Assessing the Perspectives of Ground Penetrating Radar for Precision Farming

Abstract

The United Nations 2030 Agenda for Sustainable Development highlighted the importance of adopting sustainable agricultural practices to mitigate the threat posed by climate change to food systems around the world, to provide wise water management and to restore degraded lands. At the same time, it suggested the benefits and advantages brought by the use of near-surface geophysical measurements to assist precision farming, in particular providing information on soil variability at both vertical and horizontal scales. Among such survey methodologies, Ground Penetrating Radar has demonstrated its effectiveness in soil characterisation as a consequence of its sensitivity to variations in soil electrical properties and of its additional capability of investigating subsurface stratification. The aim of this contribution is to provide a comprehensive review of the current use of the GPR technique within the domain of precision irrigation, and specifically of its capacity to provide detailed information on the within-field spatial variability of the textural, structural and hydrological soil properties, which are needed to optimize irrigation management, adopting a variable-rate approach to preserve water resources while maintaining or improving crop yields and their quality. For each soil property, the review analyses the commonly adopted operational and data processing approaches, highlighting advantages and limitations.

1. Introduction

Climate change and population growth exert great pressure on the environment and natural resources. The World Population Prospects [1] estimates that the world population will reach approximately 9.7 billion by 2050, leading to a growing demand for resources and food. In order to face this increasing demand, the Food and Agricultural Organization (FAO) forecasts a more than 50% increase in irrigated food production by 2050 [2]. Given the limited availability of arable land, this food production will determine an agricultural intensification, that means an increasing use of water, fertilizers, pesticides, energy and other inputs. Focussing on water resources, the water imbalance between water supply and demand for agriculture is destined to grow in the coming years, leading to an increase in water scarcity for this sector. Along with this aspect, the rapid urbanisation processes that are currently undergoing in some regions of the world must be taken into account, which are not only affecting how humans consume natural resources, but are also transforming land use [3,4].

In this context, the development and adoption of sustainable production processes are needed to promote an efficient use of natural resources and of water in particular.

Precision Agriculture (PA) is proposed as a solution to achieve a sustainable agriculture. PA is an ensemble of management strategies that, through advanced technologies, provides tools for a spatially variable and time-variant monitoring, deciding and application of agronomic inputs. The aim of PA is increasing the efficiency in the use of inputs, thus reducing the environmental impacts of agriculture production in order to preserve natural resources for the future. Contemporaneously, the achieved improvement in quality and quantity of crop yields and the cost reduction due to inputs saving should guarantee the economic sustainability of the PA solutions proposed to farmers. Recently, the authors of [5] described PA as the ‘use of technologies that integrate sensors, information systems, enhanced machinery, and informed management to improve production by accounting for dynamics within sustainable agricultural systems’.

Specifically, Precision Irrigation (PI) is aimed at reducing irrigation use without creating water stress for crops, by applying water at the right time, i.e., only when it is required, with the right amount, i.e., depending on the soil water depletion, and at the right place, i.e., without creating over-irrigated and under-irrigated areas in the field, thus improving yields [6]. As for water, the variable rate application of the other agronomic inputs (fertilizers, pesticides) leads to a maximization of their use efficiency of use, while reducing their losses in the environment.

The identification of optimum irrigation scheduling must take into account the space–time variability of the soil–crop system, also depending on site-specific agro-meteorological conditions affecting crop water requirements [7]. Therefore, compared to the uniform rate adopted in conventional agriculture, the variable rate and time-variant water application is based on the operators’ ability to monitor soil and crop properties and dynamics. In this context, soil and its properties play a fundamental role, because the rooted soil volume constitutes the reservoir from which crop roots extract water for crop development.

Due to the necessity of gathering a deep understanding of soil hydrological behaviour, the last decades have seen a rise of the adoption of remote sensing and geophysical sensors as data sources to support decision-making tools for site-specific irrigation water management. Therefore, geophysical measurements have now become a well-known method for soil investigation at the field scale and farm scales, thanks to their non-invasive, non-destructive, rapid and cost-effective characteristics. Their main advantage is the possibility of investigating agricultural surfaces without disturbing and consequently affecting the soil dynamics [8,9]. In addition, geophysical survey can provide extended insights on both the spatial and the temporal domains, thus offering a bridge between remote sensing observations and point-based measurements. Finally, the retrieved information could also benefit the development of accurate and robust prediction models to analyse soil properties [10,11,12].

Among one of the many challenges in implementing PI is to obtain an accurate characterisation of the spatial pattern of soil electrical conductivity, a parameter known to be significantly affected by a number of key soil features, including hydrological, structural and textural features [13]. These include, but are not limited to, soil water and organic matter content, soil density and porosity and particle size distribution [14,15]. Complex site-specific behaviours of soil properties affect soil water availability and plant rooting, hence determining crop growth and yield [16,17,18] Gathering this information has allowed PA to develop from a conceptual idea to a tool for addressing the issue of agricultural sustainability [19,20].

Within the geophysical techniques landscape, Electrical Resistivity (ERT), Electromagnetic Induction (EMI) and Ground Penetrating Radar (GPR) represent the most widely adopted geophysical methods employed for agricultural applications, due to a number of advantages, the main being their high sensitivity to variations in the electrical properties of the soil, i.e., dielectric permittivity and/or electrical conductivity, though providing only an indirect measure of such soil properties. Among them, GPR has been relevantly leveraged in the precision farming domain, as the method is sensitive to both permittivity and conductivity properties, displaying the potential of increasing the collectable information [21,22].

Considering its operating principles, GPR has always shown great promise in scenarios where water content, material porosity and stratigraphic discontinuities represent the predominantly investigated factors, such as in hydrogeological studies, near-surface aquifer delineation, groundwater characterisation and soil hydrology studies. Geological monitoring and surface mining are two additional environments where information on the presence of water and percentage of clay within the stratigraphy acquired through GPR can assist landslide characterisation and the detection of undesired geologic bodies.

GPR is also promising for agricultural applications, where soil moisture, transport and solute issues are of great interest due to their direct impact on crop yield, irrigation management and environmental effects of land usage. In particular, due to GPR’s ability to spatially map local soil attributes affecting crop performance, the methodology has been extensively researched and has emerged as possibly the most suitable geophysical technique within the domain.

The present review focuses on the main physico-chemical soil properties relevant for PA procedures, with a particular attention to Precision Irrigation (PI), and on the available advanced geophysical tools to investigate them. In particular, the contribution starts with a detailed analysis of the main soil parameters that directly or indirectly affect the hydrological behaviour of soils, highlighting the research questions that need to be addressed in order to achieve a sustainable water use (Section 2). Then, Section 3 reviews the role that geophysical measurements, in particular Ground Penetrating Radar technique, have played and are playing within the PI domain and the reasons that have secured them a leading role for agricultural soil characterisation. Starting from the core soil parameters previously identified, Section 4 provides a broad review of the different research approaches, methods and processing strategies that have been adopted within the PI domain, highlighting principal characteristics and advantages and limitations. This section, that represents the more relevant contribution of this review, portraits the current landscape of the application of GPR for PI purposes. The paper ends by discussing technological and operational challenges, as well as prospective opportunities related to the use of GPR in soil studies.

