How Do Local Economic Structures Influence the Variability of Land Sensitivity to Degradation in Italy?

Abstract

This study examines the relationship between local economic structures and environmental sensitivity in Italy, focusing on a novel indicator that estimates the spatial variability of the Environmentally Sensitive Area Index (ESAI) over time. This approach captures within-region disparities in degradation processes, addressing a key gap in the existing literature. Using a dataset covering all Italian provinces from 1960 to 2010 and considering multiple socio-economic variables, the research evaluates their impacts on ESAI variability. In particular, this study adopts a spatial autoregressive model (SAR), which allows both direct and indirect effects of selected predictors to be captured. The findings offer insights for policymakers in designing strategies to mitigate the spread of land degradation hotspots and promote strategies that balance environmental conservation with socio-economic development to ensure resource sustainability.

1. Introduction

Land degradation and early desertification are critical environmental [1,2] and socio-economic issues on a global scale [3], intensified by climate change and unsustainable economic growth [4]. These challenges threaten environmental stability and socio-economic development, particularly in rural areas that rely heavily on agriculture [4]. The consequences include job losses, population decline and economic recession [5,6]. To address the adverse effects of land degradation, continuous monitoring is essential for developing policies that support both adaptation and mitigation strategies in affected regions. Adaptation strategies aim to reduce vulnerability to the impacts of land degradation by adjusting human and natural systems to changing conditions. For instance, the promotion of modern irrigation systems in dry areas and the adaptation of grazing systems to prevent soil erosion can be considered adaptation measures. Mitigation strategies focus on reducing the drivers of land degradation and desertification. These include reforestation initiatives, soil conservation practices and improving irrigation efficiency to reduce water waste [7].

This requires a holistic approach that considers the wide range of factors influencing land sensitivity at different territorial scales. Effective policies should be tailored to the specific characteristics of each area, including its economic, environmental and social contexts that contribute to land vulnerability [8,9].

Despite global initiatives like the United Nations Convention to Combat Desertification (UNCCD) and the “Zero Net Land Degradation” (ZNLD) strategy outlined in the Sustainable Development Goals (SDGs) [3,10,11], progress toward the development of a standardized monitoring framework remains limited. The lack of a unified approach hinders the effectiveness of these initiatives [12], as inconsistent methodologies and indicators make it challenging to compare data across regions and track progress over time [13].

Several contributions have proposed quantitative approaches for monitoring land degradation [14], but these often fail to provide consistent and continuous support for local policy decision making. The literature highlights that land degradation is a multidimensional issue, making it difficult for any single measure to fully capture its complexity [15]. Therefore, a unified framework that provides consistent, high-quality and comparable indicators across different territorial levels is essential [8].

The Environmental Sensitivity Area Index (ESAI) is a recognized framework for assessing land sensitivity to degradation. It has been widely applied at local, regional and national scales in Mediterranean areas, based on four key dimensions of environmental quality: climatic conditions, soil properties, vegetation cover and human activities. By offering an overall assessment of vulnerability at the regional level, the ESAI helps identify hotspots, i.e., areas at the highest risk of desertification, and facilitates comparisons between different territories [6]. However, reliance on average vulnerability scores can limit the framework’s capacity to capture territorial spatial disparities. In heterogeneous contexts, areas of high degradation risk can coexist with areas of low risk, leading to significant variations in degradation levels even within the same administrative unit. This underscores the need for complementary measures beyond average assessments to capture both the intensity and variability of land sensitivity to degradation within each region [16].

