic matter contents; simultaneous analysis of multiple soil properties; high throughput capability; depth-specific predictions	Requires sample preparation; reduced sensitivity in environments with high humidity; instrument cost	[21,84,85,86,87]
Laser-Induced Breakdown Spectroscopy	190–1000	Rapid analysis (one minute per sample); minimal sample preparation	Spatial variability resulting from small-point sample measurements; instrument sensitivity decreases with the increase in carbon; matrix effects due to other soil elements; quantification issues due to spectral overlap; limited depth of measurement	[88,89,90,91,92,93]
Inelastic Neutron Scattering	Gamma ray (approx. 0.01 nm or less)	Non-destructive; minimum sample preparation; provides multiple elements; large area scanning	Achieving precise depth resolution; instrument complexity; matrix effects, such as interactions between soil constituents and neutron beams	[94,95,96,97]

). VNIR and MIR methods estimate soil C by analyzing the interaction between soil constituents and specific infrared radiation ranges [22,54,65]. A typical infrared spectrometer consists of light source, wavelength selector, sample holder, and a detector [77]. A light from the source is directed onto the soil sample, and the reflected or transmitted light is analyzed by the detector after passing through the monochromator, which separates the light into different wavelengths (Figure 2). The spectrometer focuses on fundamental vibrational transitions where molecules absorb infrared radiation to transition between different vibrational states [21,77]. This absorption pattern generates a unique spectral fingerprint for each substance, allowing for the precise identification and quantification of soil components. VNIR uses 400–2500 nm radiation, while MIR focuses on 2500–25,000 nm radiation [21,22]. By analyzing the unique spectral features of soil C in these ranges, spectrometers can measure and analyze the absorbance data [55,78,79]. Calibration models are then developed by comparing the absorbance data with the estimated C values, allowing for soil C modeling [65].

VNIR and MIR, both infrared spectroscopic methods, offer a quick, non-invasive, and reliable approach for assessing soil C in field conditions [98]. However, there are limitations associated to each one of them. The shallow sampling depth of VNIR limits deeper soil analysis without additional tools, like fiber optic probes [24]. VNIR accuracy is affected by SOC spatial variability, soil moisture, and particle size variations, requiring complex preprocessing and statistical models [99,100]. Jiang et al. [68] study examines how soil moisture affects SOC predictions using VNIR spectroscopy. It shows that, while partial least squares regression (PLSR) models accurately predict SOC under specific moisture ranges (50–100, 250–300 g·kg⁻¹), they struggle with transferability across different moisture levels. Orthogonal signal correction (OSC) and generalized least squares weighting (GLSW) methods effectively mitigate moisture interference. In contrast to VNIR spectroscopy, MIR spectroscopy offers comprehensive insights into a broader range of soil properties, including particle size distribution [101]. This is achieved by leveraging molecular vibrations of MIR-active components, particularly soil organic matter (SOM), allowing for a detailed analysis beyond the SOC content alone [22]. It supports in situ analyses but depends on the presence of MIR-active components and is sensitive to small-scale soil heterogeneity and environmental conditions, which can impact measurement reliability [84,102]. MIR spectroscopy is favored by researchers over VNIR for its ability to offer higher information density on SOC bands and for producing more consistent spectra with higher signal-to-noise ratios and minimal wavelength variability [85,86,103]. The accuracy of soil C estimation depends on the quality of the calibration dataset and the methods used for spectral preprocessing [104,105,106]. Fourier Transform Infrared–Attenuated Total Reflectance (FTIR-ATR) coupled with MIR spectroscopy is a highly effective method for both quantitative and qualitative analyses. This technique offers several advantages, including the ability to analyze samples directly without the need for dilution, thereby reducing the measurement time. By utilizing FTIR-ATR, researchers have achieved efficient and accurate SOC measurements [84]. MIR spectra can be affected by water vapor, necessitating careful handling in humid environments [107].

The INS method involves the use of fast neutrons, typically generated by a pulsed Deuterium–Tritium (D-T) neutron generator. These neutrons interact with C atoms in the sample, leading to the emission of gamma rays, which can be detected to determine the C content [94,108]. The detection time in this method is influenced by factors such as the energy of the generated neutrons, their interaction with the sample, and the concentration of the C element of interest [109]. In a validation study conducted by Wielopolski et al. [110], the INS methodology was evaluated at multiple field sites to verify its calibration. The measurements obtained from the INS system were compared to traditional dry combustion analysis. Measurement errors varying between 5% and 12% were identified in the study, with a minimum detection limit of 0.018 g C. The authors proposed enhancements such as augmenting detector numbers and optimizing the system setup to improve the precision of C analysis with the INS method. They also highlighted the importance of calibrating the INS system for different soil types and exploring its potential for scanning large areas. Yakubova et al. [95] showcased the reliable and efficient field performance of mobile INS (MINS), offering significant time savings compared to traditional methods like dry combustion technique (DCT). With a linear correlation between MINS and DCT results, MINS demonstrates promise for in situ soil C assessment, with an accuracy influenced by both signal precision and soil C distribution.

LIBS is an emission-based spectroscopic method that involves exciting the atoms in a sample using plasma energy and then measuring the radiation emitted as they return from higher to lower energy states [88]. Each element has unique spectral characteristics, and by measuring the specific wavelengths and intensities of the peaks, the elemental composition of the sample can be determined. LIBS application in soil C estimation in field settings has generated reliable results with a good accuracy and precision [89]. However, challenges arise from interference caused by inorganic carbonate deposits, particularly iron (Fe), at specific wavelengths, as well as the presence of soil water [111]. Glumac et al. [112] confirmed the efficacy of standard LIBS for accurately measuring soil C content, even in the presence of Fe interference. It showed a strong correlation (r² = 0.94) between the LIBS C signal and the traditional dry combustion method on soil samples with an SOC ranging from 0.5% to 3% (w/w). Additionally, this study validated the use of the 247.8 nm C spectral line in soil LIBS analysis, enabling the precise determination of the C content down to sub-percent levels through optimized signal processing techniques [112]. Other researchers have addressed these challenges by introducing a different C line at 193 nm and developing a two-element standardization factor [90]. Additionally, the spatial variability resulting from small-point sample measurements presents another obstacle. Nevertheless, the development of portable LIBS instruments offers a potential solution by facilitating the efficient measurement of multiple soil samples [89].

