Assessing Spatial Variability of Soil Properties in Mediterranean Smallholder Farming Systems

Abstract

Smallholder farming systems are typical of the European Mediterranean region. Small farms of less than 2 hectares cover approximately 15% of cropland in the southern EU and only 5% across the EU. The greater variability of cultivated species per unit of cropland (ha), the different approaches, and empirical application of cultivation practices by smallholder farmers increase the spatial variability of soil properties. Therefore, a decision support tool for effective management practices was formed based on a soil indicators set, which is sensitive to changes under agricultural management practices and different LUs. The data for this task were collected from 364 crop fields. The data were clustered and correlated based on (a) the existing soil units (SU): Fluvisols, Cambisols, Luvisols, and Calcisols, and (b) the LU: pastureland, annual, and permanent crops. Principal component analysis (PCA) identified up to seven main components that can better explain soil variability properties. The results indicated that the selected soil indicators can explain only 70.98% of soil variability. Clustering the parameters based on LU and SU can explain up to 80% and 82% of soil properties’ variability, respectively. Factor analysis could function as a decision support tool for soil fertility management by farmers or policy makers, who aim to achieve higher yields, promote sustainable practices, maintaining, at the same time, a low cost of cultivation.

1. Introduction

Nowadays, agricultural activities do not fulfil the requirement of meeting basic human needs without compromising those of future generations. Therefore, sufficiency in sustainable agriculture is not considered to be a basic requirement, however, long-term stability and efficiency are both necessary. For this purpose, any intervention in agriculture should be based on the protection of natural resources and the maintenance of the health and productivity of agricultural land [1].

However, crops and soil form a complicated nexus: plants depend on the characteristics of soil substrate, but they can also alter them [2]. Changes in agriculture land use (LU) affect both input and output fluxes of nutrients and carbon, which in turn affect soil health [3]. Agricultural land is divided mainly into three land uses, according to the data obtained by FAOSTAT (2021). The Greek agricultural land in 2019 occupied about 6,103,600 ha, of which more than 47% was covered by land with permanent meadows and pastures, followed by arable land (35%) and land with permanent crops (18%), while across the EU, agricultural land occupied about 164,925,146 ha and was covered by 33% permanent meadows and pastures, 60% arable, and 7% permanent crops [4]. On the other hand, soil units (SU) varied based on different properties, highlighting the significance of the latter to evaluate soil functions [5]. Rodeghiero et al. [5] stated that in the Mediterranean region, soils exhibit a high spatial variability of properties. In Greece, Cambisols are the dominant soil group, consisting of approximately 38.7% of soil, followed by Fluvisols (24%), Calcisols (11.1%), Luvisols (9%), Leptosols (8.4%), Regosols (4.3%), Vertisols (4.0%), Histosols (0.1%), and Gleysols (0.7%) [6]. Yassoglou et al. [7] reported that the SU affects both the quality and the sustainability of the soil, on which agricultural production is mainly based, and constitutes a crucial factor for the economic and social development in Greece. It is obvious that the soil functions are affected by the main soil-forming factors such as the climate, the parent material, the relief, the biotic factor, and the anthropogenic interventions performed.

Blum et al. [8] reported that most soils (44% of the global land area) present favorable conditions for agriculture, while some soil units (SU) are inherently deficient in plant nutrients [9]. On the other hand, different land use, long-term intensive monoculture systems, followed by environmentally unfriendly long-term agricultural practices, adversely affect both health and soil fertility. In order to achieve a multi-functional land evaluation, a soil analysis of a wide range set of physical, chemical, and biological indicators is required [3]. The assessment of a large set of soil indicators is a complicated, costly, and time-consuming process [10]. Instead, in order to facilitate the optimization of management practices in cultivated lands, a minimum dataset (MDS) of soil chemical and physical indicators can be used [11]. Takoutsing et al. (2016) [11] focused on chemical properties because they were considered the most important factors that have been affected by land management, having a great impact on crop productivity. Rezaei et al., [12] reported that the introduction of the MDS method has greatly facilitated soil assessment and the selection of soil indicators. Additionally, soil data redundancy can be diminished by using a general approach, through choosing the most representative indicators from a large existing data set. However, the weight for each soil quality indicator was obtained by its communality, calculated by mathematical statistics of standardized factor analysis [13,14], thus, helping to avoid subjective judgments. Moreover, a number of methods for a quantitative determination of MDS have been proposed, and the principal component analysis method is the most widely used.

Large-scale farmers perform regular soil analysis to improve mainly their crop production and quality, which is a useful diagnostic tool for checking both fertility and soil health. However, in Greece, small farms (farm size less than 2 hectares) typically cover 11% of the total Greek cropland, while in France and Germany they cover 0.2% and 0.1% respectively [15]. In a previous survey in the study area, 500 smallholder farmers were asked about the use of fertilizers and soil testing. Approximately 75% of them stated that they had not carried out soil analysis for a long period (>10 years) [16], which is an unacceptable agricultural practice, leading to yield reduction. In addition, uncontrolled empirical use of fertilizers without any recommendation increases the cost of fertilization and usually has adverse environmental impacts. [17].

Thus, the development of a decision support tool based on soil analysis and appropriate soil indicators would be essential in smallholder farming systems for effective best management practices. Several previous studies have focused on the effect of either land use (LU) or soil unit (SU) on the edaphic properties’ variability, mainly on natural (undisturbed) soils, while few studies have focused on Mediterranean agricultural soils [18,19,20,21].

In this study, an attempt has been made to (a) assess the current level of physical and mainly chemical soil properties that affect crop performance and (b) to identify the most reliable indicators that would support the development of a decision-making tool for “effective best management practices” in smallholder farming systems. A minimum dataset (MDS) of soil indicators was used as a decision support tool. In order to include sensitive indicators in a regional level, and plan cost-effective strategies with factor analysis, samples were clustered based on soil unit (SU), according to existing soil maps and land use (LU) as observed in the study area.