To address the gap, this study introduces an innovative approach that builds on the ESAI framework while accounting for spatial variability of land degradation. Unlike traditional methods that focus exclusively on average vulnerability scores, this approach evaluates the variability of ESAI values, identifying zones where degraded and non-degraded lands coexist within the same region. By recognizing these spatial differences, policymakers can design more targeted strategies, focusing resources on severely degraded areas to mitigate damage and restore productivity while implementing preventive measures in less affected zones [17]. In regions with low variability in ESAI scores, where land sensitivity to degradation is consistently distributed, a cohesive, region-wide strategy can be applied uniformly. This simplifies resource management by promoting standardized actions to address uniform vulnerability levels. Regions comprising severely degraded, moderately degraded and non-degraded areas exhibit high spatial heterogeneity, reflected in higher variability in ESAI scores. In such cases, diverse strategies can be adopted according to a multi-target approach tailored to the real needs of areas [18]. For severely degraded areas, immediate interventions are crucial to stop further damage and restore land productivity [19]. Preventive strategies should be prioritized in non-degraded areas to mitigate future risks, focusing on factors like climate change, land-use changes and other contributors to degradation [20]. This dual strategy enables policymakers to concentrate on high-risk “hotspots” while implementing proactive measures to safeguard less affected areas, ensuring sustainable land management. Sustainable land management requires the adoption of land-use systems that maximize economic and social benefits while maintaining or improving ecological functions and sustaining the livelihoods of local communities [21]. By aligning economic activities with sustainable practices, local communities can balance development and environmental conservation, reducing land degradation and fostering ecological resilience. Therefore, an effective approach to sustainable land management cannot overlook the local economic structure of the area. This refers to the dominant economic activities and territorial vocations that characterize the region and influence its land-use patterns. These activities include primary sectors, such as agriculture, along with secondary and tertiary sectors like industrial production, services and tourism. The distribution and intensity of these economic activities determine the pressures exerted on natural resources and influence the land’s sensitivity to degradation.

Based on the above, this study aims to examine the influence of local economic structures on the spatial variability of land sensitivity to degradation over time. As mentioned above, the analysis adopts an innovative approach based on the ESAI framework. Unlike the majority of studies on this topic, which are based on average ESAI values, this study focuses on the spatial variability of ESAI scores across the Italian territory.

Italy serves as an ideal case study due to its pronounced economic disparities, as well as its diverse geography and climate conditions across regions [10,22]. The stark differences between the highly exposed southern areas and the northern regions, which have shown increased sensitivity due to climate impacts [17], provide a fitting context for this variability-oriented approach. These factors create a complex landscape where both natural and human-induced drivers of land vulnerability contribute to the heterogeneous spatial distribution of land sensitivity to degradation. The study focuses on NUTS-3 Italian provinces and six key time points (1960, 1970, 1980, 1990, 2000, 2010), which correspond to distinct phases in Italy’s economic cycle. After assessing the most appropriate estimation approach, spatial interactions between units are modeled using a spatially autoregressive model.

2. Conceptual Framework

Land degradation results from social, economic, demographic, political and biophysical factors operating across various spatial scales [23,24]. Both anthropogenic and natural factors are the leading causes of land degradation [25]. While natural causes are beyond human control, the challenges of anthropogenic influences can be addressed through targeted efforts [26]. This duality highlights the interconnectedness of human and environmental systems, where the pressures from human activities on natural resources often result in harmful ecological consequences.

In this regard, ecological economics explores how socio-economic development impacts environmental quality, providing a framework for investigating the link between economic activities and sustainability [27]. Ecological economics offers a critical perspective that contrasts with traditional economic frameworks by recognizing the significance of environmental costs and the finite nature of natural resources [28]. It underscores the need to align economic decision making with the planet’s ecological limits and long-term sustainability goals. In this context, land degradation frequently arises from economic processes that prioritize short-term productivity over ecological resilience, resulting in soil erosion, biodiversity loss and reduced land productivity [24].

Local economic structures can influence the vulnerability of land to degradation, as industrial and agricultural activities accelerate natural resources’ depletion. For instance, agricultural practices, though essential for food security and livelihoods, can exacerbate land degradation. Over-cultivation, deforestation and intensive irrigation contribute to soil depletion, habitat destruction and erosion [29]. Similarly, industrial activities often exploit natural resources at unsustainable rates with severe environmental impacts, including pollution, deforestation and depletion of water resources [30].