3.2. Remote Sensing and GIS

While remote sensing techniques and spatial imagery methods are not typically utilized for direct, in situ SOC measurement at specific locations, they are often employed to complement in situ measurements by providing spatial data, analysis tools, and modeling capabilities for SOC estimation and mapping [36,37,49,111,113,114,115]. Remote sensing methods revolutionize in situ SOC prediction by harnessing data from satellites, airborne platforms, and spatial models like Digital Elevation Models (DEMs) [35,116,117,118]. By utilizing multispectral or hyperspectral sensors, these methods capture images of the soil surface, revealing crucial information on soil reflectance, vegetation indices, and surface temperature [119]. By analyzing these spectral signatures, remote sensing algorithms and models can infer SOC content. These algorithms are often based on statistical relationships between soil C measurements taken in the field and the corresponding spectral information obtained from satellite images. Pascucci and colleagues demonstrated the efficacy of utilizing airborne hyperspectral imagery in the thermal infrared region (8.0–11.5 μm) for predicting topsoil properties in agricultural fields, emphasizing its utility in quantitatively assessing SOC content in the topsoil [120]. However, the existence of interfering factors at the soil surface, including crop residues [121], soil temperature [119], soil wetness, and soil texture [122] introduce variability and noise into the remote sensing data, making it challenging to obtain precise SOC estimations. The integration of satellite-derived indices like the Normalized Difference Vegetation Index (NDVI) with machine learning models has shown promise in improving SOC estimation by accounting for spatial variability and soil heterogeneity [123,124,125]. Moreover, the Normalized Difference Soil Moisture Index (NSMI) offers a robust approach for estimating the surface soil moisture content, leveraging specific spectral features ideal for field SOC estimation [126]. Integrating VNIR data from aerial photographs with topographical features enhances SOC prediction models, particularly in scenarios with low SOC contents and narrow data ranges [33]. Additionally, airborne imaging spectroscopy (AHS-160 sensor; 430–2540 nm) emerges as a potent tool for mapping SOC across diverse landscapes, with local calibrations proving effective in capturing spatial patterns linked to topography and management variables [34]. These advancements in remote sensing provide unparalleled opportunities for accurate in situ SOC estimation, necessitating careful sensor selection and methodological refinement to suit the specific research objectives and environmental conditions.

GIS plays a crucial role in SOC estimation and mapping, allowing for the integration of diverse data sources, such as satellite imagery, soil characteristics, land classification, topographic information, and meteorological data mapping (

Table 3. Data, approaches, and challenges for SOC estimation in Geographic Information Systems (GIS).

Data Source	Analytical Approaches	Challenges
Satellite Imagery	Image classification, object-based image analysis (OBIA), change detection	Atmospheric interference, cloud cover, image resolution limitations
Soil Characteristics	Soil sampling and laboratory analysis, soil databases	Spatial variability, limited soil data availability
Land Classification	Land cover classification, land-use mapping	Classifying complex land cover types, integrating land-use change data
Topographic Information	Digital elevation models (DEMs), slope and aspect analysis	Data resolution limitations, accuracy of DEMs
Meteorological Data	Climate data interpolation, weather station data	Limited weather station coverage, missing data
Spectroscopic Measurements	Spectral indices, regression models (e.g., PLSR)	Calibration and validation, spectral signature interpretation challenges
Remote Sensing Data	Vegetation indices (e.g., NDVI), thermal infrared imagery, hyperspectral data	Sensor limitations, atmospheric correction, data integration issues
Spatial Information	GIS overlay analysis, spatial interpolation, geostatistical techniques	Data incompatibility, uncertainty in interpolation, scale mismatch
Machine Learning Algorithms	RF, SVM, ANN	Overfitting, model complexity, data requirements
Spatial Analysis	Spatial autocorrelation analysis, hotspot analysis, spatial clustering	Spatial dependence, data clustering interpretation

) [127]. Through spatial analysis within a GIS framework, these data layers are combined to derive estimates of the SOC content [128,129,130]. When combined with remote sensing and spectroscopic measurements, GIS enables the identification of areas with high or low SOC contents based on spectral signatures and spatial patterns [131,132,133]. Furthermore, by integrating spectroscopic measurements with spatial information, such as soil type, topography, and vegetation cover, GIS enhances the accuracy of SOC estimates and facilitates the spatial mapping of its distribution [39,72,134]. For instance, Ayala et al. [133] demonstrated the effective integration of remote sensing and GIS to predict SOC storage in herbaceous páramo ecosystems. By incorporating novel Sentinel-2-derived SOC predictors, such as vegetation cover and canopy water content, alongside traditional variables, the study improved SOC estimation accuracy and spatial resolution across different soil profiles using Gaussian processes regression (GPR). This approach offers detailed SOC mapping in challenging environments and shows promise for a broader application in similar ecosystems globally, facilitating better land management and climate change mitigation strategies. Tiruneh et al. [135] conducted a study in Aba Gerima, Ethiopia, utilizing spectroradiometric data for geospatial modeling and the mapping of SOC. Employing ArcGIS 10.5, they applied semi-variogram modeling and geostatistical methods, such as Ordinary Kriging and Least Distance Weighting, to map the variability in soil properties in the area. Their analysis revealed robust spatial patterns, effectively predicting clay and sand contents and SOC across the study area [135]. In another study, Kok et al. [136] integrated VNIR spectroscopy with proximal sensors and Sentinel-2 remote sensing data to predict the SOC content effectively. Their deep learning model utilized transfer learning to handle variations introduced by soil moisture, and sensor fusion techniques improved the accuracy of portable sensor models. In addition, van der Voort et al. [137] employed the SoilCASTOR method to estimate SOC stocks across arable fields in the USA using satellite data, VNIR scanner SOC measurements, and machine learning. This approach provided precise estimates of SOC stocks in the top 30 cm of soil, ranging from 19 to 55 t C/ha, with a high level of spatial accuracy. The method’s low cost and time investment make it a promising tool for SOC stock assessment in diverse soil types and agricultural settings. The insights from these integrative approaches provide valuable guidance for improving soil management practices to enhance agricultural productivity and ensure ecological sustainability. However, challenges such as the spatial resolution limitations of remote sensing [21], data quality variations [136,138], and the transferability of models [139] across regions need to be addressed for broader implementation and reliable soil management recommendations. Additionally, the temporal analysis of satellite data remains underexplored, with multitemporal approaches showing potential advantages [140,141].

4. Reference Laboratory Methods

Reference laboratory methods are standardized techniques used to quantitatively measure SOC in the lab. These methods involve the collection of soil samples from the field, transportation to the laboratory, and subsequent preparation. Preparation steps include drying, grinding, sieving, and weighing the samples before analysis. They are considered the “gold standard” for SOC analysis and are used to validate and calibrate new in situ SOC measurements [142,143]. These methods include techniques like wet oxidation and dry combustion, each with its own principles, procedures, and applications [144].