2. Materials and Methods

2.1. Study Site

The study area is located in the prefecture of Aitoloakarnania (21°15′ N and 21°37′ E) in Western Greece (Figure 1). Aitoloakarnania is the largest prefecture in Greece (38°37′ N, 21°22′ E) and it occupies an area of 5448 km², representing the 4% of the total area in Greece. Although it is an important agricultural area rich in water reserves, there are structural problems in agriculture, where the average farm size is approximately 2.9 ha, while at a national level it is 3.9 ha. According to the European Soil Data Centre (ESDAC) and the national soil maps [6], the dominant SU in our study area are Fluvisols (FL), Luvisols (LV), Calcisols (CL), and Cambisols (CM). The climate in the prefecture of Aitoloakarnania is typical Mediterranean, with precipitations ranging between 800 and 1000mm. Nevertheless, the uneven seasonal distribution of rain makes irrigation an obvious and necessary option to increase and stabilize crop production. Throughout the region of Aitoloakarnania, the land use (LU) consists of annual crops (arable, 23%), permanent crops (17%), and abandoned or pastureland (58%). Livestock plants such as vetch, legumes, alfalfa, oats, barley, etc., are cultivated due to the fact that sheep and goat farming is the main activity of the inhabitants of the Regional Unit of Aitoloakarnania. Maize covers the largest area of cereals and grains, while cotton is the major industrial crop of the region. Additionally, olive groves and citrus orchards cover most of the permanent crops, occupying 85 and 11%, respectively.

2.2. Data Collection-Soil Sampling

The data collection took place in the winter of the years 2019–2020. The selected fields (364) are located mainly in the area of the municipalities of Messolonghi and Agrinio in Western Greece (Figure 1). Samples were clustered based on both land use/cover and national soil maps (ESDAC) in the study area. From the total 364 samples, about 30% of the composite soil samples were collected from annual and pasture or abandoned land, while 40% were from permanent crops, in order to have a representative presence of each type of land use (LU) with the approximately same slope gradient (0–2%) and climatic conditions. Each field covered an area of approximately 0.5–3 ha. Fields of permanent crops included approximately 200 olive trees, 330 citrus trees, 450 kiwi fruits, 80–88 thousand plants of maize, 90–100 thousand plants of cotton, etc.

Soil sampling took place in the winter season, when soils are inherently variable in their distribution of plant nutrients [22]. A total number of 364 soil samples were collected following a composite sample approach and labeled with a GPS device (Figure 1). A soil sample was received from selected permanent crops as follows: a central sub-sample and then four other sub-samples within a distance of 2 m from the central one were collected—all beneath tree canopy—and mixed, as proposed by LUCAS topsoil sampling methodology [23]. Trees of about the same height, canopy use, and distance within orchard were selected, and were located in the center area of the field. No sample was taken in boundary areas or near trees out of the average tree height and canopy use. Each sample received in annual crop fields and in abandoned lands consisted of 10 cores and were well-mixed on site, which were collected from different points in the field with a zigzag method of soil sampling [18]. Sampling was conducted by using a Dutch auger to a depth of 0–30 cm for all soil samples. However, one sample from each sampling site was collected to determined soil BD. Soil bulk density (BD) was determined by collecting undisturbed soil cores from 0–30 cm depth using 100 cm³—cylinders (5 cm height and 5.04 cm diameter) [24].

2.3. Laboratory Analysis

Soil samples (n = 364) were air-dried, crushed, and passed through a 2 mm sieve. Texture analysis was carried out according to the Bouyoukos method [25]. Electrical conductivity (EC) and pH of saturated pastes were measured for each sample using conductivity (HandyLab LF1, SCHOTT AG, Mainz, Germany) and pH (GLP21, Crison, Barcelona, Spain) meters, respectively [26]. The CaCO₃ equivalent (CCE) was determined using a Bernard calcimeter [27]. Soil organic matter (SOM) was determined by the Walkley–Black method [28]. The exchangeable K and Mg were determined following extraction by 1 N ammonium acetate at pH 7.0 [29], and the available P was measured by the Olsen method [30]. Mn, Fe, Cu, and Zn availability were determined by a DTPA extraction [31], while their concentrations were determined using an Analyst 700 (PerkinElmer Inc., Waltham, MA, USA) flame atomic absorption spectrometry (AAS). Soil B content was estimated via hot water extraction and applying the azomethine-H method [32]. Determination of NO₃⁻ was performed in 1:10 water extracts using Dionex-1500 Ionic Chromatography [33].

2.4. Statistical Analysis

Analysis of variance (Kruskal Wallis Test—nonparametric test) was performed in order to determine significant differences among different LUs and SUs. Correlations among soil properties were determined using the Pearson’s rank correlation (Proc CORR, Pearson) procedure. Principal component analysis (PCA) was used to identify properties that explain most of the variability and to select the most appropriate indicators (a minimum data set MDS) that reflect soil quality. Statistical analysis was carried out using the SPSS statistical package version 20.

3. Results

3.1. Assessment of Soil Properties in the Total Study Area (Stotal)

The interpretation (

Table 1. Interpretation of soil properties in total study area (Stotal).

