3.1.

Digital Soil Mapping (DSM)

The orology DSM was initially conceived as the act of generating and fusing spatial soil information through the application of field and laboratory observational methods alongside spatial and non-spatial soil inference systems [20]. The DSM method gained increased popularity for providing detailed and up-to-date data about soils [21];[22]; [23]. DSM presents itself as a flexible, quantitative, and precise method; nevertheless, it is essential to underscore specific significant points. The applicability of DSM models to diverse regions with distinct characteristics and profiles is primarily complex, as the prediction models are developed for each particular area. Furthermore, unlike conventional soil mapping products, DSM maps are designed with specific objectives rather than for general use. A crucial point to consider is the uncertainty parameter in the DSM products since the created models are approximations of reality. Some sources of uncertainty include the covariate accuracy and the model structure [24]. Random and systematic errors found in observations and covariates also represent a significant factor in the uncertainty [25]. The quantification of uncertainty in several models applied within DSM, notably kriging models such as regular kriging and kriging with external drift, has attracted substantial attention [26]. There is a notable uptick in the utilization of machine learning algorithms (ML) within the DSM, which are employed to forecast conditional quantiles [27]. Within uncertainty quantification, DSM utilizes prediction intervals to estimate the likely range of values for an actual measurement. A commonly used threshold in this context is the 90% prediction interval (PI) [21], which denotes the range of values where one can expect to find a new measurement nine times out of ten [28]. The frameworks proposed for assessing uncertainty in DSM are organized into three main classifications: spatial prediction frameworks, probabilistic frameworks, and frameworks that synthesize both spatial characteristics and probabilistic aspects to yield spatially explicit realizations [29].

3.2.

Remote Sensing (RS)

A plethora of research studies have been conducted to explore the potential of RS in soil mapping surveys. These investigations have consistently indicated a significant increase in the effectiveness of conventional methods when RS data is incorporated [20].

Spatial models rely on accurate data regarding the continuous variations in soil properties. To ensure the utmost precision, these models require as much detail as possible when utilizing such data as input [30]. One of RS’s major drawbacks arises at this point, as the predictions are deemed valid only for the very top layer of soils. It is essential to underscore an additional condition when analyzing RS data, which pertains to making corrections for atmospheric, geometric, radiometric, and topographic effects. An example that is commonly encountered is the interference caused by vegetation, which poses challenges to investigations. Once these corrections have been adequately addressed, geostatistical techniques can be utilized for spatial interpolation, enabling the mapping of spatial patterns in regions where soil data is limited.

The reliability of thematic mapping classification achieved via RS techniques is a crucial determinant of its relevance and precision for multiple applications. No universally established standard for accuracy in thematic mapping is produced through RS methods. Nonetheless, at least 85% accuracy in correct classifications is frequently recognized as the aspirational objective target in thematic mapping [31]. The integration of the error matrix technique, commonly referred to as the confusion matrix, is employed to address the challenge of accuracy in computer-generated mapping [32]. Functioning as a critical, descriptive statistic, the confusion matrix allows for evaluating classification results against established ground truth data. However, constructing and interpreting the confusion matrix is fraught with difficulties. These difficulties are typically rooted in assumptions that are either unattainable or biased [31].

3.3.

Proximal Sensing (PS)

PS alludes to a technological advancement which enables the measurement of soil characteristics from a distance of less than 2 meters above the soil surface [33]. PS is predominantly associated with high-resolution soil mapping, usually at resolutions less than 20m. Recognizing the concept of applying PS in soil science is of utmost significance. While measurements’ precision may be lower than conventional mapping techniques, PS facilitates the collection of larger volumes of spatial data at a lower cost and through less intricate and more straightforward methods [33]. Table 2 presents the suitability of each technology, while below is a description of three primary PS techniques used.

Table 2.

Suitable PSS for acquiring soil information.

3.3.1.

The Ground-penetrating radar (GPR)

GPR is a near-surface geophysical non-destructive technique which creates high-resolution images of the Earth’s dielectric properties. The fundamental concept behind this principle is that high-frequency electromagnetic waves have the potential to reflect or refract when they encounter subsurface features that exhibit changes in their electrical properties [34]. GPR makes transmitting and reflecting high-frequency electromagnetic waves within the Earth’s subsurface possible. The accuracy and level of detail attained through GPR are determined by selecting different antenna frequencies. Higher frequencies contribute to enhancing the resolution of the images generated. GPR measures the time it takes for a direct ground wave to move from the source to the receiver antenna across the uppermost layer of the ground [35]. Figure 1 presents a simplified schematic design of a GPR.