Demographic factors also contribute to land degradation [31]. High population densities intensify the demand for land resources, increasing pressures on housing, infrastructure and agriculture [23]. The relationship between population growth and resource degradation has been debated since the Malthusian school of thought [31], which posited that unchecked population growth could exceed land’s regenerative capacity. This imbalance leads to nutrient depletion and soil erosion, ultimately reducing agricultural productivity, accelerating fertility loss and driving desertification [32]. Beyond demographic pressures, the population’s age structure also affects land sensitivity to degradation. Regions with younger populations often face heightened pressure on land resources, increasing the risk of degradation as the working-age population drives resource demand [33]. Conversely, areas with an aging or shrinking workforce may experience land abandonment or mismanagement, resulting in desertification or diminished productivity [33].

Finally, the role of biophysical factors, including geographic and climatic conditions, cannot be overlooked [34]. Altitudinal gradients, soil quality and water availability significantly shape land vulnerability to degradation and often interact with human activities, worsening their impacts [25,34]. For example, lowland areas are particularly susceptible to desertification due to intensive agriculture, which, when mismanaged, leads to soil depletion and reduced fertility. Additionally, their proximity to water bodies makes them prime urban expansion and industrial development targets. On the other hand, regions with higher altitudes and steep slopes may contend with soil erosion and vegetation loss. These areas are also vulnerable to deforestation, which reduces the soil’s ability to retain moisture and increases the risk of landslides. Figure 1 visually represents the discussed conceptual framework. It emphasizes the need to consider both natural features and anthropogenic pressures when assessing land degradation.

Land degradation is rarely uniform across or within regions, reflecting disparities in economic development, demographics and environmental conditions. Less developed areas often lack the resources to implement sustainable land management, leaving them more exposed to degradation. Conversely, wealthier regions may possess greater means to invest in mitigation and restoration but still face significant challenges from urbanization and industrial expansion.

Building on this conceptual framework and the relationships outlined in the existing literature, the following sections detail the analysis and selection of key factors that help explain the variability of land degradation in the study area.

3. Materials and Methods

3.1. Methodology

To achieve the research objective of examining how local economic structures influence the spatial variability of land susceptibility to degradation, this study adopts a spatial approach. This methodology is well suited, as land degradation is rarely confined to a single area; instead, it often spreads across neighboring areas due to shared environmental conditions, land-use practices and human activities [17]. When one local unit experiences severe land degradation, the surrounding areas with similar characteristics are likely to experience similar degradation processes. The spatial dependencies underscore the interconnected nature of regions, where environmental degradation in one location can amplify or accelerate similar patterns in adjacent areas. Ignoring the spatial interactions can lead to inaccurate assessments and biased estimates, as the effects of explanatory variables may be either overstated or understated due to unaccounted spillover effects [35].

Based on this premise, this study adopts a spatial autoregressive model (SAR), which is particularly suited for modeling endogenous spatial interactions by introducing a spatial lag of the dependent variable [36].

The choice to perform the SAR model was supported by the results of the LM-lag test and its robust version [37]. Both tests indicate significant spatial autocorrelation in the variability of land sensitivity to degradation across all six study periods. Additionally, the LM-error test was conducted to assess the presence of autocorrelation in the errors. However, this test did not yield statistically significant results, suggesting that the disturbances lack a spatial autocorrelation structure. In addition, to assess whether exogenous interaction effects should be included, we conducted the likelihood ratio (LR) test, which compares the SAR model—which only includes the spatial lag of the dependent variable—with the spatial Durbin model (SDM), which also includes spatial lags of the explanatory variables. The results indicate that the null hypothesis of no exogenous interaction effects cannot be rejected, implying that extending the SAR model to the SDM is unnecessary. Figure 2 summarizes the methodological flow that guided the model selection, supporting the choice of the SAR model as the best-fit spatial specification.

The SAR model is formally expressed as follows

$$Y=\rho WY+\alpha {\iota }_N+X\beta +\epsilon$$

(1)

where $Y$ is the $N\times 1$ vector of the dependent variable, and $X$ is the $N\times K$ matrix of exogenous covariates with which the $K\times 1$ vector $\beta$ of regression parameters is associated. W is a non-negative and non-stochastic $N\times N$ spatial weights matrix that summarizes the spatial relationships between units. $WY$ is the $N\times 1$ spatial lag vector of the dependent variable, which is associated with the scalar parameter ρ, reflecting the strength of spatial dependence. Finally, $\epsilon$ is the $N\times 1$ vector of independently and identically distributed error terms with zero mean and constant variance.