Properties	Soil Interpretation
pH	Rating	4.01 to 5.5	5.51 to 6.8	6.81 to 7.2	7.21 to 7.5	7.51 to 8.5
	Class	Moderately acid	Slightly acid	Near neutral	Slightly alkaline	Moderately alkaline
	% of S.S *	2.2	18.1	6.6	18.4	54.2 (+0.5 with pH 8.6 **)
SOM (%)	Rating	<1	1 to 2.5	2.5 to 5	5 to 10	>10
	Class	very Low	Low	Medium	High	very High
	% of S.S *	3.8	66.9	28.8	0.5
CaCO3 (%)	Rating	<1%	1–2.5 %	2.5–5%	5–10%	>10%
	Class	Poorly	Moderate	Rich	loam soils	Calcareous
	% of S.S *	24.3	12.7	11.4	21.4	30.2
EC (dS m−1)	Rating	<4	4 to 8	8 to 16	16 to 40	>40
	Class	Normally	slightly saline	moderately saline	strongly saline	Very strongly saline
	% of S.S *	100
Mg (mgKg−1)	Rating	<10	10 to 50	50 to 300	300 to 500
	Class	very Low	Low	Moderate	High
	% of S.S *		1.1	89.8	9.1
K (mgKg−1)	Rating	<100	100 to 500	500 to 700
	Class	Low	Moderate	High
	% of S.S *	10.3	83.3	6.4
NO3-N (mgKg−1)	Rating	<3	3.1–10	10.1 to 20	20.1 to 40
	Class	Low	Inadequate	Moderate	Adequate
	% of S.S *	22.3	43.6	16.8	17.3
POlsen (mgKg−1)	Rating	<5	5.1 to 15	15.1 to 25	25.1 to 45
	Class	very Poor	Inadequate	Sufficient	more than Sufficient
	% of S.S *	15.6	32.3	14.5	37.6
Cu (mgKg−1)	Rating	<0.1	0.1 to 1	>1
	Class	Low	Moderate	High
	% of S.S *		14.2	85.8
Zn (mgKg−1)	Rating	<1	1 to 10	>10
	Class	Low	Moderate	High
	% of S.S *	54.7	43.9	1.4
Mn (mgKg−1)	Rating	<1	1 to 10	11 to 100	>100
	Class	very Low	Low	Moderate	High
	% of S.S *	2.5	57.8	39.1	0.6
Fe (mgKg−1)	Rating	<0.1	0.1 to 10	>10
	Class	Low	Moderate	High
	% of S.S *		11.4	88.6
B (mgKg−1)	Rating	<0.7	0.8 to 1.2	1.3 to 2
	Class	Low	Moderate	High
	% of S.S *	67.4	31.2	1.4
BD (g cm−3)	Rating	1.20–1.40	>1.40–1.60	>1.60
	Class	Ideal BD depends on the soil texture
	% of S.S *	40.7	59.1	0.3

* S.S.:% of soil samples; in total, 347 soil samples were taken from the depth of 0–30 cm (topsoil layer) for determination of fertility status on study area. The interpretation of soil properties was in line with Triantafyllidis et al. [14]. ** Strongly alkaline.

) of soil analysis was initially performed in the total study area (Stotal) in a large (17) set of soil properties’ data, showing a high spatial variability. The soil reaction in Stotal showed that soil pH values ranged from 4.3 to 8.6 with mean values of 7.4 (slight alkaline). Soils rich in CaCO₃ equivalent (CCE) were found with mean content of 7.3%, while it ranged from 0.0 to 68.4%. Additionally, soil EC values ranged from 0.15 to 3.30 with a mean value of 0.77 dS m⁻¹. The moderate soil organic matter value was 2.2%, ranging from 0.2 to 5.6 %. The mean soil bulk density was 1.42 g cm⁻³, ranging from 1.21 to 1.62 in different soil types. Moderate concentrations (adequate levels for most plants) characterized the following nutrients: Mg (176 ± 4.75 mg kg⁻¹), K (240 ± 8.65 mg kg⁻¹), NO₃^-N (12.5 ± 0.90 mg kg⁻¹), Zn (1.8 ± 0.25 mg kg⁻¹), Mn (12.0 ± 0.60 mg kg⁻¹). High mean concentrations (30 ± 1.8 mg kg⁻¹) of available P (excellent reserve of available P for plants in acid to alkaline soils) were observed using the P Olsen method. Similarly, high copper and iron mean concentrations were found (Cu:4.5 ± 0.26 mg kg⁻¹ indicate possible toxicity to Cu-sensitive plants, probably due to soil contamination; Fe concentrations 29.8 ± 1.69 mg kg⁻¹ with effects unknown).

As shown in

Table 2. Correlation matrix based on Pearson’s correlation coefficients between soil chemical and physical properties in total study area (Stotal).