4.1. Wet Oxidation

One of the most widely used laboratory methods for SOC estimation is the Walkley and Black chromic acid wet digestion method [145]. This method involves oxidizing carbonaceous compounds present in soil samples into CO₂ using a chromic acid solution, comprising potassium dichromate and sulfuric acid. The resulting oxidized C is subsequently quantified through titration with a ferrous ammonium sulfate solution. The results are either reported as percentage SOC or grams of SOC per kilogram of soil (g/kg). However, despite its widespread use, the Walkley–Black procedure has its limitations. One major issue is that it may not fully oxidize all forms of SOC [146]. Some forms of SOC, such as humus or recalcitrant C, may not be completely oxidized under the conditions of the procedure [147]. As a result, the procedure may underestimate the actual amount of SOC. Thus, researchers have developed correction factors to account for the incomplete oxidation of SOC to address this limitation [17]. These correction factors aim to adjust the measured SOC to better reflect the true SOC content. However, it is important to recognize that these correction factors can vary significantly among different soil types, soil depths, and land-use conditions [148,149]. Researchers have used modified Walkley and Black methods to enhance the recovery of soil C by incorporating a definite heating time and temperature for soil–chromic acid mixtures in a test tube [17,150,151].

4.2. Dry Combustion

4.2.1. Loss on Ignition

The loss on Ignition (LOI) method involves heating a sample in a muffle furnace to combust SOM and decompose carbonate minerals, releasing CO₂ [143,150]. The difference in mass before and after ignition provides an estimate of the percentage of SOM [144]. Conversion equations, which vary with soil type, depth, and land use, are then applied to convert SOM to SOC [152,153]. Uniformity in sample size, exposure time, and furnace positioning significantly influences LOI results, particularly with increased sensitivity observed at 550 °C [154]. Hoogsteen et al. [155] concluded that using a sample mass of at least 20 g minimized variation in LOI measurements. They also found that an ignition temperature of 550 °C ensured the complete oxidation of SOM and emphasized that applying a clay correction factor was essential for obtaining accurate results across different soil types and climates [155].

4.2.2. Elemental Analysis

Dry combustion in an elemental analyzer involves precise sample weighing and placement into a high-temperature combustion furnace, typically operating between 950 °C and 1100 °C [143]. Within this furnace, the sample undergoes complete combustion in an oxygen-rich environment, converting C to CO₂. The resulting CO₂ is then detected using gas chromatography, with its signals proportional to the total C (TC) content present in the original sample. Subsequently, the SOC is calculated by subtracting the soil inorganic C (SIC) from TC. The measurement of SIC can be conducted using standardized methods, such as the modified pressure-calcimeter method [156]. Notably, several elemental analyzer models used in dry combustion studies—including the Analytik Jena Multi N/C 3100, Germany [50]; the Vario EL Elemental Analyzer, Hanau, Germany [55]; the Vario MAX CN Analyser, Hanau, Germany [61]; the HekaTech Euro EA elemental analyzer, Wegberg, Germany [63]; the Thermo Scientific Flash 2000, Waltham, MA, USA [68]; the Leco Truspec CHN analyzer, St. Joseph, MI [62]; and the Costech ECS 4010 CN analyzer, Milan, Italy [33]—follow similar principles, albeit with variations in design and specifications.

5. Optimizing In Situ SOC Estimation

The accurate estimation of SOC through in situ methods relies heavily on a meticulous approach across various critical steps (Figure 3). Site selection stands as the foundational pillar, requiring a site that appropriately represents the RoI. This entails considering a diverse range of soil types, land uses, and environmental conditions characteristic of the target area [157]. Furthermore, attention to spatial distribution is crucial to capture the inherent variability in soil properties effectively, ensuring robust estimations [158,159]. Equally paramount is ensuring internal homogeneity within selected sites, as variability could introduce bias into the estimations [22]. Drawing from existing data and local knowledge further enhances the precision of site selection, providing invaluable insights into the nuances of the environment [160]. Consideration of environmental factors such as vegetation cover, topography, and hydrology add another layer of complexity to site selection, influencing the accuracy of SOC estimations [161,162]. Following site selection, input variable selection becomes imperative. This involves identifying and choosing the most relevant factors that contribute to SOC prediction [103]. These factors encompass a wide range, spanning from physical properties like soil texture and color to environmental conditions like temperature and moisture levels [125]. Through a judicious combination of scientific knowledge, statistical methods, and machine learning techniques, the selection of pertinent variables ensures the development of finely tuned models capable of delivering accurate estimations [163].

Instrument method selection further refines the estimation process, necessitating the choice of appropriate technology for data collection, considering factors such as sensor type and technology, sensitivity, specificity, wavelength range, and calibration standards [164,165]. Each sensor type, be it VNIR, MIR, or short-wave infrared (SWIR) spectroscopy, possesses distinct strengths and limitations in detecting variations in SOC content [28]. Calibration standards and protocols establish the groundwork for instrument accuracy and reliability across diverse soil conditions, underlining the importance of meticulous calibration [166]. Once the appropriate technology is chosen, the selection of diverse soil samples from representative sites is crucial to capture the full spectrum of soil properties effectively [167]. Spectral data acquired using handheld devices, along with the recording of relevant metadata such as the location of the sample, land use, and moisture content, ensures comprehensive information availability, enhancing the accuracy of subsequent SOC estimations [22]. Subsequently, laboratory analysis is conducted to evaluate the soil samples for actual SOC content, providing a comprehensive understanding of SOC distribution across the study area [168]. The quality and uniformity of the reference laboratory techniques used significantly influence the precision and reliability of SOC prediction [169]. Following data collection, spectral data preprocessing plays a pivotal role in enhancing prediction accuracy. Techniques like such as moving average, median filtering, fractional order derivatives (FOD), Savitzky–Golay (SG) smoothing, multiplicative scatter correction (MSC), discrete wavelet transform (DWT) and data standardization (DS) and normalization are employed to address noise and interference present in raw spectral data, thereby improving the signal quality and reliability of the SOC content [170,171,172]. In the subsequent data modeling phase, the calibration and validation steps are pivotal. Calibration establishes a robust mathematical relationship between spectral data and actual SOC content measured in laboratory analysis [173]. Ensuring the calibration set comprehensively covers relevant variations in the data is imperative for accurate model development [174]. Validation of the model’s performance using statistical metrics like the coefficient of determination (R²), root-mean-square error (RMSE), and residual prediction deviation (RPD) provides a critical assessment of its predictive capability [175]. Various regression algorithms, including multiple linear regression (MLR), support vector machines (SVM), PLSR, gradient-boosted machines (GBM), random forest (RF), artificial neural network (ANN), and wavelet neural network (WNN) are utilized based on the spectral data’s nature and soil characteristics, further refining the modeling process [176]. Once thoroughly evaluated and optimized, the implemented model facilitates real-time SOC estimation in the field using handheld devices.