The presence of the spatial lag vector $WY$ allows for the decomposition of the total effect caused by a unit change in the explanatory variables on the dependent variable into direct and spillover effects [38]. Direct effects focus on the localized impact of changes in a covariate within a specific unit, while spillover effects capture the interregional influence that a change in one area can have on surrounding regions.

For each k-th explanatory variable, the average of the diagonal elements of the partial derivatives’ matrix of the expected value of $Y$ corresponds to the average direct effect, while the average of the row (or column) sums of the off-diagonal elements represents the average spillover effect [35]:

$$\left[{\displaystyle \frac{\partial E\left(Y\right)}{\partial x_{1k}}}\cdots {\displaystyle \frac{\partial E\left(Y\right)}{\partial x_{Nk}}}\right]={\left(I_N-\rho W\right)}^{-1}{\beta }_k$$

(2)

3.2. Data

The assessment of spatial variability in land sensitivity to degradation across Italian provinces is based on the coefficient of variation of the ESAI. The analysis is conducted at six distinct time points: 1960, 1970, 1980, 1990, 2000 and 2010. These years align with the availability of census data derived from general censuses conducted every ten years by the Italian National Institute of Statistics (ISTAT). Analyzing data at the provincial level offers a balanced perspective that combines geographic detail and administrative relevance, which is essential for designing and implementing effective land degradation mitigation and adaptation policies [17].

During the study period, the number of provinces changed from 94 to 110 due to administrative reorganizations aimed at improving the efficiency of government and public services and adapting to socio-economic changes. Significant restructuring began in the 1990s, particularly in regions with rising population density and economic growth. New provinces were created, especially in northern and central Italy, to address the demands of urbanization and the need for more decentralized administration. As a result, it was not possible to perform a balanced panel data analysis. However, given Italy’s rapid socio-economic transformations following World War II [17], the periods analyzed in this study represent distinct and independent socio-economic phases, which justifies treating each time point separately. These selected timeframes reflect critical phases in Italy’s socio-economic evolution: from post-war reconstruction and industrial growth in the 1960s, through the population and economic surge of the 1970s, the environmental and social transformations of the 1980s and 1990s, to the globalized economy of the early 2000s, culminating in the financial instability and socio-institutional challenges of 2010 [39].

The ESAI is a composite index that combines 14 variables, representing four essential dimensions of environmental quality: climate, soil, vegetation and land management. Assessing variability in desertification through the coefficient of variation of the ESAI offers several advantages. First, it accounts for a wide array of factors contributing to land degradation and early desertification processes [40]. Second, it allows for the identification of regions vulnerable not only due to environmental conditions but also because of land management practices.

Figure 3 illustrates the spatial distribution of the coefficient of variation of the ESAI by deciles across Italian provinces over the six years considered. Darker shades indicate higher variability, suggesting greater differences in environmental sensitivity within a given province, while lighter shades reflect more stable environmental conditions, potentially resulting from more consistent land-use management.

Explanatory variables were sourced from official databases, including the Agricultural Census and National Accounting Statistics provided by ISTAT (

Table 1. Explanatory variables.

Label	Definition	Source
ALTI	Median altitude of a province above sea level (m)	ISTAT
CVAL	Coefficient of variation of ALTI	ISTAT
DENS	Population density (inhabitants/km2)	ISTAT
DIFF	Range of provincial population density (difference between the maximum and minimum density level)	ISTAT
PRSI	Labor productivity in the service sector to labor productivity in the industrial sector	National accounting statistics
PROL	Indicator of land productivity, defined as gross saleable production per utilized agricultural area	Agricultural Accounting Information Network
VAIN	Industrial value added to total provincial value added	National accounting statistics
VAPR	Percentage above or below the average per-capita provincial value added	National accounting statistics
AGRP	Total agricultural area to the provincial surface area	Agricultural Census
AGRF	Total agricultural area to farms (ha)	Agricultural Census
ELDE	Population aged 65+ to population aged 0–14	ISTAT
TOUR	Total number of beds in tourism facilities per total provincial surface area (ha)	ISTAT
CRED	Proportion of credit sector employees in the total population	ISTAT

). The integration resulted in a unique dataset that combines information on territorial characteristics with demographic and socio-economic structures at a high level of geographic detail. Their selection was informed by relevant literature (see Section 2), which identifies land degradation as the outcome of complex socio-economic interactions [6].

