Scrutinizing the Statistical Distribution of a Composite Index of Soil Degradation as a Measure of Early Desertification Risk in Advanced Economies

Abstract

Using descriptive and inferential techniques together with simplified metrics derived from the ecological discipline, we offer a long-term investigation of the Environmental Sensitive Area Index (ESAI) as a proxy of land degradation vulnerability in Italy. This assessment was specifically carried out on a decadal scale from 1960 to 2020 at the province (NUTS-3 sensu Eurostat) level and benefited from a short-term forecast for 2030, based on four simplified assumptions grounded on a purely deterministic (‘what … if’) approach. The spatial distribution of the ESAI was investigated at each observation year (1960, 1970, 1980, 1990, 2000, 2010, 2020, 2030) calculating descriptive statistics (central tendency, variability, and distribution shape), deviation from normality, and the increase (or decrease) in diversification in the index scores. Based on nearly 300 thousand observations all over Italy, provinces were considered representative spatial units because they include a relatively broad number of ESAI measures. Assuming a large sample size as a pre-requisite for the stable distribution of the most relevant moments of any statistical distribution—because of the convergence law underlying the central limit theorem—we found that the ESAI scores have increased significantly over time in both central values (i.e., means or medians) and variability across the central tendency (i.e., coefficient of variation). Additionally, ecological metrics reflecting diversification trends in the vulnerability scores delineated a latent shift toward a less diversified (statistical) distribution with a concentration of the observed values toward the highest ESAI scores—possibly reflecting a net increase in the level of soil degradation, at least in some areas. Multiple exploratory techniques (namely, a Principal Component Analysis and a two-way hierarchical clustering) were run on the two-way (data) matrix including distributional metrics (by columns) and temporal observations (by rows). The empirical findings of these techniques delineate the consolidation of worse predisposing conditions to soil degradation in recent times, as reflected in a sudden increase in the ESAI scores—both average and maximum values. These trends underline latent environmental dynamics leading to an early desertification risk, thus representing a valid predictive tool both in the present conditions and in future scenarios. A comprehensive scrutiny of past, present, and future trends in the ESAI scores using mixed (parametric and non-parametric) statistical tools proved to be an original contribution to the study of soil degradation in advanced economies.

1. Introduction

Soil and landscape degradation are key ecological issues that contemporary societies have to face because of the serious consequences they pose to human health and environmental quality at large [1,2,3]. On the one hand, this issue has been sometimes raised in the midst of media attention, decision makers, and public opinion [4,5,6]; on the other hand, the essentially cyclical nature of this interest corresponded, especially in its peak stages, to the outbreak of emergency situations associated with prolonged droughts and water scarcity—all processes that are easily (but sometimes erroneously) associated with the issue of climate change [7,8,9]. This media interest has directed the attention of the general public to the desertification–climate relationship—and more generally on the intrinsic changes in the bio-physical factors underlying landscape degradation—neglecting the important role played by social, economic, cultural, political, and institutional factors [10,11,12]. This role, brought to the fore by recent environmental dynamics, requires the application of dedicated and mixed approaches to any relevant scientific perspective, aiming to ensure a more conscious and less sensationalistic dissemination of positive results, which are also of interest for policy application [13,14,15,16].

Based on a misinterpreted principle that claims man’s right and supremacy over the environment, industrial development and technological progress—especially in advanced countries—have generally been pursued at the expense of soil resources [17]. Since the 1950s, theoretical frameworks have emerged that assume the complex interplay between economic and environmental systems, in turn recognizing the influence of biophysical factors on human societies and the relevance of a bi-directional relationship between natural and social systems [18,19,20]. The notion of ‘carrying capacity’ in socioeconomic evaluations was later introduced with reference to the ineluctable submission of human societies to the limits set by natural environments and the cogency of ecological laws [21,22,23]. This ‘ecological paradigm’ highlighted the need of a complex (systemic) organization grounded on self-relations [24,25,26]. These constitute an example of the continuously evolving interplay between scientific knowledge and practice [27]. In this sense, ecology and environmental disciplines at large, easily and rapidly became the first sciences schematizing the relationship between humanity and nature, considering ‘Man as part of nature’ rather than its master [28,29,30]. In this conceptual framework, soil degradation is assumed here as the result of paradigmatic interactions between man and nature, which may also involve other aspects of the human–nature ‘complexity nexus’ [31,32,33], possibly representative of the intrinsic environmental and socioeconomic dynamics exemplified in the landscapes illustrated in Figure 1 [34]. A long-term environmental assessment scheme allowing the honest quantification of such processes is especially needed in advanced economies, where ecological information is often available but sometimes less coordinated and comparatively weak over time and/or space [35,36,37].