2. Physico-Chemical and Hydrological Soil Properties Relevant for PI

Soils are mainly classified on the basis of texture, which refers to the percentage of sand, silt and clay sized particles constituting the mineral fractions of the soil. The soil-type classification based on texture is an important starting point to obtain a first idea of the hydrological behaviour of a soil (i.e., water retention and hydraulic conductivity as a function of the soil water content). However, soil hydraulic properties do not depend only on texture, but also on soil structure and porosity. Indeed, soil structure and porosity in natural conditions are related to textural properties. Soil structure depends on the arrangement of the primary soil particles, creating secondary soil particles (aggregates, clods and pedons) [23]. The formation of these secondary particles depends on the mineralogical composition of soil and on the quality and quantity of soil organic matter. Soils can be classified as structural and non-structural (without aggregates). The latter are mostly coarse and sandy soils with low content of organic matter. Soil structure is a fundamental physical property for agricultural soil, as it is strongly related to the soil porosity (i.e., volume of soil pores over the bulk soil volume, including both soil particles and pores) and pore-size distribution, which play important roles in the hydrological soil processes. The macropores, which are mostly inter-aggregate pores, or even large pore created by dead roots, worms and soil cracks, enable fast infiltration of rain or irrigation water into soil, reducing soil erosion and enabling easier tillage. Conversely, macropores enable fast leaching of nutrients and other solutes, increasing the risk of groundwater pollution. Moreover, macropores enable favourable conditions for aeration and respiration of roots, guaranteeing the soil production functions. Soil organic matter is an important property to ensure the stability of the soil structure, given its resistance against erosion factors such as heavy rains or wind; soils with a high content of organic matter have a stable structure, preserving soil quality [24]. Moreover, soil organic matter increases the soil’s ability to retain water by electrostatic interactions. Soil bulk density, defined as the ratio of the dry weight of soil and its bulk volume (including both soil particles and pores), is an indicator of soil compaction (i.e., reduction in soil porosity). This soil property varies over time (especially in the plugged soil horizon) as a result of agricultural soil management and interactions with agro-meteorological factors. Although rarely measured, soil compaction is an important property of agricultural soils due to its variability over time, connected to soil degradation. Indeed, the reduction in soil porosity and, thus, in bulk density plays a critical role for water and solute transport processes in agricultural soils, as it determines decrease in soil hydraulic conductivity [25], which affects irrigation water infiltration and nutrient uptake, thus increasing runoff and soil erosion [26,27,28,29]. Moreover, soil porosity and bulk density are strongly related to soil structure: as already introduced, a well-developed and stable soil structure is fundamental for productive soils [30,31].

Soil texture, structure and its organic content greatly affect the soil water retention properties due to the interactions between water contained in the pores and the solid phase. The volumetric Soil Water Content (SWC) expresses the relative quantity of water in the soil, as the ratio of water contained in the soil pores and the bulk soil volume [32]; its value is maximum at saturation, when the pore system is completely filled with water. Soil Matric Potential (SMP) expresses the tension of water in the pore system, which is negative due to the attraction between water and the solid phase, becoming null at soil water saturation. The relationship between SMP and SWC is described by the Soil Water Retention Curve (SWRC). The SWRC is typical of a specific soil, and depends on its texture, structure and soil organic matter. The soil–water constants are specific points on the SWRC, with SWCs (or SMPs) corresponding to different levels of water availability to plants. The soil–water constants often used in the irrigation management are Field Capacity (FC), Point of Limited Availability (PLA) and Wilting Point (WP). FC is the maximum SWC (and SMP) available for plants after the drainage of water present in the larger pores following a rain or irrigation (i.e., water content between saturation and FC is not available for plants); PLA is the SWC (and SWP) at which the actual crop transpiration rate starts to be lower than its potential rate, thus leading to a decrease in biomass production. WP is the SWC (and SMP) at which the crop transpiration stops since crops are not able to counteract the tension with which water is retained at the solid matrix (after reaching this point, no further irrigation is effective). Furthermore, Available Water Capacity (AWC) is the range of SWC available to plants in the soil root depth, from WP to FC.

Soil hydraulic conductivity also depends on SWC (and, thus, on SMP), which is a fundamental soil variable to determine the water flow rate in soils. Again, the relationship between SWC (and SMP) and soil hydraulic conductivity is soil dependent. Particularly, saturated hydraulic conductivity (i.e., hydraulic conductivity at saturation) is expected to be higher in coarse soils than in fine soils, but the unsaturated hydraulic conductivity depends mainly on soil texture, structure and organic matter content, as well as on the agronomic operations conducted. Indeed, these factors affect porosity and pore size classes. The presence of skeleton (rock fragments or cobbles with a dimension > 2 mm) in soils deserves a separate discussion, since it modifies the volume available for water in the soil, as well as the soil hydraulic conductivity.

The knowledge of soil hydrological behaviour, that means soil water retention and hydraulic conductivity curves (as a function of SWC/SWP), allows for the accurate description of water flows in agricultural soils by using physical-based agro-hydrological models involving the solution of the Richards’ equation. Through these approaches, it is possible to support the irrigation management and planning at different spatial scales. However, measurements of soil water retention and hydraulic conductivity curves in field or laboratory are very expensive in term of time and economic costs; additionally, they concern single points in soil profiles and are consequently not applicable in real agricultural production systems. As a matter of fact, the hydrological properties of soils vary not only in the horizontal dimensions, but also along the vertical dimension, since soils are generally characterized by an alternation of horizons with different characteristics.

To overcome this fact, soil hydrologists and irrigation specialists very often describe the soil water retention and hydraulic conductivity curves through analytical models (e.g., Brooks and Corey [33], Van Genuchten [34]), and derive their parameters by the knowledge of fundamental soil properties (texture, soil organic matter and, if available, bulk density, porosity, structure, FC and WP), which are more easily measurable and often quantitatively described (at least texture and organic matter content) in the soil profiles associated with regional/national soil maps. These soil properties are inputs for empirical models called Pedo-Transfer Functions (PTFs), which are used to estimate parameters included in the analytical models describing the soil water retention and hydraulic conductivity curves [35,36,37]. In particular, PTFs are generally statistical regression models developed on the basis of large datasets including several hundred measurements of coupled physico-chemical and hydrological soil parameters.

2.1. Reference Values for Physico-Chemical and Hydrological Soil Properties

The range of variability of the main soil properties can be deduced from the wide datasets collected to develop the different PTFs. The USDA Soil Survey Manual [38] refers to the work of [39] to describe the variability of the saturated hydraulic conductivity for the USDA soil texture classes, based on bulk density and particle size distribution (percentage of sand, silt and clay). The authors of [40] illustrate the soil properties variability based on measurements of 1323 soils in the USA. More recently, a review work in 2016 [41] reports the main research that developed PTFs worldwide: based on the largest datasets, the authors of [39,40,42,43,44,45] developed PTFs for soils in the USA (5320 soil samples), while the authors of [46] developed PTFs for soils in Europe (HYPRES database, 4030 soil samples). Additionally, the authors of [47,48] developed PTFs by the neural model ROSETTA, based on a dataset of 2134 soil samples collected in North America and Europe.