4. Results

The SAR models were estimated for each of the six time points using a spatial weights matrix (W) defined by a distance-decay function. Specifically, the elements of W were set equal to the inverse of the pairwise Euclidean distances between the centroids of provinces i and j $(w_{ij}=d_{ij}^{-1})$, provided that the distance fell within a predefined threshold that ensured each province had at least one neighbor. If the distance exceeded this threshold, the weight was set to zero, assuming no spatial interaction between the provinces.

The estimation results of the SAR models are presented in

Table 2. Estimation results of the SAR model with inverse distance weights matrix.

	1960	1970	1980	1990	2000	2010
ALTI	−0.125 (0.118)	−0.170 (0.116)	−0.101 (0.118)	−0.152 (0.123)	−0.165 (0.111)	−0.020 (0.102)
CVAL	0.304 *** (0.110)	0.332 *** (0.117)	0.188 (0.123)	0.069 (0.123)	0.097 (0.101)	0.183 ** (0.088)
DENS	−0.197 (0.138)	−0.202 (0.158)	0.091 (0.214)	−0.053 (0.186)	−0.122 (0.127)	0.159 (0.114)
DIFF	−0.223 ** (0.090)	−0.237 *** (0.089)	−0.173 * (0.092)	−0.206 ** (0.096)	−0.214 ** (0.086)	−0.264 *** (0.075)
PRSI	0.003 (0.090)	0.074 (0.090)	−0.101 (0.096)	−0.161 (0.105)	−0.232∗ (0.121)	−0.147 (0.101)
PROL	−0.161 (0.136)	−0.199 (0.149)	−0.397 ** (0.202)	−0.291 * (0.168)	−0.109 (0.117)	−0.025 (0.093)
VAIN	−0.402 *** (0.137)	−0.026 (0.121)	0.323 ** (0.145)	0.184 (0.125)	0.180 (0.116)	0.131 (0.098)
VAPR	0.210 (0.200)	−0.277 (0.194)	−0.470 ** (0.219)	−0.103 (0.213)	0.186 (0.202)	0.008 (0.119)
AGRP	0.127 (0.111)	0.121 (0.125)	−0.007 (0.127)	−0.068 (0.128)	−0.279 *** (0.095)	−0.148 (0.093)
AGRF	−0.152 (0.095)	−0.195 * (0.103)	−0.195 * (0.113)	−0.280 ** (0.111)	−0.288 *** (0.098)	−0.058 (0.089)
ELDE	−0.256 ** (0.109)	−0.118 (0.110)	−0.051 (0.111)	−0.158 (0.124)	−0.150 (0.103)	−0.190 ** (0.091)
TOUR	−0.004 (0.105)	0.086 (0.103)	0.189 * (0.101)	0.163 (0.119)	−0.129 (0.088)	0.033 (0.081)
CRED	0.250 * (0.133)	0.374 *** (0.131)	0.248 (0.152)	0.200 (0.170)	0.112 (0.162)	0.202 ** (0.103)
cost	−0.012 (0.078)	−0.013 (0.080)	0.002 (0.082)	0.006 (0.084)	0.004 (0.076)	0.001 (0.067)
N	92	94	94	95	103	110
ρ	0.288 **	0.290 **	0.424 ***	0.398 ***	0.452 ***	0.590 ***
LogLik	−105.09	−110.61	−113.46	−117.96	−122.57	−122.87
R2pseudo	0.419	0.377	0.345	0.291	0.361	0.448
AIC	242.17	253.21	258.91	267.92	277.14	277.73

* p < 0.1; ** p < 0.05; *** p < 0.01.