Considering the relatively long time interval between 1960 and 2020, and incorporating a deterministic projection to 2030 under four different and simplified short-term scenarios, the spatial distribution of the Environmentally Sensitive Area Index (ESAI), a renowned proxy of land degradation vulnerability [38], was investigated at each observation year (one per decade), calculating descriptive statistics (central tendency, variability, and distribution shape), deviations from normality, and the increase (or decrease) in diversification in the index scores [39,40,41]. Based on nearly 300 thousand observations all over Italy, provinces were regarded as representative spatial units for such a kind of study because they share a sufficiently large number of observations, thus assuring statistical stability to all the adopted indicators [42,43,44]. Multiple exploratory techniques (namely, a Principal Component Analysis and a two-way hierarchical clustering) were run on the data matrix of distributional metrics and time observations [45]. A purely statistical approach, like the one proposed here, may reorient both conceptual and operational frameworks toward a permanent, quantitative assessment of soil degradation in advanced economies [46]. Re-interpreting past, present, and future trends in soil degradation and the newly emerging dynamics of human pressure and local warming in a representative Mediterranean country, may finally offer a comprehensive perspective informing zero-net degradation strategies [47] and containing early desertification risks in Europe.

2. Methodology

2.1. Study Area

The degree of soil degradation based on a composite index was studied over the whole Italian territory [48]. From an administrative perspective, Italy is divided into three geographical areas (North, Central, South), twenty NUTS-2 administrative regions, and 110 NUTS-3 provinces (2007 reference setting), covering a total surface area of nearly 301,330 km² [49]. Italy displays important disparities in economic growth, social development, and natural resource availability, and the country appears a relevant case study to address the interaction of biophysical and socioeconomic dimensions that predispose land to degradation processes [50].

2.2. Data and Variables

The analysis was carried out at a national scale using the Environmental Sensitive Areas (ESA) model [51], which is capable of estimating the vulnerability of the territory to LD (Land Degradation) through the definition of four ‘quality’ concepts referring specifically to four corresponding dimensions underlying land degradation as follows: climate, soil, vegetation/land use, land management (Figure 2). In order to study the evolution of LD over time, the period investigated (1960–2020) was divided into ten-year sub-periods.

Being specifically designed, tested, and ground validated for Mediterranean regions [52], the Environmentally Sensitive Area Index (ESAI) is a composite indicator of the level of land vulnerability to degradation in a given location as a combination of inappropriate land management, low quality soils, dry climate regime, and poor vegetation cover [53,54,55]. In line with the current literature, the present study utilizes the comparable data required to develop the standard ESAI model [56,57] by adopting information layers that are reliable, comparable, and up-to-date at both regional and local scales in Mediterranean countries [56], covering sixty years between 1960 and 2020 [58].

Climate quality was estimated adopting mean annual precipitation, a standard aridity index, and slope aspect [59]. Regionalized annual precipitations and aridity indexes (calculated as the ratio of precipitation to reference evapotranspiration according to the United Nations Environmental Program, UNEP) were computed as decadal means from data derived from the National Agro-meteorological Database of the Italian Ministry of Agriculture and Forestry [60]. The slope aspect was calculated from the ASTER (Advanced Spaceborne Thermal Emission and Reflection Radiometer) database representing a 30 m resolution Digital Elevation Model (DEM) [14]. Soil depth and texture, slope and nature of parent material, taken as proxies of soil quality, were estimated from a 1 km² resolution digital map released by the European Soil Database (Joint Research Center, Ispra) [12]. We assume soil variables remain stable over the time period investigated in this study, since they are determined by the joint action of pedogenesis [11,19].

The role of vegetation cover and land-use was estimated considering the four following standard variables [4]: vegetation cover, wildfire risk, the level of protection offered by vegetation against soil erosion, and the degree of drought resistance reflected in the current vegetation cover [10]. These indicators were derived from elaborations on comparable maps as follows: (i) a Corine-like ‘Land Use Map of Italy’, released by the National Research Council (CNR) and the Italian Touring Club (TCI) for 1960, and (ii) two Corine land cover maps, respectively dated 1990 and 2018 [3], in accordance with the technical operations described in earlier studies [16]. Land management was quantified as the result of population dynamics and land-use transformations [17], considering (i) demographic density and (ii) the annual growth rate—both measured at the municipal scale using census data derived from the Italian National Statistical Institute (Istat) [7], as well as an indicator of land-use intensity, extracted from the digital maps mentioned above by classifying each land type in accordance with the assumed intensity of economic use [56]. All the variables related to the four dimensions (climate, soil, vegetation, human pressure) were translated into numeric scores in accordance with the specifications reported in a number of earlier studies [49,50,57,61].