Table 1. Soil hydraulic properties variability for the USDA texture classes. Mean values are reported for each textural class, and ±1 standard deviation values are in brackets.

	[39,40]					[43]		[47]
USDA Texture Class	Sample Size	n (cm3/cm3) 1	FC (cm3/cm3) 2	WP (cm3/cm3) 3	Ksat (cm/h) 4	Sample Size	OM (%) 5	Sample Size	n* (cm3/cm3) 6	Ksat (cm/h) 4
Sand	762	0.437	0.091	0.033	21.00	660	0.71	308	0.375	26.78
		(0.374–0.500)	(0.018–0.164)	(0.007–0.059)			(SD 1.06)		(0.320–0.430)
Loamy Sand	338	0.437	0.125	0.055	6.11	198	0.61	201	0.390	4.38
		(0.368–0.506)	(0.060–0.190)	(0.019–0.091)			(SD 1.16)		(0.320–0.460)
Sandy Loam	666	0.453	0.207	0.095	2.59	371	0.71	476	0.387	1.60
		(0.351–0.555)	(0.126–0.288)	(0.031–0.159)			(SD 1.29)		(0.302–0.472)
Loam	383	0.463	0.270	0.117	0.68	203	0.52	242	0.399	0.50
		(0.375–0.551)	(0.195–0.345)	(0.069–0.165)			(SD 0.99)		(0.301–0.497)
Silt Loam	1206	0.501	0.330	0.133	1.32	497	0.58	330	0.439	0.76
		(0.420–0.582)	(0.258–0.402)	(0.078–0.188)			(SD 1.29)		(0.346–0.532)
Silt	-	-	-	-	-	-	-	6	0.489	1.82
									(0.411–0.567)
Sandy Clay Loam	498	0.398	0.255	0.148	0.43	250	0.19	87	0.384	0.55
		(0.332–0.464)	(0.186–0.324)	(0.085–0.211)			(SD 0.34)		(0.323–0.445)
Clay Loam	366	0.464	0.318	0.197	0.23	175	0.10	140	0.442	0.34
		(0.409–0.519)	(0.250–0.386)	(0.115–0.279)			(SD 0.51)		(0.363–0.521)
Silty Clay Loam	689	0.471	0.366	0.208	0.15	209	0.13	172	0.482	0.46
		(0.418–0.524)	(0.304–0.428)	(0.138–0.278)			(SD 0.42)		(0.396–0.568)
Sandy Clay	45	0.430	0.339	0.239	0.12	61	0.38	11	0.385	0.47
		(0.370–0.490)	(0.245–0.433)	(0.162–0.316)			(SD 1.20)		(0.339–0.431)
Silty Clay	127	0.479	0.387	0.250	0.09	-	-	28	0.481	0.40
		(0.425–0.533)	(0.332–0.442)	(0.193–0.307)					(0.401–0.561)
Clay	291	0.475	0.396	0.272	0.06	72	0.38	84	0.459	0.61
		(0.427–0.523)	(0.326–0.466)	(0.208–0.336)			(SD 0.83)		(0.380–0.538)

¹ Porosity; ² Field Capacity (evaluated as SWC at—33 KPa); ³ Wilting Point (evaluated as SWC at—1500 KPa); ⁴ saturated hydraulic conductivity; ⁵ organic matter; ⁶ porosity (evaluated as SWC at saturation).

shows the variability of the soil properties (sketched in Figure 1) relevant for PI, based on the aforementioned datasets. Totally, the dataset considered in [39,40,42] mostly covered the agricultural soils characterised in [49]. Sand, silt and clay contents were shown to range, respectively, from 0.1 to 99% (mean 56%), from 0.1 to 93% (mean 26%) and from 0.1 to 94% (mean 18%); organic matter content ranged from 0.1 to 12.5% (mean 0.66%), and bulk density from 0.6 to 2.09 g/cm³ (mean 1.42 g/cm³). Soils used for the ROSETTA PTFs were characterised by mean values of porosity and saturated hydraulic conductivity equal to 0.411 cm³/cm³ and 5.04 cm/h, respectively.

2.2. Soil Characterization and Mapping

In order to apply PI, the within-field soil variability must be accurately described. According to the traditional approach in soil detection, a higher accuracy in soil mapping requires us to increase the number of sampling points. Undisturbed soil samples are taken at different soil depths by means of an auger; samples are used for the determination of soil texture and organic carbon. Undisturbed soil samples for the quantification of bulk density and water retention and hydraulic conductivity properties require the opening of trenches to directly collect soil in metal samplers to preserve soil structure. Consequently, the achievement of a detailed description of soil variability at the field scale through a traditional soil survey is very expensive and time consuming, since a dense sampling network should be designed and applied. Innovative approaches introduced in the 1990s adopt geophysical sensors to quantify soil properties through quick and non-invasive surveys (i.e., soil proximal sensing).

3. Agricultural Geophysics and Ground Penetrating Radar

From what has been previously described, it is evident the key role that the knowledge of the electrical properties of soil plays in precision farming processes and decisions [50,51]. Soil physical properties, such as texture, structure, porosity and density, hydrologic attributes, including SWC, and chemical features, are known to relevantly affect the electromagnetic properties of agricultural soils [52,53].

The soil properties mainly affecting EC are SWC, texture (particularly clay content) and bulk density [16,17,18]. Other factors influencing EC are soil temperature and composition of soil solution. As matter of fact, EC depends on the free electrical charges either carried by dissolved ions (i.e., volume conduction), or carried by the solid/liquid interfaces (i.e., surface conduction). As volume conduction dominates in soils, EC is very sensitive to SWC and to the composition of the soil solution. The relationship between EC and SWC is described by Archie’s law; nevertheless, in the range of variation of SWC for environmental and agronomic applications, that relationship is quasi-linear [54]. The influence of soil solution composition is more important when the soil is saturated, while it can be neglected for unsaturated soils. Moreover, for salt soil or sodic soil, EC increases linearly with the increasing electrical conductivity of the soil solution [55]; otherwise, the composition of the soil solution can be neglected if the geophysical survey is carried out sufficiently far from the fertilization periods.

The surface conduction depends on the nature of soil particles and its effect on EC is not negligible in soils containing clay [56]; organic matter content also affects EC [57]. Time-lapse EC surveys on the same site allow us to detect variations of bulk density; indeed, under dry conditions, increasing porosity due to cracking, perforation by worms or roots or tillage operations causes lower EC, while reduction in porosity due to soil compaction determines higher EC [58]. Porosity is a dominant factor on EC under very dry soil conditions, as the EC value for air is very low; in this case, spatial variability in EC detection is mainly due to variations in bulk density. However, for soils with low porosity and under dry conditions, variations in EC discriminate soils with different textures. Conversely, under wet soil conditions, EC is sensitive mostly to SWC, then differences due to texture or other factors are hardly detectable. Finally, the presence of areas with soil compaction are more easily detected through EC survey under dry soil conditions.