, while

Table 3. Effects’ decomposition of the SAR model with inverse distance weights matrix.

	1960	1970	1980	1990	2000	2010
	Direct effects
ALTI	−0.128	−0.175	−0.106	−0.159	−0.176 *	−0.023
CVAL	0.312 ***	0.340 ***	0.198	0.072	0.103	0.206 **
DENS	−0.202	−0.207	0.096	−0.056	−0.130	−0.179
DIFF	−0.229 **	−0.243 ***	−0.183 *	−0.216 **	−0.227 ***	−0.297 ***
PRSI	0.004	0.075	−0.107	−0.169 *	−0.247 *	−0.165
PROL	−0.165	−0.204	−0.420 *	−0.305 *	−0.116	−0.028
VAIN	−0.412 ***	−0.026	0.341 **	0.193	0.192 *	0.147
VAPR	0.215	−0.284	−0.496 *	−0.108	0.197	0.009
AGRP	0.130	0.124	−0.007	−0.071	−0.297 ***	−0.166
AGRF	−0.156	−0.200 *	−0.206 *	−0.293 **	−0.306 ***	−0.066
ELDE	−0.263 **	−0.120	−0.054	−0.166	−0.160	−0.214 **
TOUR	−0.004	0.088	0.200	0.171	−0.137	0.038
CRED	0.256 *	0.383 ***	0.262	0.210	0.119	0.227 **
	Spillover effects
ALTI	−0.047	−0.066	−0.069	−0.093	−0.126	−0.026
CVAL	0.116	0.128 *	0.128	0.042	0.074	0.240 *
DENS	−0.075	−0.078	0.062	−0.033	−0.093	−0.209
DIFF	−0.085	−0.091 *	−0.118	−0.126	−0.163 *	−0.347 **
PRSI	0.001	0.028	−0.069	−0.099	−0.177	−0.193
PROL	−0.061	−0.077	−0.270	−0.178	−0.083	−0.032
VAIN	−0.153	−0.010	0.220	0.113	0.137	0.171
VAPR	0.080	−0.107	−0.320	−0.064	0.141	0.011
AGRP	0.048	0.047	−0.005	−0.042	−0.213 *	−0.194
AGRF	−0.058	−0.075	−0.133	−0.172	−0.219 *	−0.077
ELDE	−0.098	−0.045	−0.035	−0.097	−0.114	−0.250
TOUR	−0.001	0.033	0.129	0.100	−0.098	0.044
CRED	0.095	0.144 *	0.169	0.123	0.085	0.265

* p < 0.1; ** p < 0.05; *** p < 0.01.

shows the decomposition of total effects into direct and spillover effects. As a robustness check, the estimates were repeated using a different configuration of the spatial weights matrix.

Table A1. Estimation results of the SAR model with 5-nearest neighbor weights matrix.