The operational scheme produced four (partial) indicators—Climate Quality Index (CQI), Soil Quality Index (SQI), Vegetation Quality Index (VQI), and land Management Quality Index (MQI)—from the aggregation of the 14 variables described above. Indicators were composed of the geometric mean of the different scores assigned to each input variable, following a standard scoring system [62]. The final ESAI value was estimated at each spatial unit (1 km² grid) as the geometric mean of the four indicators transformed into a score ranging from 1 (the lowest level of degradation) to 2 (the highest level of degradation), based on a standard scoring system [63]. The ESAI scoring system was extensively verified in the field, both directly [64] and indirectly [65]. The ESAI was subsequently estimated at each i-th spatial unit and j-th year as the geometric mean of the four partial indicators [66] as follows:

ESAI_i,j = (SQI_i,j × CQI_i,j × VQI_i,j × MQI_i,j)^1/4

The ESAI scores range from 1 (the lowest land sensitivity to desertification) to 2 (the highest sensitivity to desertification). Four classes of land vulnerability were identified based on the ESAI figures, which reflect the most used classification thresholds [48] as follows: (i) unaffected areas (ESAI < 1.17), (ii) potentially affected areas (1.17 < ESAI < 1.225), (iii) ‘fragile’ areas (1.225 < ESAI < 1.375), and (iv) ‘critical’ areas (ESAI > 1.375). Intermediate and final maps were produced (Figure 3) after the various elementary layers were registered and referenced to an elementary pixel of 1 km² [63].

2.3. Indicators and Future Scenarios

In order to acquire further evaluation elements, we considered the future evolution of the ESAI on the basis of statistical projections to 2030. Four different scenarios were considered (S1, S2, S3, and S4), reflecting an increase in the average aridity index (both (i) +5%, intended as a moderate increase, and (ii) +10%, intended as a more marked increase) and the demographic dynamics that relate to (iii) population stability, and (iv) the trend reflecting an identical rate of demographic growth observed between 2011 and 2021 in each municipality of Italy (

Table 1. The main characteristics of 2030 land degradation (ESAI) scenarios for Italy.

Scenario	Increasing Climate Aridity		Demographic Dynamics
	+5%	+10%	Stable	Increasing
S1	●		●
S2		●	●
S3	●			●
S4		●		●

). Scenarios were applied from a purely ‘what … if’ (mechanistic) perspective [47] and should not be regarded as stochastic forecasts based on a given probability threshold [65]. Nevertheless, they are inclusive of the most plausible hypotheses in both the climate and demographic fields for the study area (Italy as a whole). This is widely demonstrated by both the continuous increase in air temperature caused by global warming, which is reflected by worse soil aridity conditions, and the spatially varying demographic conditions, which may result in polarization between the most accessible locations (with population still growing after decades of continuous expansion) and the more remote places (with population declining rapidly) because of land abandonment and aging [67].

2.4. Data Analysis

A mixed data analysis strategy was carried out considering sequential stages [68] that included the use of (i) descriptive statistics, (ii) inferential tests, (iii) ecological diversity metrics, and (iv) a multivariate approach based on three separate procedures summarizing the complex relationship at the base of distributional attributes over time [69]. Descriptive metrics, tables, and graphs were initially used to illustrate the statistical distribution of the ESAI scores across the Italian provinces (n = 110). In this study, spatial calculations were carried out using ArcGIS release 10 (Redwoods, CA, USA), specifically the ‘zonal statistics’ and ‘tabulate area’ commands [49]. More specifically, descriptive metrics were divided into three sub-groups [13], namely, (i) simple descriptive statistics of central tendency, dispersion, and distributional shape; (ii) inferential statistics testing the departure of the given statistical distribution from normality; and (iii) ecological metrics evaluating the intrinsic diversity and evenness of the statistical distribution under scrutiny [16]. Such measures were calculated at ten points in time (every decade from 1960 to 2020 and at four different scenarios (S1–S4) for 2030).

More specifically, the descriptive statistics of the ESAI distribution by year included specific measures of central tendency (arithmetic and geometric means, median), dispersion (minimum, maximum, 25th and 75th percentile, absolute (normalized) range, interquartile range, coefficient of variation), and shape (skewness, kurtosis, median-to-mean ratio). Inferential tests included four statistics for normal distribution of several samples of univariate data [4]. For all tests implemented here, the null hypothesis is H₀ (the sample was taken from a population with normal distribution); it means that, if the given p (H₀: normal) is less than 0.05, the normal distribution can be rejected [9]. Of the given tests, the Shapiro–Wilk and Anderson–Darling tests were assumed to be precise, and the Lilliefors and Jarque–Bera tests were given for reference and internal control [14]. Since these tests were running jointly on seven samples, the multiple testing issue was controlled using Bonferroni’s correction for multiple comparisons [12]. Finally, the input of the five metrics derived from information theory and ecological science were identified, as follows [3]: (a) evenness (namely Pielou’s evenness) index, computed as the Shannon diversity index (H’) divided by the logarithm of T, the number of non-null observations; (b) Brillouin’s index (B), another measure of diversity in the rates of the ESAI calculated as follows: B = (ln(T!)-∑_iln(ESAI!))/T; and (c) Menhinick’s richness index (M), providing a gross estimation of heterogeneity in a statistical distribution as M = T/√N, where N is the total number of records [15]. The Fisher (alpha) and Berger–Parker diversity indexes completed and enriched the picture provided by the statistical metrics of the ESAI distribution over time [46].