It is important to consider the temperature effect for time-lapse EC surveys, particularly in the first 20 cm of soil depth, as temperature can variate a lot during the day, especially in summer. In these cases, EC must be assessed at the reference temperature usually equal to 25 °C by applying a correction equation [59].

Concerning dielectric permittivity, it is highly correlated to SWC [60,61], as dielectric permittivity for water is much higher than that for other soil constituents; indeed, the relative dielectric permittivity of air is 1, while it is 80 for water, and 3–8 for dry mineral materials (with air in pore spaces); hence, the addition of water in the soil pores drastically increases the dielectric permittivity of soil. Soil texture, organic matter content, porosity and temperature affect soil dielectric permittivity as well, even though these factors are usually negligible as their effects are secondary compared to the effect of water. The relationship between SWC and soil dielectric permittivity is described by empirical models, as Topp’s model calibrated for soils with different textures. Under completely dry soil conditions, mapping dielectric permittivity provides information on the spatial variability of porosity and soil mineralogy [62,63,64]; moreover, the vertical variability of the dielectric permittivity allows us to describe soil layering.

The geophysical methods predominantly used for agricultural purposes are ERT, EMI and GPR; additional methodologies, such as magnetometry, self-potential and active seismic methods, have the potential for substantial future use in agriculture, but at present are being limitedly employed for agricultural purposes [21,65,66].

In agriculture applications, the importance of measuring the electrical properties of soils dates back to the end of nineteenth century [67,68,69], with the first electrical prospection experiments being performed by Conrad Schlumberger in France in 1912 [70,71] and Frank Wenner in the United States in 1916 [72]. The earliest application was to determine the soil water content, while advancements have shifted towards the measurements of soil salinity [73,74], which can be considered the beginning of the agricultural geophysics research branch. ERT methods (Figure 2) are based on the measurement of the soil resistance to current flow across (typically) four electrodes inserted along a line on the soil surface at a specified distance. Despite its wide adoption within the field, mainly due to the simplicity in converting sensor measurements to EC values and the possibility of repeated measurements over time, the technique cannot be considered strictly non-invasive, as it requires good contact between the soil and the electrodes and the non-uniqueness of the inversion schemes, hence limiting its applicability and measurement reliability [56].

The use of EMI dates back to the late 1970s, and particularly to 1976, when the first EMI sensor was patented [75] with the main purpose of assessing soil salinity features as a response of the growing need for information on soil properties and behaviour [76,77]. EMI sensors typically transmit a primary EM field inducing electrical currents in the soil, and a receiver records the secondary field, which is used to estimate the EC values. The first reported use of such technique in 1979 highlighted the suitability of the methodology in providing fast and continuous measurements due to its non-invasive configuration, though at the same time stressing the limited depth information that can be gathered [78]. Principal evolution of the technique encompassed the estimation of EC through the development of site-specific empirical relationships for inverting both low conductivity media and higher conductivity values, i.e., when the linearity assumption between the quadrature component of the received EM field and the measured EC does not hold anymore [79,80,81].

ERT and EMI on-the-go sensors (Figure 2a and Figure 3a) are mainly employed in soil survey at the field scale for PA applications. These sensors measure the soil electrical conductivity (EC) at different depths, which is related to both physical and chemical properties of the soil profiles. Since the latter are relevant conditions known to adopt a variable rate management of irrigation and fertilization, according to PA practices, high resolution EC maps (Figure 3b), quite easily obtained through geophysical sensors, are used to delineate the different homogeneous Management Zones (MZ) within the field [82,83,84], where soil variability is expected to be less than among MZs. As a matter of fact, the delineation of MZs can be used to design a soil sampling network with a limited number of points, optimally located within each MZ and distributed among the MZs, thanks to the soil variability scouting provided by the EC maps. In each sampling point, soil water retention and hydraulic conductivity properties can be directly measured on undisturbed soil samples collected at different depths, or otherwise indirectly evaluated from texture data measured on disturbed soil samples. These measures allow us to characterize the soil in each MZ as well as to derive maps for specific soil properties through calibrating empirical models [83,85,86,87,88,89,90,91,92].

The accuracy in soil mapping for specific properties depends on their relationship with EC measurements, as well as on the ability of the geophysical survey to describe the soil variability in both horizontal and vertical directions.

Magnetometers measure anomalies in the strength of the Earth’s magnetic field due to subsurface magnetic heterogeneities, including soil iron content and drainage pipes [93,94]. Instead, seismic methods are based on the analysis of the propagation of elastic waves into the ground, thus measuring density and elastic property anomalies; from laboratory studies, it has been shown that wavelet characteristics (velocity and amplitude) can be correlated with soil compaction, porosity, bulk density and soil water content [95,96,97]. Finally, the self-potential method, one of the few passive geophysical techniques within the domain, records the naturally occurring electric potential difference between two locations; such measurements are associated with electrokinetic processes generated by leakages, water flows, water table depression and horizontal spatial patterns for soil salinity and soil clay content [98,99].

Along with the emerging trend in sensor development for agricultural applications, the decade between 1970 and 1980 has seen the consistent introduction of GPR as a tool to study the variability of soils [100,101,102]. The GPR technique utilises high frequency EM wave reflections at boundaries between materials exhibiting discontinuity with different electrical properties to produce a continuous vertical profile of the investigated area in a cross-section along a transect. GPR is an electromagnetic method similar in principle to the seismic reflection technique, except that it is based on the propagation and reflection of electromagnetic waves rather than acoustic ones. Essentially, a transmitter antenna radiates an electromagnetic wave that propagates into a lossy dielectric material until it encounters a change in material properties, represented by a contrast in the electromagnetic impedance feature. At this interface, a part of the wave is scattered back, and its reflection is recorded by the receiving equipment. The other portion of the wave might instead be transmitted and, therefore, it will continue to propagate. The magnitude of the reflection and transmission phenomena depends on the sharpness of the contrast, as well as on the physical and geometrical features of the discontinuity structure. The amplitudes of the received echoes and the corresponding arrival times are then processed to determine characteristics, nature and location of the discontinuity.

A standard GPR system (Figure 4) consists of a signal generator, a separate transmitter and receiver antennas, typically housed in a single case, and a portable laptop for real-time data display and further data processing. According to the signal generation system, GPR systems are classified into time domain and frequency domain radar, the first category being the most widely adopted in commercial systems. An additional distinction is made depending on whether the equipment is working in contact with the surface (ground-coupled) or above it (air-launched). The choice of one or the other approach represents a trade-off between survey logistics and data interpretability.