6. Discussion

6.1. Handheld Devices for SOC Estimation

Handheld devices offer a convenient, cost-friendly, and effective way of estimating soil C in field settings that can be utilized for modeling and mapping (

Table 4. Comparison of handheld devices for in situ SOC.

Device Name	Type	Spectral Range (nm)	Spectral Resolution (nm)	Manufacturer	Light Source	Detector	References
Agilent 4300 Handheld Spectrometer	Handheld FTIR Spectrometer	2500–15,000	1.25 × 106	Agilent, Santa Clara, CA, USA	QCL	DTGS	[63,179]
Terra Spec 4 Hi-Res Mineral Spectrometer	Portable Near-Infrared Spectrometer	350–1000	3–6	PANalytical Boulder, USA	Halogen	512 element silicon array	[99]
VERIS P4000 soil profiler	Soil Electrical Conductivity Meter	350–2200	8	Veris Technologies, Salina, KS, USA	Halogen	CCD, InGaAs	[74]
ASD FieldSpec3 Spectroradiometer	Portable Near-Infrared Spectrometer	350–2500	3–6	Analytical Spectral Devices, Boulder, CO, USA	Halogen lamp	Silicon-CCD	[69,180]
Model SAS3000 Soil sensor	VNIR Spectrophotometer	320–1700	-	Shibuya Seiki Co., Ltd., Japan	Tungsten halogen lamp	Si-CCD	[178]
Mobile CompactSpec	VNIR Spectrophotometer	305–1700	1	Tec5 Technology, Germany	Xe flash lamp; halogen	CCD	[57]
Nirone sensor S2.2	SWIR	1750–2500	20	Spectral Engines Steinbach, Germany	2 tungsten vacuum lamps	InGaAs	[51]
Neospectra Scanner	FT-NIR MEMS	1350–2500	16	Neospectra Si-Ware Systems (Menlo Park, CA, USA)	LED	InGaAs	[53]

Note: QCL—Quantum Cascade Laser; CCD—Charge Coupled Device; InGaAs—Indium Gallium Arsenide; DTGS—Deuterated-triglycine sulfate.

). These devices are designed with enhanced portability, allowing for on-site measurements that eliminate the logistical challenges of sample transportation to laboratories [79]. Moreover, modern handheld devices are equipped with robust data processing capabilities and connectivity features [42], enabling real-time analysis and integration with GIS [133]. This functionality facilitates immediate decision-making processes in soil management practices based on up-to-date information. These innovations significantly enhance the accuracy and precision of SOC estimation compared to traditional laboratory-based methods [177]. They are calibrated against standard laboratory techniques to ensure reliability across various environmental conditions [178]. Additionally, by reducing the need for sample transportation and laboratory analysis, these devices minimize environmental impact and support sustainable practices [37]. Several studies that are discussed in this section have demonstrated the good accuracy and precision of portable devices for SOC estimation. These devices have been extensively tested and validated in different field settings, showcasing their reliability and robust performance. Field-portable spectrometers, soil C analyzers, and handheld probes are the three types of portable devices commonly used for in situ SOC estimation. Each device has its unique features and measurement capabilities, and applications, providing researchers and practitioners with a range of options to suit their specific needs (Figure 4). These devices are designed with enhanced portability, allowing for on-site measurements that eliminate the logistical challenges of sample transportation to laboratories.

6.1.1. Field-Portable Spectrometers

These devices use spectroscopic techniques, such as VNIR, MIR, and portable X-ray fluorescence (pXRF), to estimate SOC content. The performance of VNIR spectroscopy in predicting SOC was evaluated by collecting the reflectance spectra of 250 soil samples utilizing a portable Terra Spec 4 Hi-Res Mineral Spectrometer [99]. Upon comparing results with PLSR and SVM calibration techniques, SVM regression achieved a good prediction accuracy with the R² value of 0.84. In a UK study, TC was estimated using on-line VNIR spectroscopy with a portable fiber-type spectrophotometer (AgroSpec from tec5, Germany) [181]. A comparative analysis was conducted to evaluate the effectiveness of GBM, ANN, and RF as modeling techniques in predicting soil properties. The calibration models, developed using a combination of local farm samples and a European dataset, showed that the RF models utilizing the spiked European dataset achieved the highest accuracy (R² = 0.98). In a Belgium study conducted by Nawar and Mouazen in 2022, the MIR spectra of soil samples were used along with the stacked generalization machine learning (SG-ML) approach, which combined SVM, GBR, and RF [182]. Linear ridge regression was employed as a meta-learner to predict SOC with improved accuracy. The results showed that the SG-ML approach compared to the individual ML models enhanced the accuracy of MIR spectra, resulting in reduced prediction errors and highlighting its effectiveness for precise SOC estimation. Portable MIR spectrometers demonstrated its crucial role in predicting soil properties by outperforming other miniaturized VNIR spectrometers [65]. In contrast to Soriano-Disla et al. (2017), VNIR spectroscopy demonstrated comparable or slightly better accuracy than MIR spectroscopy for soil property estimation [179].

Lin and Liu (2022) [183] employed the soil-moisture-index spectrum reconstruction (SSR) method with PLSR to improve SOC estimation accuracy. They used an ASD FieldSpec 3 spectrometer to acquire soil spectra and determined SOC contents using the potassium dichromate capacity method. The results demonstrated that the SSR-PLSR model, incorporating the spectral data and soil moisture information, outperformed the PLSR model, providing more accurate SOC estimates [183]. The impacts of spectrometers, namely ASD FieldSpec 3, ASD FieldSpec 4, and FOSS XDS, as well as scanning conditions on SOC prediction models were examined [165]. Spectra were collected from 143 samples across three different laboratories. The findings illustrated that the FOSS setup provided more accurate SOC predictions compared to other setups. Merging uncorrected spectra yielded poor results but merging soil spectral libraries with an internal soil standard improved prediction accuracy [165]. Chatterjee et al. (2021) [184] investigated the potential of combining proximal sensing with pXRF spectra and electromagnetic induction (EMI) data with remote sensing information for predicting soil physicochemical properties. The outcomes indicated that pXRF spectra, when combined with PLSR models, can effectively predict SOC. Additionally, the fusion of EMI, Sentinel-2, and DEM data shows promise in mapping soil properties across a heterogeneous crop field [184]. The potential of two hyperspectral sensors (STS-VIS and STS-NIR) and two multispectral cameras (Parrot Sequoia and Mini-MCA6) was investigated for predicting SOC content in a study conducted by Crucil and others [61]. Both indoor laboratory environment and varying the outside conditions were considered in the study as illumination and the atmospheric conditions affect the spectral data. The data unveiled that the VNIR range provided the most accurate estimation of SOC content under outdoor conditions and narrow bands models from the multispectral cameras performed comparably or better than that of hyperspectral sensors that utilized continuous spectra [61]. Portable gamma sensors, like the Medusa MS-700, offer a rapid and efficient means of estimating SOC and assessing soil health parameters in agricultural fields [185]. By emitting gamma rays from a radioactive source and measuring their intensity after passing through the soil, these sensors detect the attenuation, or reduction, in gamma-ray intensity caused by interactions with soil particles. The gamma-ray attenuation densitometer within the sensor enables the quantification of this intensity, which can be correlated with the SOC content and other important soil properties [186].