pH

pH

1.000

SOM

SOM

0.063

1.000

CaCO3

CaCO3

0.440 **

0.012

1.000

sand

sand

−0.190 *

−0.117 *

−0.197 **

1.000

silt

silt

0.169 **

−0.156**

0.166 **

−0.653 **

1.000

clay

clay

0.089

0.308 **

0.108 *

−0.661 **

−0.117 *

1.000

EC

EC

0.098

0.163 **

0.080

−0.278 **

0.217 **

0.155 **

1.000

Mg

Mg

−0.136 *

0.143 **

−0.220 **

−0.257 **

0.016

0.308 **

0.075

1.000

K

K

−0.085

0.267 **

−0.088

−0.048

−0.087

0.160 **

0.164 **

0.167 **

1.000

NO3-N

NO3-N

−0.134 *

−0.044

−0.076

−0.122 *

0.088

0.072

0.553 **

0.087

0.038

1.000

P

P

−0.435 *

−0.007

−0.358 **

0.085

−0.138 **

0.023

0.059

0.143 **

0.355 **

0.224 **

1.000

Cu

Cu

−0.342 *

0.027

−0.168 **

0.173 **

−0.196 **

−0.033

−0.093

0.012

0.155 **

0.115 *

0.192 **

1.000

Zn

Zn

−0.107 *

0.051

−0.122 *

0.059

−0.078

0.002

−0.012

0.016

0.128 *

−0.023

0.101

0.025

1.000

Mn

Mn

−0.326 *

0.160 **

−0.306 **

0.070

−0.084

−0.006

0.030

0.142 **

0.069

−0.049

0.118 *

0.013

0.205 **

1.000

Fe

Fe

−0.600 *

0.068

−0.220 **

0.107 *

0.016

−0.157 **

0.016

0.098

0.089

0.062

0.347 **

0.243 **

0.027

0.345 **

1.000

B

B

−0.184 *

−0.096

−0.055

−0.013

−0.157 *

0.105

0.102

0.178

0.213 **

0.308 **

0.280 **

0.161 *

−0.022

−0.055

0.127

1.000

BD

BD

−0.111 *

−0.610 **

−0.122 *

0.689 **

−0.233 **

−0.672 **

−0.234 **

−0.271 **

−0.162 **

−0.069

0.091

0.035

0.021

−0.083

−0.019

−0.003

1.000

** Correlation is significant at the 0.01 level (2-tailed). * Correlation is significant at the 0.05 level (2-tailed).

, significant correlations were observed among the soil properties in the total study area (Stotal). The most significant correlations among the parameters were the following: soil pH values were positively correlated (p <0.01) with the total CaCO₃, while being negatively correlated with Fe and Olsen P content. Soil bulk density values were positively correlated (p <0.01) with sand content, and negatively correlated with soil organic matter and clay content. Nitrate soil concentration (NO₃-N) was significantly (p <0.01) correlated with EC.

Finally, significant (p < 0.01) correlations were observed between available soil K and Fe concentrations with Olsen P concentrations (Table 2). As shown in Figure 2a, the higher CaCO₃ content decreased the available Olsen P, while the lower soil pH values increased the available Olsen P. These high P concentrations were positively correlated with high Fe and K availability (Figure 2b). Moreover, the results showed that 17.3% of the soil samples were characterized as clayey and 82.7% of them as loamy according to their soil texture in 364 fields with different land use (Figure 3).

Principal component analysis (PCA) was used to identify the most appropriate soil indicators (a minimum data set MDS) that reflect soil function from a large (17) data set of soil properties. A minimum data set (MDS) of these indicators could provide us with accurate and up to date information about the soil variability of fields in the study area, which can be used as a decision support tool for agriculture management practices. After interpreting all components (Table 2), the results are as follows: PC1 was attributed to bulk density, soil texture, and soil organic matter, explaining 16.4% of variance; PC2 was attributed to soil reaction and its effect on soil nutrient concentration and probably to Cu fungicide application (14.7% of variance); PC3 was attributed to anthropogenic interventions such as mineral fertilization and the high content of CaCO₃ in the study site (11.6% of variance); PC4 was attributed to N mineral fertilization and the significant correlation of nitrate soil concentration (NO₃-N) with EC (10.5% of variance); PC5 was attributed to geogenic factors such as soil texture; PC6 was attributed to CaCO₃ soil content and its effect on soil micronutrient availability. However, factor analysis of soil properties in Stotal reported that upon varimax rotation, six principal components could explain only 70.987% of the total variance (

Table 3. Results of the principal component (PC) analysis of soil properties in the total study area (Stotal).

Soil Properties	Stotal
	PC1	PC2	PC3	PC4	PC5	PC6	CE
BD	−0.910						0.887
clay	0.838						0.806
SOM	0.684						0.680
Fe		0.789					0.702
Cu		0.724					0.601
pH		−0.645					0.671
K		0.428	0.412				0.490
P			0.734				0.645
Mg	0.438		0.651				0.690
CaCO3			−0.552			−0.440	0.671
B			0.536	0.430			0.543
NO3-N				0.870			0.783
EC				0.804			0.741
silt					−0.935		0.912
sand	−0.652				0.662		0.937
Zn						0.740	0.561
Mn		0.575				0.594	0.747
Eigenvalue	2781	2498	1981	1786	1605	1417
% of Variance	16,360	14,694	11,653	10,507	9439	8334
Cumulative %	16,360	31,054	42,707	53,214	62,653	70,987

The Rotation Method was the Varimax with Kaiser Normalization. Rotation converged in six interactions. CE stands for communalites estimates. Coefficients with absolute values less than 0.4 were suppressed. Significant loadings in PCA factors are shown in bold.

).

3.2. Interpreting the Spatial Variability of Soil Properties Across Different Land Uses (LU)

Soil samples (364) were taken from 29% of the fields covered by annual crops, 29% of pasture / abandoned areas, and 42% of fields covered by permanent crops as presented in Figure 3. Among different land uses (pastureland, annual, and permanent crops), statistically significant differences were observed in 12 out of the 17 analyzed properties, showing that the availability of soil nutrients was higher in pastureland than in annual and permanent crops (

Table 4. Least square means with standard error (SE) of soil chemical and physical parameters in different land uses (LU).