The first reported feasibility study was performed in 1979 in a joint effort by NASA, USDA-SCS and the Florida Department of Transportation [103,104]. The main outcome of such a study was that the penetration performance of the methodology was not consistent throughout all the US states (Figure 5), suggesting a distinct dependency with local soil properties and, therefore, the suitability of the GPR methodology for estimating variations in soil properties to classify soils and develop taxonomic composition of soil units, as for example the USDA-NRCS soil suitability map [105,106,107]. In particular, what emerged was the sensitivity of GPR to detect abrupt changes in physical properties, such as texture and bulk density, and in SWC and chemical properties, including organic matter and minerals. Subsequent analysis and studies have demonstrated a decrease in cost (reported 70%) and an increase in survey efficiency (reported 210%) comparing GPR profiling against alternative transect techniques [108].

At the earliest stages, GPR technology was adopted to determine organic layer thickness and depth of stratigraphic discontinuities [109,110], to identify and classify soil types [111] and to detect hard pans [112]. In addition, the effectiveness of GPR in providing time-lapse monitoring of water movements and water table depth variations [113,114], and in identifying preferential flow paths and moisture-front movements [115,116,117] became evident.

Table 2. The main features of geophysical methodologies.

Geophysical Method	Physical Property	Potential Application
Resistivity	Electrical resistivity	Soil drainage Soil salinity Spatial variation Soil water content
Electromagnetic induction	Electrical conductivity	Clay-pan depth Soil nutrient Soil salinity Spatial variations Soil water content
GPR	Dielectric constant and electrical conductivity	Soil classification Vertical microvariability Bedrock depth Plant root biomass Flow pathways Drainage pipes
Magnetometry	Magnetic susceptibility	Drainage pipes Soil pollution and iron content
Self-potential	Electric potential gradient	Soil salinity Soil water content
Seismic	Density and elastic moduli	Soil compaction Soil porosity

summarises the main features of each geophysical methodology.

Ground Penetrating Radar Application for PA

Even from these early studies, a number of clear advantages of the GPR methodology emerged: (1) the capability of providing large quantities of continuous, high-resolution subsurface data, (2) the possibility of jointly estimating the geometrical and the geophysical characteristics of the site, (3) a completely non-destructive and potentially non-contact operation, and as a consequence of that, (4) improved speed of investigation. All these features have gained GPR technology a wide recognition within several application areas, from engineering to hydrogeology, archaeology and forensics [118,119,120]. Successful application of GPR has been reported for subsurface heterogeneities characterisation, as well as for delineating complex subsurface geometries [121].

What can be considered an added value that GPR has brought into the agricultural domain is its sensitivity to both the properties which characterize the electrical behaviour of soils, i.e., the dielectric permittivity and the electrical conductivity. The combined estimation of these two properties by GPR measurements allow for a more detailed soil characterization [122,123,124,125]. Through GPR detection, soils are fundamentally described in terms of their electrical properties, i.e., permittivity and electrical conductivity values, and in terms of some coefficients describing the rates of change of those properties with temperature, density, composition and stratigraphy. Such attributes determine how soils react to an electric field and are inherently related to soil physical properties and, therefore, to its constituents [126]. In particular, soil dielectric permittivity is mostly dominated by soil porosity, water saturation and texture [127,128,129], as a consequence of the large contrast in the dielectric values of air and water. Moreover, simultaneous measurements of soil electrical conductivity allow for the evaluation of the type of fluid contained within the pore-space (and its chemical composition), as well as the soil mineralogy and the clay content [130,131,132].

In addition, a strong relationship has been well documented between soil permittivity, SWC and phase velocity, while the same direct effect does not exist with power attenuation [133,134]. Since the soil dielectric constant is one of the key factors governing wave propagation, reflection and scattering phenomena, GPR measurements are clearly a reliable and accurate way to estimate such attributes [135]. Finally, it should be also mentioned that a number of materials, including water and clay, exhibit a frequency-dependent impact on dielectric permittivity. This means that analysing the spectrum of the received signal, in addition to its amplitude, can provide a further tool to determine the presence of specific constituents and better characterise soil type [136,137,138,139].

Table 3. Typical range of electrical characteristics of different materials.

Material		Conductivity S/m	Relative Permittivity
Air		0	1
Water	Fresh	1 × 10−5	81
	Sea	1 × 103
Sand	Dry	1 × 10−6	4
	Wet	1 × 10−2	20
Clay	Dry	1 × 10−2	4
	Wet	1	25
Limestone	Dry	1 × 10−8	7
	Wet	1 × 10−3	8
Soil sandy	Dry	1 × 10−4	7
	Wet	1 × 10−2	20
Soil loamy	Dry	1 × 10−4	7
	Wet	1 × 10−2	20
Soil clayey	Dry	1 × 10−2	7
	Wet	1	20

describes the electrical conductivity and the relative permittivity values for different materials and soils [140].

From those values, two main considerations emerge. Firstly, comparing the dry/wet characteristics, it is evident how the presence of water impacts both the electrical properties of the material. While a major increase is experienced by the permittivity, variations are also evident in the conductivity values (e.g., dry and wet clay). Salinity, instead, severely affects the conductivity only, leaving the dielectric features almost unchanged (e.g., fresh and sea water) [141].

The feasibility of a GPR investigation, and, thus, its suitability, is primarily determined by (1) the existence and the strength of an EM impedance contrast on sides of the interfaces between different materials, (2) material attenuation and (3) the resolution performance within the depth of interest.

The magnitude of the impedance contrast defines the amplitude of the reflected wave [142,143], thus effectively affecting the possibility to have a record from specific subsurface materials (on each side of the interface), and it is directly related to the ratio between the dielectric constant of those materials [144]. The reflection amplitude of the GPR signal increases for water-saturated soils and for high levels of salinity and clay content. Therefore, since the effect of saturation on the relative permittivity is dominant, the reflection amplitude could give an estimate of the SWC of the underlying soil layer [145,146,147].

Material attenuation controls the wave amplitude loss during propagation, hence determining the maximum achievable penetration depth [148]. Attenuation is directly dependent on both permittivity and conductivity; hence, it is highly sensitive to SWC, clay and chemicals [149,150,151]. The attenuation is also linearly dependent to the operating frequency; therefore, the higher the frequency, the more pronounced will be the attenuation and the achievable penetration will be reduced [152,153]. Additionally, the attenuation of GPR energy shifts the amplitude spectrum of the radar pulse to lower frequencies, known as absorption [154,155,156].

Theoretically, resolution is generally defined as the capability to distinguish, in time and space, two events that are close to each other and it determines to what extent the GPR system is capable of delineating thin layers of heterogeneous soil [157,158,159]. The value is mainly governed by the propagating wavelength and the reflection depth, implying that a wave propagating in different materials exhibits different resolution performance [160,161], and that, despite typically considered a constant value, resolution performance will degrade with depth due to the absorption effect which limits the high frequency content of the propagating wave [162,163].

Finally, though it is not possible to accurately determine GPR performance in terms of maximum penetration, since it depends also on radar-system-dependent parameters and ground- and target-dependent ones, the combined effect with attenuation allows us to recognize different soil conditions [163,164]. Indeed, wet soils quickly attenuate radar signals and, hence, significantly limit the penetration depth (high conductivity, high permittivity), at the same time allowing us to appreciate finer details and features (low velocity and short wavelength). As a final consideration, the high spatial and temporal resolution achievable allows us to cope with the variability of the measured physical properties over very short horizontal and vertical distances. This GPR performance in spatial resolution is an important feature for application in the PA domain, bearing in mind that the soil thickness of interest in agricultural geophysics is within a 2 metre zone directly beneath the ground surface [165,166].