6.1.2. Soil C Analyzers

These devices measure CO₂ emissions from the soil as an indirect indicator of the TC content in the soil. They are commonly used in field studies to assess soil C dynamics and are designed specifically for on-site SOC estimation. Tóth et al. [187] introduced a novel method for the rapid assessment of TC using a portable infrared gas analyzer. This technique involves measuring the emitted CO₂ concentration resulting from the reaction of the soil sample with an acidic solution of potassium permanganate. Additionally, the method includes a separate measurement of the SIC content. The difference between the two measurements provides an estimate of SOC. By measuring CO₂ concentrations and fluxes, non-dispersive infrared (NDIR) gas analyzers, such as the soil microbial activity assessment contraption (SMAAC), help to quantify the amount of C being respired by soil microorganisms, offering a non-destructive and real-time assessment of soil C dynamics [188,189]. The portable tunable diode laser absorption spectroscopy (TDLAS) sensor designed by Gu et al. [190] showed promising performance as a gas analyzer for the online monitoring of CO₂ and H₂O concentrations. It exhibited comparable accuracy to the commercial NDIR, highlighting its potential role in monitoring soil C flux. Brecheisen et al. [191] presented an innovative approach utilizing the Field-Portable Gas Analyzer (FPGA) platform to investigate the dynamics of belowground aerobic respiration and estimate soil C in deep soil systems. By analyzing the relationship between apparent CO₂ accumulation and O₂ consumption, the FPGA platform enables a detailed analysis of soil C dynamics. While the portable chamber utilizing infrared gas analyzer provided simple and rapid measurements of soil CO₂ flux, there were variations compared to the static chamber using gas chromatography, indicating the need for further research and calibration for accurate soil CO₂ flux estimation [192]. The comparison of static and dynamic chamber methods for measuring the soil surface CO₂ flux highlighted differences in flux estimates and emphasized the importance of considering instrument-specific characteristics and variability factors [193].

6.1.3. Handheld Soil Probes

These probes typically utilize spectroscopic or EC measurements to determine SOC content. Capacitance-resistance probes, through calibration procedures, can play a vital role in estimating SOC by continuously monitoring soil water content and EC [194]. Reeves and McCarty [195] demonstrated the successful application of VNIR spectroscopy with a fiber-optic probe for determining soil composition, particularly SOC and total nitrogen (TN), although caution is needed in handling outliers. To achieve accurate measurements of SOC using compact VNIR spectrophotometers, it is proposed to employ a rotational–linear sample probing device [196]. This specialized instrument allows for the systematic and controlled sampling of the soil by rotating and linearly moving the probe. By doing so, it ensures the comprehensive coverage of the soil, facilitating the acquisition of high-quality spectral data. The use of a portable field spectroradiometer with an illuminating contact probe facilitated the collection of soil reflectance data at multiple locations, allowing researchers to capture the spatial heterogeneity of soil properties such as SOC content [197]. Shonk et al. (1991) developed a prototype sensor for the real-time measurement of SOM. This innovative sensor utilizes a narrow-band light source and detects the reflectance from the soil using a photodiode [198]. The initial field tests using the prototype sensor yielded positive results, with a high R² (>0.83). However, the further development and refinement of the sensor are necessary to enhance its performance and capabilities.

The combination of the 3S-HeD (Sensing, Scanning, and Synthesis of Soil Hyperdimensional Data) optical probe and a field spectrometer shows promise for quantitative and objective soil profile characterization in situ [199]. Nonetheless, further research is necessary to refine the instrument, validate its performance across diverse soil conditions, and extend its usage to different geographic locations. The integration of the vertical probe with the Veris P4000 soil profile instrument, along with the incorporation of spectral data, depth information, and additional soil parameters such as soil texture and EC, led to improved accuracy of in situ SOC measurements [74]. The proposed method for the rapid assessment of the SOC content utilizes a specialized instrument consisting of a metal tube equipped with a heating probe [177]. This instrument is designed to heat the soil sample to approximately 400 °C in a controlled manner. The combination of the heating probe and the CO₂ sensor enables the estimation of the SOC content by measuring the concentration of CO₂ released during the oxidation process. In an Arctic study, researchers employed an in situ VNIR probe to develop calibration models for predicting the SOC content in polar desert soils [200]. The local models successfully predicted SOC at the target sites, and the inclusion of a soil probe enhanced the accuracy of regional models. This approach enables efficient and accurate SOC estimation in remote Arctic regions using field-portable instruments.

6.2. Improving Prediction Accuracy of SOC Content

The accuracy of handheld devices for SOC prediction depends on the calibration models, preprocessing techniques, and data interpretation [201]. The quality of the calibration dataset directly impacts the accuracy of the device’s predictions [22,202,203]. Preprocessing methods significantly contribute to improving accuracy prediction by increasing the R² and reducing the RMSE [172,204,205]. Many studies have investigated different combinations of regression models and preprocessing techniques within in situ approaches. A prevalent trend has been the utilization of VNIR spectroscopy coupled with PLSR to enhance the SOC accuracy (Table 1). Among the 32 studies reviewed and detailed in Table 1, 14 demonstrated good accuracy, with an R² of 0.80 or higher, by aiming to strike an optimal balance between reducing noise, identifying significant features, extracting crucial wavelengths, and improving overall prediction accuracy. For instance, in soil profiles and subsurface measurements, the combination of VNIR and PLSR with absorbance transformation yielded the best results [206]. On the other hand, for surface soils, VNIR with spectral normalization using standard normal variate (SNV) demonstrated superior performance. SNV preprocessing may have reduced the impacts of soil heterogeneity and scattering, leading to improved accuracy in surface soil SOC prediction [206]. In contrast, few studies [83,207,208] reported no improvement in the spectral properties for the prediction of SOC content using pretreatment. Illumination normalization is a valuable preprocessing technique utilized to improve the accuracy of SOC content prediction from images captured through cellular phones [209]. This technique effectively eliminates accuracy errors caused by various factors, including dark gaps captured in images and the presence of small crop residues [209]. Moreover, it compensates for variations in lighting conditions, ensuring consistent and reliable predictions [209,210]. Heil and Schmidhalter [205] utilized NIR spectroscopy to compare various preprocessing transformations and found that derivations with Savitzky–Golay provided the best fit and the highest model accuracy. In the study by Biney et al. [211], combining multiple calibration techniques and employing several pretreatment algorithms, such as in an ensemble model, demonstrated a superior performance compared to those by individual methods. Overall, studies have carefully optimized factors, such as spectral range, pretreatment methods, and calibration models, to enhance the accuracy of SOC content measurements. This demonstrates both the methodological diversity and ongoing pursuit of precision in direct field measurements.