Multidimensional Analysis

The latent relationship between the selected distributional metrics (described above) for the ESAI scores in Italy at ten specific observation times (1960, 1970, 1980, 1990, 2000, 2010, 2020, and four short-term scenarios (S1 to S4) for 2030) was reflected in a 10 (columns) per 20 (rows) row-standardized matrix subjected to a sequential multivariate statistical analysis [68], namely, (i) a Principal Component Analysis (PCA) reducing the dimensionality of the input matrix and extracting the most relevant, non-redundant analysis’ components based on a correlation philosophy (specifically, exploring the correlation among input variables) [66], and (ii) a two-way clustering delineating convergence or divergence across distributional metrics and years via a similarity philosophy (namely, the use of agglomeration techniques based on Ward’s metric), used as a confirmative tool of the previous results [61]. In line with the foundational literature of multivariate statistics [44], two criteria identifying the latent relationship between variables and cases (as described above) were adopted [17], considering separately the correlation and similarity among variables and cases in the same geometric (representation) space [45].

Principal Component Analysis (PCA) identifies hypothetical variables (namely, components, illustrated as a ‘biplot’ axis) that explain the largest percentage of the variance in a given data matrix, organized as a two-way dataset with variables by column and observations by row [63]. Axes were artificially constructed as linear combinations of the original variables maximizing the individual information from the data matrix [56]. Eigenvalues, reflecting the variance associated with each axis, were obtained from the spectral decomposition of the correlation matrix using the Singular Value Decomposition (SVD) algorithm [62]. Axes with eigenvalues >1 were selected and analyzed via loadings (reflecting the intrinsic correlation between observation years and axes) and scores (that reflect the intrinsic linkage between distributional metrics and axes) into a biplot summarizing their position in the same geometric plane [12]; this plot was adopted to summarize the outcome of PCA [4]. Two-way hierarchical clustering based on Euclidean distances under the operational assumption of Ward’s agglomeration rule was additionally adopted here as an assumption-free, flexible technique exploring the multivariate similarity among distribution metrics varying over time and providing an aggregate (graphical) representation of the metrics’ similarity based on dendrograms [57]. Similarity was represented in the two-way dendrogram plot directly based on Ward’s metric [60].

3. Results

3.1. A Descriptive Analysis of a Composite Index of Soil Degradation

Figure 4 delineates the statistical distribution of the ESAI by year (1960–2020), considering in turn four short-term scenarios (S1–S4) for 2030. A graphical representation based on the mean and whisker plots was used here. The graph illustrates a rather clear trend of increasing ESAI scores over time, on average, shifting from scores around 1.34 to scores close to 1.37. Based on the standard classification of the ESAI, a score ranging from 1.225 to 1.375 reflects a ‘fragile’ area, basically representing an intermediate vulnerability class; a score above 1.375 indicates a ‘critical’ area, more specifically, the most exposed class to soil degradation. The average shift illustrated in this graph suggests that, on average, a substantial portion of the Italian provinces moved from a ‘fragile’ class to a barely ‘critical’ class. The following analysis will provide further details from this perspective.

3.2. Distributional Metrics Delineating Trends over Time in a Composite Index of Soil Degradation

Table 2. Analysis of the statistical distribution of the ESAI score across Italian provinces (NUTS-3 level, n = 110), considering metrics of central tendency (minimum, maximum, arithmetic mean, median, geometric mean), dispersion (coefficient of variation, absolute range, 25th percentile, 75th percentile, interquartile range), and shape (Mean to Median: MtoM, skewness, kurtosis) by year (1960–2020) or 2030 scenario (S1 to S4).