4. Review of GPR Applications to Soil Properties Estimation

The majority of the soil parameters described in Section 2 relies upon the arrangement of solid particles and voids in the soil, as well as from the presence in the void space of water or other fluids. Hence, their determination has been typically accomplished by exploiting relative changes in the soil dielectric constant that caused the reflections and EM wave velocity. In any event, the choice of the optimal survey approach depends on the specific purpose and site condition. Many successful experiments exploited reflection coefficients and early-time signal characteristics to develop site-specific relationships between amplitude attributes and shallow-soil permittivity. This applies not only for the soil hydraulic properties, which can very often be considered the final information to be gathered, but also for the estimation of the physical soil properties, such as soil porosity, compaction and structure.

Inferring specific quantities or constituents, such as soil wilting point and organic matter content, and determining hydrodynamic parameters, including water retention characteristics, has been so far accomplished in a semi-quantitative manner. This mainly includes the development of numerical models of GPR signals and electromagnetic field propagation to explore the relationships between specific subsurface properties and GPR data, and inversion schemes to iteratively reconstruct the true geometrical and amplitude data from the reflections picked from the GPR data. These partial achievements are due to the fact that their quantitative definition by interpreting GPR profiles alone is still not feasible, and that some of the most widely used models relating dielectric constant to the soil constituents are purely empirical.

After a brief description of the currently available GPR technology and the most successful acquisition strategies, this section provides a comprehensive review of the different research approaches, methods and processing strategies that have been adopted for soil characterization within this research domain, highlighting principal characteristics, advantages and limitations. Four main topics will be covered: the determination of the soil textural properties and of the main soil constituents, the estimation of the attributes defining the soil structural properties and the characterisation of the soil hydraulic properties. Despite not strictly being a soil property but a time-varying soil condition, the first sub-section focusses on the review of the SWC estimation techniques. This inclusion has been deemed necessary due to the fact that it represents the starting point wherefrom the aforementioned soil properties are consequentially estimated.

4.1. GPR Equipment Characteristics and Survey Strategies

To derive the most from a GPR survey, key issues to be considered include subsurface structure characteristics, GPR antenna selection and survey design.

From what has been previously described, the two main requirements for a successful GPR survey for agricultural applications are the capability of penetrating down to the first metre below the surface and a centimetre-scale resolution to adequately characterise the specific soil property. For these reasons, the majority of the GPR equipment adopted for agricultural investigations exhibits an operating frequency ranging from approximately 200 MHz to 900 MHz, corresponding to a vertical resolution (calculated as a quarter of the wavelength) of 3 to 14 cm for dry soils, and of 8 to 18 cm for wet soils, thus making the GPR technique the geophysical proximal sensing method with arguably the highest spatial resolution in describing vertical soil profiles. Concerning penetration, lowering the frequencies will ensure a proper performance even in less favourable soil conditions (e.g., clay material).

Regarding the system architecture, time domain or pulsed radars constitutes the majority of commercially available GPR devices due to the ease of engineering, limited cost and fast data acquisition. On the contrary, the principal advantages of frequency domain radar are a larger frequency bandwidth, a wider dynamic range and the possibility of shaping the power spectral density. Optimising the frequency range means that the system operational parameters can be optimised depending on the survey scenario: energy at high frequency is saved if the soil exhibits high attenuation, at the same time maintaining sufficient resolution performance. Until recently, the time-consuming calculations associated with the frequency domain data inversion limited its application, but due to continuously increasing computational capabilities, this limitation no longer applies.

Typical GPR survey collects transects along a single, fixed direction by deploying a single transmitter and receiver pair moving together over the surface and maintaining the same mutual distance. This architecture is the most commonly adopted approach, as this allows for relatively fast and operationally straightforward data acquisition. Alternatively, a multi-static survey geometry, in which the transmitter and the receiver are independently managed and moved, can be adopted. These strategies have been extensively employed for the characterisation of soil velocity, particularly for the reconstruction of vertical and lateral distributions, as they could provide an accurate and precise estimate, and to reduce random and non-coherent disturbances. However, such choice is clearly more time-consuming and more complex to implement.

Operationally, equipment should be able to survey rugged topography and unfavourable terrain conditions, as well as allow the efficient execution of a large number of profiles. These requirements have greatly benefitted from one of the major GPR advancements of the last few years, that is the implementation of a multichannel (i.e., array) air-launched GPR system. Such devices have not only allowed for the simultaneous acquisition of multiple GPR profiles, spatially consistent and highly correlated, but the possibility of mounting the antennas on a vehicle has enabled operationally suitable stand-off distances and the rapid survey of wide areas (Figure 6).

4.2. Soil Water Content

Among the reviewed studies, applications to determine SWC are by far the most frequent [167,168,169], probably as a consequence of the fact that detecting soil moisture levels represent a task which spans several domains, including engineering and other environmental applications [170,171] (Figure 7a).

As the presence of water predominantly affects the dielectric permittivity of the soil, a feature that has gained GPR an advantage over both ERT and EMI [172,173], one of the most adopted approaches is the analysis of GPR wave velocity via reflection data [174,175,176,177,178,179,180,181]. It works by first converting the EM wave velocity into soil dielectric permittivity by inverting the measured travel time, and then translating it into a volumetric water content through an empirical relationship [61,182,183].

To properly assess the EM wave velocity, the depth of the reflector should be known. Consequently, information on the subsurface velocity at certain locations are obtained by varying the spacing between the transmitter and the receiver. These multi-offset configurations, known as common source, common receiver and common mid-point schemes, additionally permit us to estimate the ground wave amplitude and frequency attributes [184,185,186,187,188,189,190] (Figure 7b). Such an acquisition strategy allows for the simultaneous determination of the reflector depth and the wave velocity; hence, it is much more suitable to estimate SWC, but the price to pay is a more complex system design and a labour-intensive acquisition [191]. Another strategy is cross-borehole radar tomography, both vertical and horizontal, a method that provides higher resolution insights into the subsurface, but is limited in the maximum lateral extent that it can afford [192,193,194,195,196]. It can, therefore, be inferred that these two approaches are most suitable for small-scale investigation or numerical model calibration.

Figure 7. GPR prospecting for SWC estimation. (a) SWC distribution determined by the surface reflection method [170]. (b) Velocity spectrum resulting from common mid-point acquisition [189].

Figure 7. GPR prospecting for SWC estimation. (a) SWC distribution determined by the surface reflection method [170]. (b) Velocity spectrum resulting from common mid-point acquisition [189].