Soil Properties	Land Use (LU)
	Annual Crops (n = 105)	Permanent Crops (n = 152)	Pasturelands (n = 107)
pH	7.58 ± 0.10 *	7.46 ± 0.04 *	7.17 ± 0.12 *
SOM (%)	2.23 ± 0.13 *	2.21 ± 0.06 *	2.40 ± 0.15 *
BD (g cm−3)	1.41 ± 0.01 *	1.42 ± 0.01 *	1.41 ± 0.01 *
CaCO3 (%)	6.35 ± 0.94 *	8.04 ± 0.56 *	4.96 ± 0.93 *
Sand (%)	32.5 ± 1.77 *	36.9 ± 0.89 *	35.9 ± 1.83 *
Silt (%)	37.7 ± 1.28	33.6 ± 0.65	33.3 ± 1.64
Clay (%)	29.8 ± 1.39 *	29.4 ± 0.67 *	30.7 ± 1.32 *
EC (dS m−1)	0.70 ± 0.04	0.79±0.03	0.73 ± 0.06
Mg (mg kg−1)	171 ± 11.4 *	172 ± 5.28 *	194 ± 15.8 *
NO3-N (mg kg−1)	7.68 ± 1.09	12.9 ± 1.13	14.1 ± 2.17
P (mg kg−1)	30.2 ± 4.01 *	29.2 ± 2.05 *	33.7 ± 5.89 *
K (mg kg−1)	216 ± 18.6 *	238 ± 9.69 *	269 ± 29.3 *
Cu (mg kg−1)	2.69 ± 0.27 *	4.54 ± 0.30 *	5.45 ± 0.84 *
Zn (mg kg−1)	1.25 ± 0.18	1.82 ± 0.33	1.98 ± 0.38
Mn (mg kg−1)	14.7 ± 2.47 *	10.8 ± 0.63 *	15.5 ± 1.44 *
Fe (mg kg−1)	31.8 ± 4.9	27.9 ± 1.88	37.0 ± 5.04
B (mg kg−1)	0.61 ± 0.021 *	0.65 ± 0.02 *	0.64 ± 0.05 *

(*) indicates significant differences at significance level p < 0.05 between values in rows for each parameter among different land use (LU).

).

To explain most of the spatial variability of soil properties in 364 fields, these fields were grouped according to existing land use (LU), while principal component analysis (PCA) was used. Factor analysis of soil properties among different LU such as pastureland/abandoned, annual, and permanent crops showed that upon varimax rotation, the first four, five, and six principal components were characterized as indicators, explaining 80.04, 71.30, and 76.84% of the total variance for each LU, respectively (

Table 5. Results of the principal component (PC) analysis of soil properties for different land uses.

Soil Properties	Land Use (LU)
	Pastureland					Permanent							Annual
	PC1	PC2	PC3	PC4	CE	PC1	PC2	PC3	PC4	PC5	PC6	CE	PC1	PC2	PC3	PC4	PC5	CE
pH	−0.912				0.902	−0.757						0.682		−0.694		−0.502		0.857
SOM		0.785			0.824		0.542			0.422		0.589	0.848					0.820
CaCO3	−0.493	0.747			0.826	−0.614			−0.433			0.644		−0.858				0.795
sand	0.764	−0.518			0.896		−0.714		0.558			0.910	−0.771					0.879
silt	−0.425		0.479		0.485				−0.926			0.892			−0.401	0.529	0.583	0.859
clay	−0.650	0.451	−0.517		0.893		0.869					0.827	0.949					0.912
EC		0.556	0.451		0.674			0.753				0.700		−0.465		0.551		0.622
Mg				0.690	0.682	0.463	0.540			−0.407		0.769	0.539					0.580
K	0.675			0.652	0.964					0.470		0.468			0.871			0.846
NO3-N			0.872		0.775			0.876				0.780				0.831		0.749
P				0.839	0.845	0.694						0.650		0.680				0.651
Cu	0.897				0.875					0.776		0.646					0.809	0.793
Zn		−0.409	0.759	0.407	0.909						0.807	0.674	0.449		0.661			0.724
Mn	0.839				0.747	0.488					0.650	0.760		0.815				0.743
Fe	0.978				0.978	0.777						0.738			0.538	0.494		0.644
B		−0.624			0.522			0.533				0.523			0.810			0.722
BD		−0.889			0.811		−0.891					0.868	−0.902					0.865
Eigenvalue	5.63	3.62	2.4	1.92		2.81	2.80	1.90	1.7	1.48	1.40		4.04	2.88	2.66	2.16	1.33
% of Variance	33.1	21.3	14.3	11.3		16.5	16.5	11.2	10.0	8.76	8.25		23.7	16.9	15.6	12.71	7.81
Cumulative %	33.1	54.4	68.7	80.04		16.5	33.0	44.2	54.3	63.0	71.30		23.7	40.7	56.3	69.0	76.84

The Rotation Method was the Varimax with Kaiser Normalization. Rotation converged in six interactions. CE stands for communalities estimates. Coefficients with absolute values less than 0.4 were suppressed. Significant loadings in PCA factors are shown in bold.

). After interpreting all components, the results are as follows:

In pastureland, PC1 was attributed to soil reaction, and its effect on soil nutrients concentration, explaining 33.1% of the variance. PC2 was attributed to soil organic matter and soil texture, explaining 21.3% of variance, and PC3 was attributed to manure due to pasture, explaining 14.3% of the variance. PC4 was attributed to conservative agriculture, explaining 11.3%.

In permanent crops, PC1 was attributed to soil reaction and its effect on soil nutrient concentration (16.5% of variance), and PC2 was attributed to soil organic matter and soil texture (16.5% of variance). PC3, PC5, and PC6 were attributed to anthropogenic interventions such as mineral fertilization (NO₃^-) and pesticide application (Cu and Zn, respectively), while PC4 was attributed to geogenic factors such as soil texture.

In annual crops, PC1 was attributed to soil organic matter and soil texture (23.7% of variance), and PC2 was attributed to soil reaction and its effect on soil nutrient concentration (16.9% of variance). PC3 and PC4 were attributed to mineral fertilization (K, NO₃⁻), while PC5 was attributed to pesticide application (Cu).

3.3. Interpreting the Spatial Variability of Soil Properties across Different Soil Units (SU)

Based on the national soil maps, the distribution (percentage) of fields in different soil units (SU) was as follows: 41% Fluvisols (FV), 29% Luvisols (LV), 19% Cambisols (CM), and 16% Calcisols (CL), as presented in Figure 3. Among different SUs, statistically significant differences were observed in 16 of the 17 analyzed properties. The only soil parameter that did not differ significantly among soil types was the soil electrical conductivity (ED), as shown in

Table 6. Least square means with standard error (SE) of soil chemical and physical parameters in different soil units (SU).