Statistic	1960	1970	1980	1990	2000	2010	2020	S1	S2	S3	S4
Min	1.261	1.273	1.258	1.248	1.255	1.263	1.244	1.255	1.256	1.266	1.266
Max	1.487	1.542	1.523	1.480	1.475	1.494	1.466	1.483	1.485	1.488	1.485
Mean	1.345	1.367	1.365	1.358	1.363	1.371	1.368	1.371	1.382	1.369	1.380
Median	1.337	1.361	1.366	1.359	1.368	1.372	1.369	1.372	1.385	1.366	1.382
MtoM	1.006	1.004	0.999	0.999	0.996	0.999	0.999	0.999	0.998	1.002	0.998
Geom. Mean	1.344	1.366	1.364	1.357	1.362	1.370	1.367	1.370	1.381	1.368	1.379
Coeff. Var.	0.034	0.040	0.042	0.039	0.039	0.038	0.036	0.037	0.036	0.037	0.036
Abs. Range	0.168	0.197	0.194	0.171	0.161	0.168	0.162	0.166	0.165	0.162	0.158
25 prcntil	1.309	1.321	1.319	1.317	1.317	1.327	1.330	1.333	1.345	1.332	1.342
75 prcntil	1.375	1.403	1.408	1.403	1.407	1.411	1.406	1.410	1.419	1.408	1.419
Intq. Range	0.049	0.060	0.065	0.063	0.065	0.060	0.055	0.056	0.053	0.055	0.055
Skewness	0.728	0.657	0.242	0.0188	−0.149	−0.026	−0.090	0.067	−0.054	0.088	−0.030
Kurtosis	0.412	0.213	−0.396	−0.866	−0.984	−0.701	−0.602	−0.628	−0.627	−0.655	−0.699

provides more explicit values delineating the conditions that predispose soil to degradation, as observed over the last 60 years. As a matter of fact, the average ESAI increased from 1.345 (1960) to 1.37 (2020) and to 1.37–1.38 (2030), depending on the considered scenario (S1–S4). S2 and S4 were the worst scenarios; S1 and S3 were assumed to be neutral scenarios, basically remaining in line with 2020 or only slightly worse conditions. Considering median values over time, this latent trend turned out to be even more evident. At the same time, considering the minimum and maximum values of the ESAI statistical distribution revealed a substantial stability of minimum scores, around 1.26, meaning that the number of unexposed areas remained stable, or slightly improved over time. Interestingly, the highest values in the statistical distribution were observed in the 1970s and the 1980s (around 1.54).

Values around 1.48–1.49 were, instead, found both at the beginning and the end of the study period, indicating a sort of distributional stability at the highest scores, i.e., those reflecting the greatest intensity of soil degradation at the local scale. Consequently, the coefficient of variation, and other dispersion metrics, exhibited the same pattern over time, displaying a moderate increase between 1960 and 1990 and a more evident stability between 1990 and 2020, which was in turn maintained in the four 2030 scenarios. The highest symmetry in the statistical distribution of the ESAI was found between 1990 and 2020, in contrast with the asymmetric distribution observed in the early 1960s. Skewness and kurtosis finally documented a moderate shift between rather asymmetric distributions in the first observation years (1960–1990) and more regular and shaped distributions in the more recent observation years, both observed (1990–2020) and predicted (2030: S1–S4).

Table 3. Analysis of the statistical distribution of the ESAI score across Italian provinces (NUTS-3 level, n = 110), considering inferential tests for normality by year (1960–2020) or 2030 scenario (S1 to S4).

Inference	1960	1970	1980	1990	2000	2010	2020	S1	S2	S3	S4
Shapiro–Wilk W	0.9623	0.9647	0.9801	0.9802	0.9684	0.9852	0.9887	0.9871	0.9882	0.9851	0.9862
p (normal)	0.0034	0.0051	0.0987	0.1005	0.0103	0.2669	0.4906	0.3742	0.4487	0.2631	0.3215
Anderson–Darling A	0.9838	0.8787	0.6150	0.6830	1.1520	0.4221	0.2770	0.4061	0.3069	0.4339	0.3547
p (normal)	0.0130	0.0237	0.1070	0.0725	0.0050	0.3164	0.6479	0.3454	0.5582	0.2966	0.4554
p (Monte Carlo)	0.0119	0.0252	0.1077	0.0744	0.0038	0.3220	0.6777	0.3485	0.5767	0.2895	0.4567
Lilliefors L	0.0785	0.0671	0.0730	0.0794	0.0913	0.0829	0.0473	0.0582	0.0531	0.0669	0.0559
p (normal)	0.0921	0.2515	0.1536	0.0851	0.0242	0.0597	0.7870	0.4691	0.6183	0.2567	0.5353
p (Monte Carlo)	0.0935	0.2599	0.1585	0.0861	0.0254	0.0642	0.7857	0.4899	0.6167	0.2597	0.5386
Jarque–Bera JB	9.9810	7.7950	1.9040	3.5670	4.9290	2.4170	1.9600	2.0440	2.0120	2.2580	2.4080
p (normal)	0.0068	0.0203	0.3859	0.1681	0.0851	0.2986	0.3752	0.3598	0.3657	0.3234	0.3000
p (Monte Carlo)	0.0153	0.0283	0.3118	0.1047	0.0652	0.2125	0.2931	0.2725	0.2840	0.2418	0.2143

illustrates the results of the four inferential statistics testing for departures from a normal distribution, considering the statistical distribution of the ESAI scores across Italian provinces (n = 110). Taken together, the four tests document a significant departure from normality for the years 1960 and 1970. No evidence of significant departures from a normal distribution was observed in 1980 and 1990. Some slight evidence was, instead, observed again in 2000. The statistical distribution of the ESAI scores in the following decades was presumably normal, confirming the progressive shift toward a more symmetrical distribution as revealed in the shape metrics (see above).