Research efforts have also been focussed on detecting and quantifying changes in SWC through time-lapse investigations, an approach considered to be a valuable strategy to monitor subsurface hydrological processes. Time-lapse GPR has demonstrated its capability in resolving small size features even in challenging surroundings where many reflectors are present [197,198,199,200,201]. The method works through a differencing approach by interpreting changes in travel time between reflection horizons or sample-to-sample variations. This aspect represents the main advantage of the methodology: being based on a change-detection approach limits the potential bias introduced by the choice of the specific dielectric mixing model (e.g., Topp’s, Archie’s or suitable alternatives), a mandatory requirement when the aim is to provide an absolute estimate of the investigated feature. Instead, the main prerequisite to ensure the correctness of the differencing scheme is the fact that highly reproducible data are necessary; hence, data consistency (including measurements location and signal sampling) must exist between individual time steps.

Concerning the signal processing algorithms, taking advantage of the well-known frequency-dependent behaviour of the GPR signal, the water content estimation has been correlated to a shift in the spectral characteristics of GPR signals [202,203,204,205,206], appreciable through a time–frequency decomposition [207] or through the application of automated learning algorithms [208,209]. Early-time attributes of GPR surface reflection, a method developed to overcome difficulties in separating the overlapping air and ground waves and that is based on the analysis of the envelope of the signal performed in different time windows, has shown some potential in determining the moisture content [210,211,212,213,214]. Since antenna radiation characteristics are affected by the EM properties of the shallower layer, changes in amplitude, shape and time duration of the ground wave signature are expected to occur for changing soil conditions. However, in addition to a limited accuracy for highly heterogeneous soil, it should be mentioned that the early-time signal is affected by both conductivity and dielectric permittivity but that their effects cannot yet be differentiated.

4.3. Soil Textural Properties

In principle, the capability of GPR in providing information on soil texture depends on the magnitude and sharpness of its variation; therefore, it is possible to locate soil horizons with strongly contrasting textures only [215,216,217,218]. For this reason, a limited number of works have vertically addressed the topic, resulting in a commonly accepted high-level characterisation of soils into coarse-grained and fine-grained texture [219,220,221].

To overcome these limitations, considering the small differences in permittivity values described in Table 3, textural estimation has been indirectly obtained from variations in the SWC. In particular, the most successful approach has been the determination of the soil clay content. From a radar perspective, with respect to the other two soil constituents (i.e., silt and sand), clay has a predominant effect on GPR wavelet amplitude and frequency characteristics, as well as producing a relevant increase in soil conductivity [222,223,224].

Clay content has been successfully delineated at both the field and laboratory scale mainly through two different approaches: the analysis of the early-time GPR signal amplitude (Figure 8a) and the observed signal attenuation; the evaluation of the received signal in the frequency domain. In the latter case, a significant shift towards lower frequencies has been observed for increased clay content [225,226,227,228,229,230].

Amplitude-based analysis is based on the principle that a signal propagating through a clayey soil is more attenuated, and, therefore, penetration depth decreases. This consideration is at the foundation of the work of Doolittle and Collins [106], where the soil textural properties were indirectly estimated by evaluating the penetration performance of GPR. Essentially, bearing in mind a quasi-homogeneity assumption, abrupt variations in the recorded amplitude can be associated to soil textural transitions, especially between sand and clay textures [231,232,233,234,235,236,237] (Figure 8a).

Effects due to organic matter content are similar to those of clay [238,239,240]; hence, increased content generally produces a soil with increased water-holding capacity and conductivity [241,242] (Figure 8b). If, on one side, this dependency suggests that measurements of permittivity values could provide a way to estimate the organic content feature [243,244,245,246,247], at the same time it represents a challenge in scenarios where organic matter occurs in association with clay and water [173,248,249]. In those situations, GPR results alone are not able to conclusively resolve the different contributions; therefore, it is not possible to differentiate their impact on the collected GPR signals [250,251,252,253]. As a consequence, most of the advancements have been achieved by refining and improving the existing dielectric mixing relationships in order to account for both the soil texture and the organic content [254,255,256]. Nevertheless, some successful results have been obtained in the case of large impedance contrast through the analysis of GPR wavelet amplitude analysis [257,258,259,260,261].

Figure 8. GPR prospecting for soil textural properties. (a) Prediction of clay content from analysis of GPR reflection data signal amplitude [236]. (b) Variations of soil hydraulic conductivity and moisture by organic matter [251].

Figure 8. GPR prospecting for soil textural properties. (a) Prediction of clay content from analysis of GPR reflection data signal amplitude [236]. (b) Variations of soil hydraulic conductivity and moisture by organic matter [251].

4.4. Soil Structural Properties

Identifying soil structural properties, including compaction, porosity, density and vertical stratification in different horizons, represents a task that well matches the operational capabilities of GPR, mainly as a consequence of the fact that the high-resolution feature exhibited by the methodology makes it capable of following the spatial and temporal variations to which such properties are subjected [166,262,263,264].

Until now, the observation of soil compaction (e.g., decrease in porosity, increase in density and change in soil hydrological properties functions), has been accomplished by exploiting relative changes in soil dielectric constant and EM wave velocity, due to the fact that soil water content varies with soil compaction condition, therefore indirectly affecting dielectric permittivity [265,266,267,268,269] (Figure 9a). The majority of experiments have approached the problem by analysing the wavelet characteristics (amplitude and spectrum) of early-time GPR signals, for which a strong correlation has been found with the dielectric constant, and by establishing the corresponding empirical relationships [270,271,272,273,274,275,276,277,278,279].

Due to the direct impact that the structural properties have on the mechanical ones, several works have combined in situ mechanical and penetrometric probes with GPR measurements, with the primary aim of estimating the required effort for land regeneration processes [280,281,282,283].

What all these studies have in common is the fact that they demonstrated the suitability of GPR in detecting soil layers affected by compaction, or generally to detect less permeable layers (Figure 9b), but at the same time they highlight the necessity of adequate and additional ground truth data to retrieve an accurate relationship between soil compaction and GPR signal, as well as the knowledge of previous or current land use. For these reasons, conducted research typically involved dedicated compaction-induced processes and sample collections for calibration.

The characterisation of porosity features has been usually obtained through the analysis of GPR reflection amplitude in combination with a number of petrophysical relationships and signal inversion models to convert the measured dielectric permittivity into porosity information [284,285,286,287,288]. In addition, wave velocity variations due to the differences in porosity have been tracked through common mid-point and wide-angle reflection and refraction acquisition schemes [289,290,291,292,293] and cross-hole tomography [294,295,296,297]. To improve the accuracy of the process, Full Wave Inversion (FWI) techniques [298,299,300,301] and automated feature extraction schemes [302,303] have also been developed.