Figure 1.

Schematic design of a GPR.

3.3.2.

Electrical resistivity (ER)

Soil ER refers to the capacity of soil to resist the flow of electrical current. To determine the apparent ER, various methods utilize Ohm’s law in conjunction with the measured injected current, the potential difference, and a geometric factor. A total of four electrodes are required to perform resistance measurements. Two of these electrodes are designated as current electrodes (A and B) and are responsible for injecting the supply current. The other two electrodes, known as potential electrodes (M and N), record the potential difference [36]. The calculation of the potential difference in the following formula determines ΔV:

AM, BM, AN, and BN refer to the geometrical distance between the electrodes A and M, B and M, A and N, and B and N, respectively. The formula then expresses the ER:

The geometric coefficient K depends on the arrangement of electrodes A, B, M, and N.

ER varies widely, from 1Ω for saline soils to 105Ω per meter for dry soil overlying crystalline rocks. Figure 2 illustrates the typical ranges of resistivity values.

Figure 2.

ER region of earth materials (after Palacky, 1987) [36].

3.4.

Soil Spectroscopy

The application of soil spectroscopy has become prevalent as a quantitative approach for estimating diverse soil properties. Spectroscopy is a quantitative method used in physical and analytical chemistry based on the principle that atoms and molecules absorb radiation in distinct wavelengths, resulting in their unique signature spectra [39]. Spectroscopic methods, such as mass spectroscopy, visible, near-infrared, and mid-infrared spectroscopy, are being acknowledged as potential substitutes for traditional laboratory and field approaches. They aim to enhance and, in some cases, even supplant our comprehension of soils.

Imaging spectroscopy employs airborne or satellite-based hyperspectral sensors to capture the pertinent spectral data spatially. Spectroscopy offers several advantages; however, there are certain drawbacks associated with this technique. One limitation is that it solely relies on reflectance measurements from bare surfaces to assess soil properties.

Consequently, it cannot estimate vertical gradients in SOC within the topsoil. The method is classified as less accurate than conventional methods, such as wet oxidation and dry combustion [9].

3.5.

Soil Texture

Vaudour et al. (2019) emphasized the advantages presented by the Sentinel mission in evaluating key top-soil attributes. A multitude of methodologies advocates for the application of chemometrics techniques or the identification of specific absorption features to compute soil properties using the reflectance regions of visible and near-infrared (VIS-NIR, 400- 1200nm) and shortwave infrared (SWIR, 1200-2500nm) [40]. Distinctive absorption features can differentiate soils rich in clay from those rich in quartz. Typically, clay minerals are identified by a hydroxyl absorption at 2200nm, whereas thermal bands displaying a reflectance peak corresponding to silica in the range of 8000nm-9500nm indicate the presence of quartz [41].

The correlation between surface soil properties and remotely sensed spectral data at various wavelengths has been explored using various laboratory statistical methods. The two most prominent techniques, multiple linear regression (MLR) and partial least-square regression (PLSR), are commonly used.

3.6.

Soil Moisture

Measurement of soil moisture content on the near-surface can be achieved using optical and thermal infrared RS, as well as passive and active microwave RS techniques. Although imaging spectroscopy indices to calculate surface soil moisture content have been adopted using the reflectance in the SWIR region, the accuracy of the results was constrained, primarily due to shallow soil penetration and the interference caused by cloud cover.

The Microwave RS method is a supplementary technique for predicting soil moisture. It involves measuring the contrast in dielectric properties between dry soil and water, which is derived from the reflected data.

The Synthetic Aperture Radar (SAR) is the most extensively employed active microwave configuration for imaging purposes. Despite weather conditions, SAR systems offer high-resolution images suitable for day and night operations. These systems can achieve resolutions in the tens of meters range, covering swath widths up to 500 kilometres, with the best performance observed in space-based systems. SAR technology is widely applied in geoscience, climate change research, 2-D and 3-D mapping, and even 4-D mapping. The estimation of soil moisture has been facilitated by the utilization of various SAR satellites, including the European Remote Sensing (ERS) [42], the Environmental Satellite/Advanced SAR (ENVISAT/ASAR)[43], the RADARSAT [44], the Advanced Land Observing Satellite/Phased Array L-band SAR(ALOS/PALSAR) [45] and the Terra SAR-X platforms [46].