Table 4. Analysis of the statistical distribution of the ESAI score across Italian provinces (NUTS-3 level, n = 110), considering ecological diversity metrics by year (1960–2020) or 2030 scenario (S1 to S4).

Metric	1960	1970	1980	1990	2000	2010	2020	S1	S2	S3	S4
Evenness	1.4440	1.4350	1.4360	1.4390	1.4370	1.4340	1.4350	1.4340	1.4300	1.4350	1.4310
Brillouin	2.6380	2.5840	2.5900	2.6060	2.5940	2.5750	2.5820	2.5750	2.5500	2.5790	2.5540
Menhinick	9.0430	8.9690	8.9770	8.9990	8.9830	8.9570	8.9660	8.9560	8.9210	8.9620	8.9270
Fisher_alpha	194.20	184.40	185.40	188.30	186.20	182.90	184.10	182.90	178.60	183.70	179.30
Berger-Parker	0.0068	0.0066	0.0067	0.0067	0.0067	0.0066	0.0066	0.0066	0.0066	0.0066	0.0066

illustrates the main values of the selected (ecological diversity) indexes over time. A common trend was observed here, delineating a progressive reduction in the overall diversification of the statistical distribution of the ESAI scores. All the selected indexes provided the same (convergent) evidence, suggesting that the level of soil degradation of Italian land homogenized over time, becoming more symmetric, and thus indicating a coherent increase in values both at the two tails (the lowest and the highest levels of soil degradation) and in the central tendency of the distribution (corresponding broadly with the average ESAI).

3.3. A Multivariate Analysis of Distributional Metrics in a Composite Index of Soil Degradation

Figure 5 provides a graphical representation of a Principal Component Analysis (PCA) run on the input data matrix of distribution metrics and years. The corresponding biplot illustrates the distribution of columns (years) and rows (metrics) together, explaining cumulatively 68% of the total variance along the first two axes. A scrutiny of the graph made evident the latent trend over time in the ESAI scores already observed in earlier analyses (see Section 3.1 and Section 3.2). Axis 1, explaining half of the total variance, discriminated the two following distinctive patterns in the spatial distribution of the ESAI score: the former associated with years from 1960 to 1990–2000, and the latter more clearly associated with 2010, 2020, and the forecasts (S1–S4) for 2030. The biplot structure documents how the most recent years were associated with high (i.e., insignificant) p-levels from normality tests, thus confirming the shift toward a normal distribution of the ESAI scores. Scenarios for 2030, and especially S2 and S4, were more evidently associated with higher median values and the 75th percentile, thus indicating a substantial increase in medium-high scores of soil degradation; the 75th percentile, in fact, corresponded with an ESAI score of around 1.42–1.43, indicating a land that is categorized as fully ‘critical’ based on the standard ESAI classification system.

Conversely, diversity metrics were clearly associated with the third quadrant of the biplot (negative side of Axis 1 and Axis 2), together with kurtosis and asymmetry, being in turn linked with the spatial distribution of ESAI characteristically observed in 1960 and, more marginally, in 1990, when the average ESAI score was significantly lower. These findings, taken together, indicate that lower average ESAI scores, i.e., less intense predisposing factors to soil degradation, were associated with the more asymmetrical and diversified statistical distribution of the ESAI scores. Conversely, with increasing ESAI scores (i.e., a high level of soil degradation), the statistical distribution of the scores themselves became more regular, normal, and thus symmetric. Regularity, symmetry, and normality are thus characteristic elements of the statistical distributions of the ESAI scores, indicating a medium-high (and possibly increasing over time) level of soil degradation. In agreement with such evidence, the most representative dispersion metrics were associated with the fourth quadrant (negative side of Axis 1, positive side of Axis 2), being in turn linked with the 1970, 1980, and 2000 observation years. These intermediate periods featured a characteristic regime of soil degradation, being primarily associated with intense dispersion around the mean, documenting a particularly high unpredictability in the spatial distribution of the ESAI scores.