4.5. Soil Hydraulic Properties

For what concerns the characterisation of the soil hydraulic parameters (Figure 10), mainly represented by the soil water retention and hydraulic conductivity functions [304,305,306,307,308], the majority of works in the literature have been carried out at laboratory or small scales [309], with the main purpose of developing accurate hydrodynamic models that could match pedotransfer functions and experimental data [310,311,312]. Reference models have been developed in [34,313,314]. In particular, the characterisation of soil hydraulic conductivity has been typically accomplished through the analysis of the observed signal attenuation in both unsaturated and saturated soil samples, as well as by processing time-lapse GPR data, and in particular GPR groundwave data, in search for changes in SWC with time [315,316,317,318,319]. The estimations are then coupled with the analytical relationships previously mentioned (principally the Van Genuchten model and its modifications) to finally obtain a soil-specific empirical description. The determination of the parameters involved in the analytical formulations is usually performed by artificially forcing a saturated soil condition, either through artificial infiltration, percolation or infiltrometer methods [320,321,322,323,324], in which the volume of infiltrated water versus time is fitted to infer soil hydraulic conductivity at or close to saturation. However, in addition to being time-consuming and difficult to replicate on a large scale, such methodologies are limited by a significant uncertainty in accurately contouring the infiltration bulb shape.

A similar approach has been followed to identify the parametrisation of the soil water retention curve, i.e., its functional form [325,326,327,328]. Specifically, the soil water retention curve relates the soil water pressure head to its volumetric water content, and its shape is determined by soil texture and pore size distribution. Considering that wave propagation velocity decreases due to an increase in water content and the signal frequency spectra allows us to characterise the unsaturated and the transition zone [329,330], GPR data inversion has been applied to find the water retention curve parameters. GPR reflection measurements have been successfully adopted to determine the location of the groundwater table, while numerical models have been developed to estimate the wave propagation in the transition zone. By combining these results and the analytical relationships previously mentioned, it is possible to reconstruct the water retention curve by data fitting [48,331,332,333,334,335,336,337,338]. Despite showing promising results, significant errors have been observed for deeper layers, where noise in the measurements drastically reduces the accuracy of the inversion schemes. In addition, a great deal of ancillary information is required to optimise the conceptual design of the GPR inversion model, such as details on soil texture, compositions and stratigraphy [339,340].

These results indicate that GPR is a useful tool to capture the trend in hydraulic conductivity and retention with reasonable accuracy for agricultural applications at the field scale, but the estimation of the actual values is still subject to high variability due to the assumptions and simplifications that are included in the inversion model, such as homogeneity, isotropy and non-dispersive 1D propagation [326,328].

Finally, very few works have addressed the determination of critical SWC values, such as soil WP and soil water content at Field Capacity (FC), needed to determine the available water capacity. The main reasons for this are the fact these soil hydraulic conditions are temporary and not uniformly distributed in the field given soil heterogeneity (FC), and difficult to reach in real field conditions (WP) [341,342]. Moreover, WP depends not only on the soil hydric regime, but also on the interaction with crop type [343,344].

Works on the topic have mainly addressed the development of specific dielectric mixing models, or the modification of existing ones, that could account for this feature when building the forward model of the permittivity values to soil water content. The estimation of the available water content property is then performed by matching the model functions with surface soil moisture measurements, obtained either through GPR reflection or early-time signal analysis [345,346,347,348]. Given the limitations of GPR in providing an absolute value of water content alone, i.e., without relying on external or ancillary information, it is clear that all the processes that follow will hardly produce quantitative representation of the exploited feature.

5. Discussion

The principal outcome that can be derived from the exploited panorama is the fact that the majority of the described soil parameters stem, directly or indirectly, from the measurement of the soil water content feature, which clearly represents the most-researched attribute. Under the assumptions that variations in the apparent soil permittivity and conductivity are dominated by changes in the water content, such information has been mainly retrieved through the analysis of the variations in GPR wave velocity and reflection coefficient magnitude. Though theoretically reasonable, the mentioned hypothesis means that small-scale spatial heterogeneities, lateral effects, interdependencies and mutual correlation among different parameters (textural or small structural changes and organic contents) are often overlooked, if not completely neglected.

This is exemplified by the achieved results in the field of textural attribute characterisation, where most of the research has been carried out to improve and refine the existing dielectric mixing and pedotransfer relationships. Considering that, up to some degree, sand, silt and clay exhibit a comparable effect on soil permittivity and conductivity, their individual contribution to the GPR wave characteristics can hardly be separated, thus limiting the extraction of information from the GPR profile alone.

This last point is even more stressed for the extraction of specific values or soil conditions, which is typically performed by inverting functions and detailed models to obtain the distribution of subsurface permittivity and conductivity. Between these two situations, research carried out on soil structural property classification, and in particular on soil density and compaction attributes, has shown that it is possible to infer the physical attributes from the interpretation of GPR results, despite being through the detection of relevant changes in the amplitude characteristics of GPR images.

Finally, it can be stated that GPR has proven its validity in providing deeper information on subsurface properties variations than those from alternative geophysical methodologies. On top of that, the possibility of densely surveying large 3D areas in a timely fashion thanks to the deployment of multi-channel systems represents a relevant operational benefit towards its implementation for irrigation optimisation to ensure water sustainability and enhanced agricultural productivity. The ability to obtain dense estimates of soil features holds potential for improving agricultural ecosystem investigation and, hence, the delineation of MZs, at the same time limiting resource wasting. Moreover, the higher the sensitivity of the crop to soil moisture, the larger the benefits will be in employing the GPR methodology.

6. Conclusions

The accurate characterisation of the spatial and temporal variation of soil properties (texture, structure and hydraulic) at the field scale, as well as the determination of soil moisture dynamics, are fundamental in PA. This contribution has reviewed the current state of the GPR research for applications in precision agriculture, and specifically for the domain of water management and precision irrigation. After a preliminary excursus on the motivations that have pushed the deployment of sustainable agricultural techniques and practices, this review has assessed the role that geophysical measurements have played throughout the last decades and beyond.

In particular, the focus has been put on the different operational and processing approaches and methodologies developed to characterise textural, structural and hydraulic soil properties, which affect the electromagnetic behaviour of soils and can, therefore, be inferable from GPR profiling results.

Summarising the outcomes of the review, it can be concluded that the field of precision farming, despite being extensively developed by the introduction of proximal geophysical sensors, poses a fundamental challenge that still hampers a full exploitation of the potential of the GPR technique, mainly with respect to the data interpretation capability: the natural heterogeneity of soils. Wave propagation, amplitude absorption and GPR reflections are due not only to soil water content, but they are also impacted by other factors that might occur at different temporal and spatial scales, as well as with different magnitude. Therefore, the difficulty in separating and isolating the different contributions is clear, which in turns could cause uncertainties in soil parametrisation and classification. Advanced modelling and wave inversion schemes have clearly mitigated the impact of such errors, but they still suffer from several assumptions, including homogeneity or geometrical simplification, and severe computational requirements. Opportunities exist within the domain of quantitative analysis, with the intent of increasing the level of information that can be extracted from the received signal, as well as from the integration of GPR and other techniques, including semi-invasive ones, as a means to better constrain the problem and reach a more accurate solution. In particular, acknowledging the main limitation that each subsurface layer can respond to more than one soil property variation, combining conceptually different sensing techniques, and, hence, put emphasis on different soil properties, holds the promise of improving the accuracy in property estimates. Such an approach could lead to the investigation of a wide range of physical, chemical and mechanical characteristics, which in turn will allow for a deeper understanding of soil relationships.