6.3. Comparing Portable Device Methodologies

In the pursuit of the accurate estimation of the SOC content, a range of spectroscopic techniques have been examined in recent studies. Bricklemyer et al. [212] compared LIBS and VNIR spectroscopy, finding that LIBS performed better for SIC estimation, while VNIR spectroscopy yielded superior results for SOC prediction. The integration of VNIR and LIBS, however, did not consistently improve the accuracy of SOC prediction, although it showed an acceptable performance given the study’s challenging conditions. Regarding sensor performance, Xu et al. [213] observed that MIR sensors outperformed other sensors in estimating the SOC content, followed by VNIR, LIBS, and pXRF sensors. Similar findings were reported by Hutengs et al. [22] and Soriano-Disla et al. [65], with handheld MIR spectroscopy demonstrating a higher calibration accuracy compared to those from VNIR measurements. Teixeira et al. [214] evaluated three proximal sensors (pXRF, VNIR, and NixProTM) for soil property prediction, including SOM. Combining data from all sensors yielded optimal results, but pXRF alone performed comparably. Soil-order-specific models improved predictions for Ultisols (R² > 0.90), and categorical predictions achieved 100% accuracy for some properties [214]. Similarly, Mahmood et al. [215] found that combining data from the VNIR spectrometer and EM38 sensor significantly enhanced the accuracy of soil property predictions. The use of PLSR proved effective in handling the complexity of predictor variables from both sensors, resulting in improved soil sensing outcomes. Reeves et al. [216] emphasized the advantages of near-infrared diffuse reflectance, especially in accounting for particle size and moisture variations, while MIR spectroscopy showed promise in samples with a wide range of SOM content. In the field of remote sensing, different studies have highlighted the varying accuracy levels of different sensing techniques in SOC prediction. Gomez et al. [32] found that remotely sensed Hyperion data provided less accurate SOC predictions compared to data collected using the AgriSpec portable spectrometer. Conversely, Biney et al. (2021) [117] demonstrated the potential of small unmanned aircraft system (UAS) imagery combined with spectral indices, outperforming Sentinel-2 and proximal soil-sensing approaches in SOC estimation. Žížala et al. [217] found that, while multispectral remote sensing data from Sentinel-2, Landsat-8, and PlanetScope showed lower prediction accuracy compared to hyperspectral data, they still exhibited a similar performance, making them cost-effective alternatives for local-scale SOC mapping. Gholizadeh et al. [218] showed that the soil color reflectance measured with a digital camera yielded superior results in predicting the SOC content compared to other techniques. The results from these studies showcase the wide range of capabilities and potential offered by spectroscopic and remote sensing methods for SOC estimation. It emphasizes the importance of carefully considering study objectives and environmental factors to choose the most appropriate approach.

6.4. Error Sources and Improvement Strategies

The in situ estimation of the SOC content using portable/handheld sensing devices is a valuable approach, but it comes with several challenges that require careful consideration to ensure accurate and reliable results. The variation in the spectral measurements of soil samples, along with differences in sampling locations, can lead to errors in estimating the SOC content using in situ spectrometry [219]. The sensors employed for SOC quantification are susceptible to capturing noise from soil properties, including moisture and roughness, which can impact the accuracy of the measurements [57,220,221,222]. To account for the non-linear relationship between the spectral reflectance and moisture content, specific calibrations for moisture content are necessary [223]. At 14% median gravimetric moisture content, Greenberg et al. [179] observed a decreased accuracy, with VNIR showing less decline compared to MIR. Seidel et al. [224] investigated the influence of the soil moisture content on VNIR and MIR portable instruments, revealing texture-dependent distortion in MIR spectra that affects predictive accuracy. Their study demonstrated the superior performance of MIR models on air-dried samples, while VNIR excelled under uniform moisture conditions. Combining VNIR and MIR data produced highly accurate predictions for SOC (RMSE = 0.22–0.27%), essential for estimating soil properties amidst a range of moisture contents. Soil surface roughness, influenced by minerals like large soil aggregates, Fe oxides, and calcium carbonates, also plays a role in the accuracy of spectral reflectance [225]. Soil warming affects the stability and decomposition of SOC, with micro-aggregates being more sensitive to temperature changes [226]. Understanding the distribution of Fe-bound organic C in soil aggregates based on their size can enhance the accuracy and reliability of in situ SOC measurements. When using VNIR spectroscopy, the presence of vegetation residue on the soil surface can introduce inaccuracies in the estimation process. The spectral response of green leaves and straw can resemble that of SOC, leading to a significant overestimation of SOC prediction, with errors of up to 200% if left uncorrected [227]. Soil moisture, texture, and SOM content are known to interfere with the accuracy of pXRF spectrometry for SOC prediction [228]. Additionally, the variability in soil properties with depth poses challenges in accurately estimating SOC content using field portable sensing devices, as it can affect spectral reflectance and introduce potential inaccuracies in the measurements of SOC contents [168]. Many studies [59,168,229,230,231] have recognized the importance of incorporating soil depth as a factor to enhance the accuracy of SOC models. The specific location of soil samples can affect the accuracy of spectral models in SOC prediction, as soil properties and environmental factors vary across different regions [232]. The precision of SOC prediction using VNIR spectroscopy is closely tied to the calibration sample set, with a larger calibration set leading to improved results [176]. Similarly, selecting an appropriate RoI size in hyperspectral imaging plays a crucial role in SOC prediction accuracy, with larger RoIs yielding better estimations [233]. However, further research is needed to fully understand the implications of varying the size of the study area and the density of soil sampling on SOC prediction using VNIR spectroscopy [234]. In addition to variability in soil properties and environmental conditions, the accuracy of SOC prediction using in situ methods is hindered by technological constraints. For instance, handheld devices or probes used in the field may not offer the same sensitivity or resolution as laboratory instruments, thereby compromising the quality and reliability of the collected data [31,179]. This discrepancy can lead to less precise measurements and potentially skewed interpretations of SOC content in soil samples taken in situ. Therefore, improving the capabilities and calibration of field-deployed instruments is essential to enhance the accuracy and consistency of SOC estimations directly in the field.