Figure 6 delineates the results of a two-way clustering that classified years and metrics separately. More specifically, years were classified in two main categories, basically more recent years (from 2010 to 2030) and less recent years (from 1960 to 2000). Metrics were in turn clustered based on their intrinsic similarity; normality tests were associated with the median, 75th percentile, and geometric mean of the ESAI distribution. Dispersion indexes formed a separate (smaller) cluster. Diversity metrics formed a third, autonomous group, and symmetry indicators clustered all together with the minimum ESAI score. These results clearly resemble those already observed from the Principal Component Analysis, suggesting a substantial stability, considering both the correlation and similarity criteria of the statistical amalgamation.

4. Discussion

Using descriptive (and inferential) statistical tools as well as simplified metrics derived from the ecological diversity analysis, the interpretative framework introduced here comparatively analyses the distribution of the ESAI, a well-known proxy of the level of soil degradation (widely used in the permanent monitoring of early desertification risk all over Europe and the Mediterranean basin, including Italy). The empirical analysis was carried out on a decadal scale from 1960 to 2020 at the province (NUTS-3 sensu Eurostat) level and benefited from a short-term forecast for 2030 based on four simplified assumptions grounded on a purely deterministic approach [70].

Early desertification risk requires the permanent monitoring of its implications in the stability of ecosystem balance and quality of life of the local communities [71]. Strategies mitigating such a phenomenon recommend the urgent development of calibrated medium- and long-term intervention strategies, finely tuned with the intrinsic (local and global) transformations in socioeconomic dynamics and territorial contexts [72,73,74]. Landscapes exposed to incipient soil degradation are relatively widespread in Southern Europe, as a specific annex (IV) of the United Nations Convention to Combat Desertification (UNCCD) dealing with the specific issues of the Mediterranean region has clearly recognized [75]. Here, soil degradation leading to early desertification processes proved to be locally significant in some parts of individual countries, irrespective of the state of the economy (e.g., advanced vs. emerging). Across countries and regions, it appears that desertification is characterized by specific forms and processes associated with climate aridity and drought, poor and degraded soils, sparse vegetation, and rugged topography, in turn associated with water (and wind) erosion risk [76,77,78].

In line with these complex perspectives, the empirical findings of the analytical techniques—spanning from descriptive statistics to multidimensional exploratory methodologies, and from parametric to non-parametric inference approaches—delineate worse predisposing conditions for soil degradation over time, as reflected in the sudden increase in the ESAI scores in Italy, observed in both average and maximum values [79]. This trend is particularly evident in all metrics of central tendency when scrutinizing the statistical distribution of the ESAI scores over time [63]. At the same time, evident changes over time were also recorded in other characteristics of the statistical distribution of the ESAI scores at the country scale, evidencing important variations for both dispersion and shape metrics [80]. These variations were also reflected in the different statistical behavior of the random process underlying the observed distribution of the ESAI score [49]. In principle, we assumed a normal distribution for the statistical distribution of the ESAI scores over the sample of Italian provinces investigated in our work [61].

Taken together, these trends may suggest the existence of latent environmental dynamics leading to early desertification risks, demonstrating how the interpretative and predictive framework proposed in this study (and based on a mixture of different statistical techniques) represents a valid tool, informative of soil degradation in both past and present conditions, as well as in future (short-term) scenarios [47]. An additional novelty of the study lies in the use of a ‘what … if’ scenario, giving some predictive information that can guide future decisions about soil degradation strategies in light with a zero-net land degradation strategy, as requested by UNCCD and Agenda 2030 Sustainable Development Goals (SDGs) [57].

The proposed scenarios—based on four different (basic) hypotheses dealing with the impact of (i) climate warming on soil aridity and of (ii) population dynamics on human pressure levels—are, in principle, very simplified, since they are grounded on linear, additive assumptions on the most relevant dimensions of change in the Mediterranean basin (e.g., climate, demography). However, to our knowledge, this is the first work providing a short-term forecast of the level of soil degradation in European countries [60]. A partial, or limited, availability of basic information layers, the complexity of long-term spatial trends (including both linear and non-linear behaviors), and the feedback interaction among factors of change reflected in elementary variables that is sometimes difficult to ascertain or estimate with considerable sampling errors, have often constrained some refined projection techniques associated with the precise quantification of occurrence probability or risk levels [46].

Despite its simplified and basically descriptive design, the interpretative framework introduced theoretically in this work, when applied to a representative case study such as Italy, documents the immediacy and appropriateness of information derived from both metrics, comparing the statistical distribution of the ESAI scores over time, and the intrinsic potential for an indirect evaluation of evolving risks, considering simultaneously departures from normality, the presence of outliers, spatial disparities, and increases in the mean and maximum values associated with the highest level of soil degradation [48]. A comprehensive scrutiny of past, present, and future trends in the ESAI scores using mixed statistical tools proved to be an original contribution to the study of soil degradation in advanced economies [44]. A particularly high rate of growth in the ESAI values, especially observed in more recent decades, is assumed to be significantly associated with both the evolving socioeconomic context and the increasing anthropogenic pressure [56]. These premises paved the way for general strategies and more specific measures addressing territorial peculiarities and delineate the specific trajectories of change reflected in the transforming distributional properties of the ESAI statistics [25,58,60,76].