Soil Properties	Soil Unit (SU)
	Fluvisols (n = 148)	Calcisols (n = 52)	Cambisols (n = 59)	Luvisols (n = 105)
pH	7.30 ± 0.07 *	7.98 ± 0.04 *	7.23 ± 0.07 *	7.45 ± 0.06 *
SOM (%)	2.45 ± 0.08 *	1.71 ± 0.09 *	1.44 ± 0.07 *	2.6 ± 0.11 *
BD (g cm−3)	1.42 ± 0.01 *	1.44 ± 0.01 *	1.45 ± 0.01 *	1.40 ± 0.01 *
CaCO3 (%)	5.59 ± 0.44 *	17.4 ± 1.60 *	4.80 ± 0.88 *	5.84 ± 0.67 *
Sand (%)	38.9 ± 1.19 *	35.1 ± 2.08 *	35.7 ± 1.99 *	33.1 ± 1.13 *
Silt (%)	33.3 ± 0.80	37.1 ± 1.56	34.2 ± 1.68	33.4 ± 1.01
Clay (%)	27.7 ± 0.84 *	27.8 ± 1.49 *	30.1 ± 1.49 *	33.1 ± 0.93 *
EC (dS m−1)	0.73 ± 0.04	0.87 ± 0.08	0.74 ± 0.05	0.80 ± 0.04
Mg (mg kg−1)	176 ± 7.58 *	148 ± 13.8 *	167 ± 13.13 *	193 ± 7.21 *
NO3-N (mg kg−1)	11.1 ± 1.10	11.8 ± 2.45	14.6 ± 2.75	13.4 ± 1.86
P (mg kg−1)	29.1 ± 2.55 *	8.75 ± 0.86 *	36.1 ± 5.51 *	38.4 ± 3.69 *
K (mg kg−1)	222 ± 10.7 *	181 ± 16.0 *	282 ± 34.5 *	272 ± 14.5 *
Cu (mg kg−1)	4.67 ± 0.44 *	3.41 ± 0.70 *	4.64 ± 0.61 *	4.51 ± 0.41 *
Zn (mg kg−1)	1.48 ± 0.13	0.79 ± 0.07	2.99 ± 1.44	2.00 ± 0.26
Mn (mg kg−1)	13.6 ± 1.25 *	8.25 ± 0.68 *	10.9 ± 1.18 *	12.1 ± 0.73 *
Fe (mg kg−1)	36.9 ± 3.63	22.1 ± 3.33	24.0 ± 1.93	26.6 ± 1.73
B (mg kg−1)	0.63 ± 0.02 *	0.56 ± 0.03 *	0.71 ± 0.05 *	0.68 ± 0.02 *

(*) indicates significant differences at significance level p < 0.05 between values in rows for each parameter among different soil units (SU).

. However, the results showed that Cambisols are characterized by limited organic matter, in Fluvisols the fertility status varies due to alluvial deposits, Calcisols have a significant amount of calcium carbonates, and Luvisols are characterized as the most fertile soils (Table 6).

To explain most of the spatial variability of soil properties in 364 fields, these fields were grouped according to soil unit (SU) and principal component analysis (PCA) was used. Factor analysis of soil properties among SUs concluded that upon varimax rotation, the first six (for FV, CL CM) and the seventh (for LV) principal components were retained as indicators with high factor loading, explaining 78.9%, 81.1%, 82.0%, and 80.5% of the total variance for each SU, respectively (

Table 7. Results of the principal component (PC) analysis of soil properties for different soil units (SU).

Soil Properties	Soil Unit Types (SUT)
	Fluvisols (FV)							Luvisols (LV)
	PC1	PC2	PC3	PC4	PC5	PC6	CE	PC1	PC2	PC3	PC4	PC5	PC6	PC7	CE
pH	−0.888						0.826		-0.615						0.663
SOM		0.736	−0.414				0.855	0.778							0.837
CaCO3	−0.541						0.606		−0.819						0.751
sand		−0.669	−0.650				0.962	−0.740					0.432		0.915
silt			0.867				0.804				−0.680		−0.426	0.422	0.909
clay		0.842					0.828	0.899							0.856
EC				0.804			0.804			0.893					0.869
Mg					0.803		0.755	0.460						0.626	0.797
K	0.600						0.641					0.751			0.770
NO3-N				0.883			0.854			0.932					0.895
P					0.697		0.744				0.625	0.463			0.775
Cu	0.777						0.757					0.776			0.679
Zn						0.810	0.815						0.908		0.879
Mn	0.846						0.771		0.727						0.651
Fe	0.857						0.769							0.821	0.796
B						0.699	0.727				0.818				0.777
BD		−0.914					0.898	−0.909							0.873
Eigenvalue	4.09	2.90	1.76	1.68	1.54	1.44		3.40	2.01	1.91	1.81	1.57	1.51	1.49
% of Variance	24.0	17.0	10.4	9.90	9.07	8.49		20.0	11.8	11.2	10.6	9.21	8.88	8.74
Cumulative %	24.0	41.1	51.5	61.3	70.4	78.9		20.0	31.8	43.1	53.7	62.9	71.8	80.5

The Εxtraction Method was the Principal Component Analysis. The Rotation Method was the Varimax with Kaiser Normalization. Rotation converged in six interactions. CE stands for communalities estimates. Coefficients with absolute values less than 0.4 were suppressed. Significant loadings in PCA factors are shown in bold.

and

Table 8. Results of the principal component (PC) analysis of soil properties for different soil units (SU).