Researchers have actively sought ways to enhance the accuracy of SOC prediction by addressing challenges such as soil moisture, soil roughness, particle size, surface topography, and soil color. Several studies have successfully employed correction methods to mitigate the errors arising from variations in soil moisture content. Among these methods, external parameter orthogonalization (EPO) has emerged as a popular choice for mitigating soil moisture-related inaccuracies [52,222,235]. Combining EPO spectral transformation with Bayesian modeling techniques, Veum et al. [236] showcased the power of overcoming environmental factors and in situ data collection limitations when utilizing diffuse reflectance spectroscopy data. In certain studies, direct standardization (DS) has been employed as a technique to mitigate the influence of environmental factors, such as the soil water content and surface roughness, on field spectral measurements [237,238]. DS and EPO are effective strategies for mitigating the effects of soil water content on VNIR spectra, with EPO demonstrating a superior performance in SOC predictions with less samples [239]. Furthermore, EPO and OSC models exhibited a better prediction accuracy compared to the DS corrected models [240]. Different algorithms may yield contrasting results depending on the moisture content present in the soil [241]. Denis et al. [242] emphasized soil roughness’s significant influence on the accuracy of the SOC content using VNIR and SWIR spectroscopy, introducing a novel shadow correction method that improves SOC accuracy by 27% in field data and 25% in airborne data, albeit marginally outperforming simpler spectral treatments in mitigating roughness-induced variability. Variations in the particle size within soil samples can introduce distortions in the spectral baseline, known as multiplicative scatter [100,243]. To address this issue, MSC preprocessing technique is employed. This method effectively corrects the tilting in the spectral baseline caused by particle size differences, resulting in an improved accuracy and reliability of spectral data [244,245]. The utilization of the PoLiS optical setup, coupled with VNIR spectroscopy, proved to be a valuable tool in mitigating the multi-scattering effect induced by soil particles and successfully generated absorbance signals that exhibited a stronger linear relationship with SOC [246]. SNV and detrending offer effective solutions for addressing challenges related to particle size, scatter, and multi-collinearity in diffuse reflectance spectrometry, thereby improving the interpretation of near-infrared spectra [247]. Shahrayini et al. (2022) [230] evaluated the effectiveness of VNIR spectroscopy in estimating SOC across multiple depths (0–15, 15–40, 40–60, and 60–80 cm) using four modeling algorithms. The results indicated that the RF model performed the best, particularly without preprocessing, except for the depth range of 40–60 cm, where SG first deviation preprocessing achieved the highest accuracy. Duro et al. [82] introduced ΔI+ as an empirical topographic correction method, enhancing SOC predictions from hyperspectral imaging (HSI) by considering slope, aspect, and wavelength interactions. It addresses surface roughness effects on HSI spectra, improving SOC predictions particularly on non-flat surfaces with slope angles over 30°, making it possible to analyze soil properties on rough soil samples [82]. These studies collectively suggest that the utilization of advanced preprocessing techniques holds promise for enhancing the accuracy of in situ SOC prediction across diverse environmental conditions.

6.5. Case Study: Using Portable Hyperspectral Imaging for SOC and Nitrogen Estimation

Background: In a pioneering study [248] conducted in Taita Taveta County, Kenya, researchers investigated the efficacy of a mobile hyperspectral camera operating in the VNIR wavelength range for estimating SOC and nitrogen (N) content. This technology aimed to support the efficient monitoring of soil properties across diverse land-use types and altitudinal gradients.

Methodology: The study involved collecting 191 soil samples from five distinct land use categories: agroforestry, cropland, natural forest, shrubland, and sisal estate. Variability in spatial location, altitude, and land cover types was deliberately integrated into the sampling strategy. This approach naturally induced differences in soil characteristics such as texture, color, and C content among the samples collected. In a controlled laboratory setting, the Specim IQ hyperspectral camera captured high-resolution spectral images of the soil samples. Ground truth data on C and N contents were meticulously obtained using the Leco combustion analysis method, serving as reliable benchmarks for both model training and validation purposes.

Analysis Approach: Advanced machine learning techniques were utilized to analyze the spectral data and estimate SOC and N contents. The study centered on automating the selection of relevant wavelengths and assessing prediction uncertainty to improve model accuracy and reliability. Five distinct methods—Lasso Regression (LR), GPR, RF, PLSR, and Convolutional Neural Network (CNN)—were evaluated. The evaluation emphasized performance metrics including cross-validated R² values and RMSE, employing ten-fold cross-validation for robust assessment.

Results: The comprehensive analysis demonstrated a strong performance across all evaluated machine learning methods. LR achieved a cross-validated R² of approximately 0.8, with an average RMSE of 1.02 and an RPD of 2.25. These results underscore the reliability and effectiveness of mobile hyperspectral imaging for accurately estimating SOC and N contents across diverse soil and land use conditions.

Conclusion: This case study underscores the transformative potential of mobile hyperspectral imaging in advancing soil science and agricultural research. By integrating cutting-edge imaging technology with sophisticated data analytics, researchers can enhance productivity, sustainability, and resilience in agricultural systems. This non-invasive, rapid, and cost-effective technology enables the real-time monitoring of SOC and nutrient dynamics, facilitating informed decision-making processes in agriculture, forestry, and environmental conservation sectors. The findings emphasize the critical role of technological innovation in addressing global food security challenges and advancing environmental sustainability.