From this perspective, the European Community Action programs (and, for instance, the Sixth Program entitled ‘Environment 2010: our future, our choice’) have identified the issues to be addressed as a priority by the European Union as (i) climate change, (ii) nature and biodiversity, (iii) environment and health, as well as (iv) sustainable use of natural resources [7]. To this end, the following actions have been identified: integrating the environmental dimension into social and economic policies, improving environmental legislation in the Member States, and encouraging environmental protection interventions [19]. The interpretative framework developed in the present study may provide a meaningful knowledge base informing all these issues (and the consequent policy strategies) from a regional/local perspective [54].

Additionally, the European Commission’s statistical program attributes an important role to territorial and ecological (official) statistics as tools supporting community environmental policies [6]. To achieve this objective, Eurostat has been increasingly asked to proceed with standardized procedures in making the available environmental statistics more comparable between countries and regions [15]. The increased demand for statistical information measuring environmental and socioeconomic sustainability conditions, as well as the considerable supply in response to requests at local, regional, and national levels, prefigure a significant growth in the sector, which will still be characterized for some years by an imbalance between the demand for environmental statistics and the statistics actually available [4]. As in the organizational context previously described, the present conditions—and the need for statistical improvements over time and space—may confirm the importance of the information framework proposed here as a basic tool assuring the permanent monitoring of early desertification preconditions—based on the advancement of soil degradation processes [1,2,3].

The mixture of different methodologies also contributes to a refined integration of different data sources, a typical issue that characterizes the permanent monitoring of soil degradation in Europe for a long time [5,8,18]. The integration of data sources of different origins allows us to obtain a considerable density of measurement occurrences (e.g., over both time and space) that may generate homogeneous estimations at the required (hopefully detailed) geographical scale [11,14]. This approach, carried out with the use of Geographical Information Systems tools, definitely allows for a disaggregated analysis of the interaction between biophysical variables and socioeconomic forces [47,52,65]. The latter aspect seems to be less explored in many agro-environmental studies [62]. Our framework may contribute to this research direction, with the aim of rebalancing this situation, also in the face of the strong linkages between climate and social (e.g., migrations) phenomena, as well as demographic (e.g., human pressure) and economic (e.g., impact of agriculture and tourism, water pollution, and industrial risk) processes [52].

Assuming the territory is a synergic and rapidly evolving system, where different productive, institutional, and contextual aspects act on environmental conditions, can significantly contribute to geographic information on environmental degradation indicating, through a multi-temporal approach, latent trends in soil depletion, since the predisposing factors act on generally different time scales [26]. For instance, human pressures generally act more rapidly than changes in climate, soil composition, vegetation, and landscape characteristics [42]. Providing spatial information on the speed of degradation processes and on the related human causes thus represents a scientific and cultural challenge but also an effective policy tool [43], which may more effectively direct mitigation interventions in the exposed territories and, above all, in those locations that—although demonstrated to be not affected today—may become potentially susceptible to the phenomenon in the short- and medium-term [73] because of climate warming, the economic crisis, and social change.

5. Conclusions

The significant growth over time in the ESAI scores for Italy confirms the appropriateness and informative potential of an interpretative framework of early desertification processes based on a mixture of quantitative metrics estimating changes over time in the statistical distribution of soil degradation scores. More specifically, changes over time in the spatial distribution of the ESAI were reflected in the central tendency, dispersion, and shape patterns, as well as in the latent behavior of the selected ecological metrics and the explicit degree of departure from a normal distribution. Due to the nature of the adopted variables and metrics, these evidence characteristics regarding both the anatomy (physical features) and the essence (conceptual elements) of the process under scrutiny, shedding further light on early desertification signs and ways to identify them on the field, based on permanent monitoring efforts.

In this sense, understanding the short-term evolution of soil degradation may require enhanced statistical procedures capable of delineating (even small) changes in distributional features that may reflect implicit risk indicators, as illustrated in the present study. Despite the existence of a large consensus on the appropriateness of the methodology adopted in this study to monitor land degradation vulnerability under Mediterranean ecological conditions, future research should investigate the possibility of integrating the ESAI framework into more flexible methodology, possibly grounding them on field surveys and making them better suited to capture local specificities in the real processes of soil degradation, moving from a vulnerability assessment to a continuous risk monitoring. Soil degradation processes were intrinsically demonstrated to adapt to (and interact with) differentiated territorial regimes, in turn demanding specific policies and (spatially explicit) planning measures. These actions have to effectively contain soil degradation by reorienting local development paths toward more sustainable (environmentally friendly, socially cohesive, and economically viable) paths.