Soil Properties	Soil Unit Types (SUT)
	Calcisols (CL)							Cambisols (CM)
	PC1	PC2	PC3	PC4	PC5	PC6	CE	PC1	PC2	PC3	PC4	PC5	PC6	CE
pH			0.884				0.831		0.415		−0.465		−0.437	0.684
SOM	0.548				−0.667		0.770		0.855					0.776
CaCO3						−0.890	0.823		0.795					0.845
sand	−0.848						0.967	−0.881		0.427				0.972
silt			0.668		0.443		0.861			−0.919				0.918
clay	0.818						0.828	0.895						0.913
EC					0.821		0.749		0.750				0.404	0.883
Mg						0.830	0.807			0.626				0.673
K		0.708					0.815					0.757		0.741
NO3-N				0.736	0.427		0.843						0.899	0.833
P				0.768			0.860					0.680	0.407	0.816
Cu				0.831			0.770	0.463		0.410			0.484	0.704
Zn		0.786					0.798				0.937			0.886
Mn			0.716				0.725				0.893			0.872
Fe		0.819					0.716					0.816		0.790
B		0.800					0.724			0.587				0.718
BD	−0.928						0.900	−0.929						0.919
Eigenvalue	3.17	2.77	2.13	2.11	1.84	1.77		3.30	2.59	2.26	2.03	1.89	1.88
% of Variance	18.7	16.3	12.5	12.4	10.8	10.4		19.4	15.2	13.3	11.9	11.1	11.1
Cumulative %	18.7	35.0	47.5	59.9	70.7	81.1		19.4	34.6	47.9	59.8	71.0	82.0

The Εxtraction Method was the Principal Component Analysis. The Rotation Method was the Varimax with Kaiser Normalization. Rotation converged in six interactions. CE stands for communalites estimates. Coefficients with absolute values less than 0.4 were suppressed. Significant loadings in PCA factors are shown in bold.

).

After interpreting all components, the results of the study were the following:

In Fluvisols (FV), PC1 was attributed to soil reaction, and its effect on soil nutrients concentration, explaining 24.0% of the variance. PC2 and PC3 were attributed to soil organic matter and soil texture, explaining 27.4% of variance, PC4 was attributed to the correlation of nitrate soil concentration (NO₃-N) with EC (9.9% of variance), and PC5 and PC6 were attributed to anthropogenic interventions or conservative agriculture, explaining 17.6% of variance (Table 7).

In Luvisols (LV), PC1 was attributed to soil texture and to soil organic matter content, explaining 20.0% of variance, PC2 was attributed to CaCO₃ soil content and soil reaction (11.8% of variance), PC3 was attributed to correlation of nitrate soil concentration (NO₃-N) with EC (11,2% of variance), and PC4, PC5, PC6, and PC7 were attributed to geogenic factors or anthropogenic interventions such as mineral fertilization (P, K) or to pesticide application (Cu), as well as to the soil texture effect on nutrient availability, explaining 37.5% of variance (Table 7).

In Calcisols (CL), PC1 was attributed to soil texture and less to soil organic matter, explaining 18.7% of variance, PC2 was attributed to geogenic factors or anthropogenic interventions (16.3% of variance), PC3 was attributed to soil reaction (12.5% of variance), PC4 and PC5 were attributed to anthropogenic interventions or conservative agriculture (23.2% of variance), and PC6 was attributed to CaCO₃ soil content and its effect on soil nutrients availability, explaining 10.4% of variance (Table 8).

Finally, in Cambisols (CM), PC1 and PC3 were attributed to soil texture and its effect on soil nutrients availability, explaining 32.7% of variance, PC2 was attributed to soil organic matter and soil reaction (15.2% of variance), and PC4, PC5, and PC6 were attributed to geogenic factors (e.g., Mn and Fe), or anthropogenic interventions such as mineral fertilization (N, P, K), as well as to the soil reaction effect on nutrient availability, explaining 34.1% of variance (Table 8).

4. Discussion

Through exploring the spatial variability of soil chemical and physical properties in Mediterranean small farms, it was observed that they were significantly affected by both land use and soil unit. Therefore, soil function and fertility status, which are vital for crop production and soil conservation, could be better explained using soil and land use/cover maps. In large agricultural areas occupied by small farms size (aver. 2.9 ha), high spatial variability of soil properties was observed, probably due to the greater variability of cultivated species per unit of cropland (ha), as well as the different approaches and empirical applications of cultivation practices by smallholder farmers. Also, PCA analysis was used as a tool to select soil indicators from a large dataset (17 soil parameters), creating a minimum dataset (MDS) of soil-sensitive indicators in a regional level, capable of being used to explain the variability of most soil properties, and as a decision support tool to implement effective best management practices.

As know, soil properties vary in different spatial areas due to the combined effect of physical, chemical, and biological processes over time [34]. However, the spatial variability of soil properties in agricultural areas is greater when compared to other natural (undisturbed) soils, as it is affected by cultivated plants and agricultural practices [35]. Similar results were observed in our study, showing significant differences in 12 of 17 soil properties analyzed after sampling in 364 farms of small size with different land use/cover (Table 4). Detailed soil fertility situation of the study area is described in Table 1. However, in the total study area, our results showed that in Calcaric soils of the Aitoloakarnania region [7] the high content of CaCO₃ increased soil pH (Figure 2a). However, lower soil pH values increased the available Olsen P concentrations (Figure 2a), and these high P concentrations were associated with high available Fe and K concentrations (Figure 2b), which is in accordance with Moore et al. [36]. Moreover, significant correlations were observed between EC values and the concentration of nitrates, the soil bulk density, the organic matter and the soil texture, as well as among other soil physicochemical properties (Table 1). These findings coincide with the findings of Patriquin et al. [37] and Arshad et al. [38]. Sillanpää [39] reported that it is often difficult to define which soil factor affects directly the availability of a certain nutrient and which factor is only in “pseudocorrelation” with that nutrient, though “pseudocorrelations” might be equally informative. The aforementioned highlight the complicated relationships among the physicochemical soil properties and suggest that soil quality cannot be measured directly, but could be deduced from chemical, physical, and biological soil properties (soil health indicators), environmental conditions, and local cultivation practices [40,41].