6.6. Role of Handheld Devices in C Farming

C farming, encompassing a range of practices such as conservation agriculture, cover cropping, crop rotation, intercropping, and agroforestry, aims to sequester C in soil and biomass, thereby offering additional income-generating opportunities for land managers [249]. The optimization of these strategies and the maximization of C sequestration potential heavily rely on accurate data regarding soil properties, including pH, SOC, nitrogen (N), phosphorus (P), and potassium (K) [99,143]. Handheld devices facilitate the efficient implementation of C farming practices by providing real-time measurements of these crucial properties directly in the field [37,79,250]. Additionally, they offer a convenient means of monitoring changes in soil composition, compaction, structural degradation, and biological processes over time [17,52,251]. By enabling the on-site detection of pH and K contents, handheld devices streamline soil testing processes, providing accurate data essential for informed fertilizer application decisions [250]. This capability allows farmers and land managers to customize fertilizer application rates, thereby minimizing input costs and mitigating environmental impacts such as nutrient runoff and soil degradation. Moreover, handheld spectroradiometers prove particularly useful for determining SOC concentration in low fertility soils [252]. Their portability and ease of use make them well-suited for diverse agricultural landscapes and land management contexts characteristic of C farming initiatives, whether in agroforestry systems, cover cropping rotations, or regenerative grazing practices [37,200]. By quickly assessing soil nutrient status on-site, stakeholders can make informed decisions supporting soil health, biodiversity, and C sequestration [37,253]. Research validates the effectiveness of handheld hyperspectral cameras in predicting SOC and N contents, indicating potential improvements in C farming practices, especially in higher elevations [248]. Additionally, portable gamma spectrometers like the Medusa MS-700 show promise for the rapid spatial analysis of TC and TN in agricultural fields, highlighting its role as a valuable tool for informing targeted regenerative actions in C farming initiatives [185]. These devices’ sensitivity to soil characteristics and accuracy in estimating TC and TN underscore their significance in advancing soil health and C sequestration efforts. Moreover, SOC assessment using handheld devices is integral to developing policies related to C farming, land tenure, and environmental regulations, as they provide essential information for quantifying C stocks and assessing the potential impacts of land-use changes on C sequestration [254].

7. Recent Advances

7.1. Wireless Sensing Technologies

Wireless sensing technologies, such as Wi-Fi signals and smartphones, have gained prominence for in situ measurements of soil parameters, with several studies focusing on soil moisture measurements [42,255,256]. Advancements in soil moisture sensing are crucial for improving the accuracy of SOC content estimates and developing robust predictive models [257]. Ding and Chandra developed Strobe, a system that utilizes existing Wi-Fi bands for soil moisture and EC sensing through RF propagation [42]. Josephson et al. created a cost-effective soil moisture sensing system using wireless backscatter tags and an ultra-wideband RF transceiver, overcoming obstacles like high sensor costs and difficult deployment [255]. Khan and Shahzad introduced CoMEt (Contactless Moisture Estimation), a novel approach using radio frequency signals to assess soil moisture at varying depths, demonstrating exceptional accuracy [256]. Smartphone-based digital images provide a rapid and cost-effective alternative for SOC analysis [210,258]. Gozukara et al. [259] found that lower illumination levels, particularly at 300 lx, led to an improved prediction accuracy for the SOC content using various color parameters. Smartphone-mediated soil analysis utilizing the smartphone camera as a reflectometer and machine learning algorithms offers accurate SOC estimations, with the accuracy ranging from 85% to 97%, transforming soil testing in agriculture [260]. Additionally, researchers have proposed a smartphone-based method for rapid SOC prediction, utilizing image segmentation and machine learning with a strong prediction accuracy of R² = 0.88 and RMSE = 0.28% [261]. Considering inter-smartphone variability and environmental factors such as temperature is crucial in smartphone-mediated soil analysis. However, this method still holds promise as an effective screening tool for assessing soil nutrient concentrations, ultimately contributing to improved crop outcomes and SAPs [262]. These innovative approaches, leveraging wireless sensing technologies and smartphone-based methods, hold great potential for advancing precision agriculture and sustainable practices by providing efficient and cost-effective solutions for in situ soil parameter measurements and SOC analysis.

7.2. Commercial Devices

Commercial devices have emerged as valuable tools for in situ SOC estimation and soil health assessment. Yardstick, for instance, utilizes VNIR spectroscopy to develop handheld soil probe that provides non-destructive and cost-effective measurements of SOC [263]. The Teralytic Probe goes a step further, offering wireless sensing capabilities to measure multiple soil parameters, including soil respiration, moisture, pH, temperature, and nutrients [264]. AgroCares Scanner, on the other hand, employs VNIR spectroscopy for on-site soil sample analysis, delivering quick results for sustainable agriculture decision-making [265]. The veris MSP3 soil scanner stands out with its unique ability to map three crucial soil characteristics, namely soil pH, SOC, and the EC, while being driven through the field [266]. CropX Soil Sensor provides real-time data on soil moisture, temperature, and EC, including estimates of SOM, optimizing irrigation and nutrient management [267]. Lastly, the FieldScout TDR 350 handheld device utilizes time-domain reflectometry for soil moisture and EC measurements, widely applied for moisture estimation in SOC calibrations and soil health assessments [268]. These commercial devices offer practicality, efficiency, and user-friendly features, paving the way for improved precision agriculture. As their adoption increases, they are expected to revolutionize in situ SOC monitoring and encourage the scaling of SAPs.

8. Conclusions

The in situ estimation of the SOC content using portable and handheld devices leverages spectroscopic and remote sensing techniques, offering non-destructive and efficient alternatives to traditional ex situ measurements. Remote sensing, though indirect, complements in situ spectral measurements by enhancing SOC estimation accuracy. Spectroscopic methods like VNIR, MIR, LIBS, and INS each have their strengths and challenges. VNIR is effective in handling moisture and texture variations, while MIR performs well with different SOM contents and concurrently assessing other soil properties, such as particle size variations. LIBS showed promising results in SIC detection, and INS has potential for scanning larger areas.

Among the 32 studies reviewed, 14 demonstrated strong accuracy with an R² of 0.80 or higher, underscoring the potential of handheld in situ methods for precise SOC assessment. Despite their advantages in time and cost efficiency, the accuracy of these approaches can vary due to factors such as soil properties and calibration methods. Innovative preprocessing techniques like OSC, EPO, DS, and FOD address environmental influences on spectral reflectance, thereby enhancing SOC prediction accuracy. Calibration methods such as MLR, PLSR, SVM, and RF establish robust relationships between spectral reflectance, soil properties, and SOC content. PLSR is widely used across different soil depths, while SVM shows efficacy in surface soils.

Researchers, land managers, and policymakers must consider factors like calibration sample size and variability, soil type, and climate factors, along with specific environmental conditions such as soil moisture contents and vegetation cover when utilizing handheld devices for SOC estimation. Implementing rigorous validation processes and advancing preprocessing techniques are vital for mitigating challenges and ensuring accurate measurements.

Integrating handheld devices into soil analysis processes has the potential to enhance C farming initiatives. By enabling faster, more efficient, and accurate assessments of soil properties, these devices empower stakeholders to make informed land management decisions. Moreover, they play a crucial role in reinforcing C sequestration efforts, thereby advancing environmental sustainability.

Future research should expand the geographical application of portable, in situ methods to diverse ecological settings and soil types worldwide. Enhancing multi-sensor data fusion and developing user-friendly analytical tools will be key to improving the practical utility and scalability of these approaches in soil and environmental sciences. These advancements promise to enhance SOC monitoring capabilities, supporting sustainable land management practices and climate change mitigation efforts globally.