3.7.

Soil Organic Carbon (SOC)

The colour of the soil serves as an initial indicator for estimating the levels of soil SOC. Soils with darker hues tend to contain higher quantities of SOM. As a result, the assessment of SOM using imaging spectroscopy proves to be an excellent approach, given its strong correlation with soil colour [8]. VIS-NIR spectroscopy is a valuable method that enables the estimation of soil’s physical, chemical, and biological properties based on its reflectance characteristics. Notably, the SWIR (1300-2500nm) and NIR (700-1300nm) regions have been identified as particularly sensitive to SOC levels, with reflectance decreasing as SOC content increases [47].

The endeavor to map SOC has been facilitated by the implementation of sensors, such as the Sentinel series, which have strategically integrated bands that are capable of detecting SOC [6]. The European Space Agency’s (ESA) Sentinel-3 Ocean and Land Colour Instrument (OLCI) is a cutting-edge technology that has recently joined the Sentinel series. With an extended sweep width of 1270km, the Sentinel-3 can collect data within the electromagnetic spectrum’s VIS to NIR wavelength range, specifically from 400 to 1020nm. These VIS-NIR wavelengths are pivotal in offering reflectance information on SOC and are widely acknowledged as the most sensitive region for accurately assessing SOC concentrations [48].

A diverse set of multivariate statistical techniques and ML algorithms have been employed to map SOC. The Principal Component Analysis [49], Regression Trees [50], Support Vector Machines [51]; [52], and Partial Least Squares Regression [53], are commonly found in the relevant literature for SOC estimations.

3.8.

Soil Salinity

Two methods can be used to identify soil salinity: direct detection on exposed soil using remotely sensed data or indirect detection through the analysis of vegetation type, which is influenced by the presence of salts in the soil.

Salt mineralogy determines concentrations of salt through identified absorption bands in the electromagnetic spectrum [54]. Spectral regions that identify soil salinity levels include the visible (0.55–0.77Am), near-infrared (0.9–1.3Am), and middle-infrared (1.94–2.15, 2.15–2.3, 2.33–2.4Am) bands. Six RS indices (Table 6) can be computed to assess salinity, which can then be used to map areas with salt-affected soils.

Table 6.

Indices to analyze soil salinity and vegetation indices using Landsat 4-7 sensors [11].

The application of RS is not exempt from difficulties, as the ever-changing process of surface salinity can lead to inaccuracies in identification outcomes. Salt detection may be impeded by vegetation or other surface attributes, causing spectral confusion with salt reflectance. Using space-borne SAR systems represents an approach to mapping areas impacted by salinization. RADARSAT images [55] and Phased Array Type L–L-Band synthetic Aperture Radar (PALSAR) data [56] are commonly employed in the estimation of soil salinity. The justification for employing microwave RS to analyze the salinity factor is rooted in the dielectric characteristics of the soil, as salinity plays a crucial role in electric conductivity. Building upon this rationale, the ML regression model emerges as a primary method for addressing soil salinity, while the random forest (RF) models are frequently employed in ensemble learning approaches [57].

3.9.

Iron Content

Employing multispectral or imaging spectrometer images, it becomes feasible to measure the absorption features related to iron oxide and iron hydroxides [8]. Iron exhibits three wide diagnostic absorption bands in the VIS/NIR in its trivalent state. These bands can be the main constituents (in iron oxides) or impurities (in iron smectites). In the case of hematite, these absorption bands are located at 550nm, 630nm, and 860nm, while for goethite, they occur at 480nm, 650nm, and 920nm. The reflectance spectra are predominantly influenced by the first absorption band (Fe-VIS:~500nm), and the third absorption band (Fe-NIR:~900nm) dominate the reflectance spectra [13].

A method to generate the redness index for Hematite content, derived from Thematic Mapper of Landsat, is the following formula: Redness Index (RI) = R2/(B*G3), while from Sentinel-2 general formulas to assess Ferric Oxides and Ferrous Iron is the B11(1600:1700)/B08(760:860); and B12(2145:2185)/B08/(760:860) +B03(520:600)/B04(630:690) respectively.