Taking into consideration the uncontrolled empirical use of fertilizers from smallholder farmers in the study area, without any recommendations and the need to implement effective best management practices, principal component analysis (PCA) was used to define soil indicators from a large data set (17 soil properties) that make up a minimal data set (MDS), which would facilitate the identification of the most appropriate indicators that can better explain soil variability [42]. However, factor analysis of soil properties in the total study area (Stotal) reported that upon varimax rotation, six principal components could explain only 70.987% of the total variance (Table 3). Therefore, our results suggest that in different land use/cover and soil units, and especially when agricultural land is occupied by smaller farms, it is difficult to explain the variability of soil properties. As it was observed in a recent Decision Support System (DSS) which was developed in Northern and Central European countries [43], there was a lack of the farm size parameter in the soil navigator due to the fact that it is not considered an important factor in these EU regions. Our results show that this parameter is crucial for the development of DSS in the Mediterranean region due to the existing small farms. Therefore, the structural differences of the European agricultural sector at a regional or spatial scale should be taken into account for the implementation of European agri-environmental policies and measures for long term soil conservation [15].

Among different land uses (LU), such as pastureland/abandoned, annual, and permanent crops, factor analysis of soil properties showed that upon varimax rotation, the principal components were characterized as indicators, explaining 80.04%, 71.30%, and 76.84% of the total variance for each LU, respectively (Table 5). The soil assessment, based on LU, revealed a greater homogeneity in pasturelands, explaining 80.04% of the total soil properties’ variance (Table 3), indicating the application of non-intensive agriculture practices. Our results showed that the highest percentage of soil organic matter and the lowest pH values were observed in the pastures when compared to the annual and permanent crops, confirming the above indication (Table 4). Our findings are in accordance to Goulding [44], who reported that soil pH in arable soils (mean pH 6.71) was higher than in grassland soils (mean pH 5.98) and observed that soil pH can be strongly influenced by the installation of a crop and the application of fertilizers. For instance, studies have demonstrated that pH decreases due to the continuous cultivation of clover or other legumes [45,46,47]. The results of the present study also indicate that micro-nutrients appear as indicators only in permanent and annual crops (in permanent crops Zn and Cu, while in annual crops only Cu, due to continuous use of mineral fertilizers and pesticides). In some cases, these changes can mask the parent material’s contribution, reflecting soil health [48,49]. However, Navas and Machinto [50] reported that the natural levels of Cu and other elements varied among different soil groups. Among different soil units (SU), factor analysis of soil properties showed that upon varimax rotation, the principal components were characterized as indicators, explaining 78.9, 81.1, 82.0, and 80.5% of the total variance for Fluvisols, Calcisols, Cambisols, and Luvisols respectively (Table 7 and Table 8). As our findings showed, the use of soil maps can efficiently deliver soil information to meet user needs for soil management and crop performance decisions, which is in accordance with previous studies [7,51]

As was observed, the spatial variability in agricultural land is attributed both to geogenic factors and anthropogenic interventions. Principal component analysis (PCA) showed that the clustering of soil properties based on the SU and LU could better describe the variability of soil properties in Mediterranean smallholder farming, which was in line with previous studies [52,53]. Bünemann et al. [54] mentioned that the frequency (percentage) with which soil indicators are used to assess soil quality was as follows: soil organic matter 90%, pH 81%, bulk density 53%, soil texture 44%, P 72%, K 49%, N 40%, EC 33%, heavy metals 22%, other macronutrients and micronutrients 16%. The frequency (percentage) of the soil indicators was presented with high factor loading and better explained the soil properties’ variability in different land use/land cover and soil units in the study area as follows: bulk density 86%, NO₃-N 86%, Fe 71%, soil texture 57%, Cu 57%, Zn 57%, pH 29%, soil organic matter 14%, available P 14%, K 14%, EC 14%, B 14%. These soil indicators will allow farmers to assess soil health and fertility on existing small holdings, thus enabling them to make more cost-effective and efficient management decisions, maintain sustainable productivity and avoid adverse environmental impacts. It is believed that the analysis and interpretation of soil parameters in different land use/cover and soil units, as well as the identification of sensitive soil indicators after analysis (PCA) at a regional level, can be visualized through the development of digital soil maps and decision support systems in order to assess and optimize soil functions, aiming for the sustainable management of rural areas.

5. Conclusions

Significant spatial variability of soil physical and chemical properties was observed in Mediterranean agricultural areas occupied by small farms. To avoid the uncontrolled empirical use of fertilizers without any recommendation or the use of nutrient site-specific management to improve the soil fertility and crop productivity as required, a holistic approach was attempted. Principal component analysis (PCA) showed that the clustering of soil properties based on soil units (SU) and land use/cover (LU) could better describe the variability of soil properties. Regarding SU, our results indicate that six to seven soil indicators with high factor loading can explain up to 82.0% of the variability of soil properties analyzed. Furthermore, regarding LU, soil indicators with high factor loading can explain 80.0% of soil properties’ variability in pasturelands, as well as 76.8% and 71.3% in annual and perennial crops, respectively. Our study proposes that an MDS approach of soil chemical and physical indicators could explain the spatial variability of soil physical and chemical properties, providing baseline information, whilebeing at the same time an appropriate and effective management practice in the smallholder farming systems.