	1960	1970	1980	1990	2000	2010
ALTI	−0.096 (0.120)	−0.152 (0.117)	−0.156 (0.113)	−0.194 (0.118)	−0.080 (0.121)	0.100 (0.122)
CVAL	0.330 *** (0.111)	0.368 *** (0.117)	0.207 * (0.117)	0.107 (0.119)	0.131 (0.109)	0.273 *** (0.105)
DENS	−0.190 (0.139)	−0.179 (0.160)	0.072 (0.204)	−0.043 (0.179)	−0.059 (0.139)	−0.106 (0.137)
DIFF	−0.210 ** (0.090)	−0.227 ** (0.090)	−0.156 * (0.088)	−0.192 ** (0.092)	−0.208 ** (0.094)	−0.312 *** (0.090)
PRSI	0.003 (0.091)	0.090 (0.091)	−0.088 (0.091)	−0.146 (0.102)	−0.277 ** (0.132)	−0.146 (0.122)
PROL	−0.183 (0.137)	−0.230 (0.150)	−0.400 ** (0.192)	−0.304 * (0.161)	−0.158 (0.128)	−0.075 (0.113)
VAIN	−0.417 *** (0.139)	−0.044 (0.122)	0.302 ** (0.138)	0.161 (0.120)	0.203 (0.127)	0.153 (0.118)
VAPR	0.226 (0.202)	−0.273 (0.195)	−0.445 ** (0.209)	−0.118 (0.205)	0.037 (0.221)	−0.025 (0.144)
AGRP	0.127 (0.112)	0.096 (0.126)	−0.042 (0.122)	−0.106 (0.123)	−0.259 ** (0.103)	−0.123 (0.112)
AGRF	−0.213 ** (0.097)	−0.238 ** (0.104)	−0.208 * (0.108)	−0.272 ** (0.107)	−0.333 *** (0.106)	−0.066 (0.107)
ELDE	−0.245 ** (0.110)	−0.083 (0.111)	−0.046 (0.106)	−0.125 (0.120)	−0.067 (0.113)	−0.167 * (0.090)
TOUR	−0.017 (0.106)	0.066 (0.104)	0.175 * (0.096)	0.139 (0.115)	−0.035 (0.095)	0.154 (0.098)
CRED	0.271 ** (0.134)	0.378 *** (0.132)	0.233 (0.145)	0.181 (0.163)	0.162 (0.177)	0.204 * (0.124)
cost	−0.011 (0.079)	−0.006 (0.081)	−0.008 (0.078)	−0.003 (0.081)	0.001 (0.083)	−0.005 (0.081)
N	92	94	94	95	103	110
ρ	0.298 **	0.295 **	0.529 ***	0.497 ***	0.117	0.275 **
LogLik	−105.66	−111.18	−110.24	−118.20	−128.75	−138.42
R2pseudo	0.411	0.370	0.388	0.283	0.279	0.268
AIC	243.32	254.35	252.48	262.40	289.50	308.84

* p < 0.1; ** p < 0.05; *** p < 0.01.

and

Table A2. Effects’ decomposition of the SAR model with 5-nearest neighbor weights matrix.

	1960	1970	1980	1990	2000	2010
	Direct effects
ALTI	−0.098	−0.155	−0.170	−0.209	−0.081	0.101
CVAL	0.338 ***	0.375 ***	0.225	0.115	0.128	0.276 **
DENS	−0.194	−0.183	0.079	−0.047	−0.060	−0.107
DIFF	−0.214 **	−0.232 **	−0.170 *	−0.206 **	−0.214 **	−0.316 ***
PRSI	0.003	0.092	−0.095	−0.156	−0.277 **	−0.149
PROL	−0.187	−0.235	−0.435 **	−0.327 **	−0.155	−0.076
VAIN	−0.427 ***	−0.045	0.328 **	0.173 **	0.200	0.156
VAPR	0.231	−0.279	−0.484 **	−0.127	0.032	−0.025
AGRP	0.130	0.098	−0.045	−0.114	−0.256 **	−0.124
AGRF	−0.218 **	−0.243 **	−0.227 **	−0.293 **	−0.338 ***	−0.067
ELDE	−0.251 **	−0.084	−0.050	−0.134	−0.071	−0.169 *
TOUR	−0.017	0.067	0.190 *	0.149	−0.034	0.156
CRED	0.276 *	0.386 ***	0.253 *	0.194	0.166	0.207 *
	Spillover effects
ALTI	−0.039	−0.060	−0.161	−0.178	−0.011	0.037
CVAL	0.133	0.146 *	0.214	0.098	0.017	0.100
DENS	−0.077	−0.071	0.075	−0.040	−0.008	−0.039
DIFF	−0.085	−0.090 *	−0.162	−0.176	−0.028	−0.115
PRSI	0.001	0.036	−0.091	−0.133	−0.036	−0.091
PROL	−0.074	−0.091	−0.414	−0.278	−0.020	−0.028
VAIN	−0.168	−0.018	0.312	0.147	0.032	0.093
VAPR	0.091	−0.108	−0.461	−0.108	0.004	−0.009
AGRP	0.051	0.038	−0.043	−0.097	−0.033	−0.045
AGRF	−0.086	−0.094	−0.216	−0.249	−0.044	−0.024
ELDE	−0.099	−0.033	−0.048	−0.114	−0.009	−0.047
TOUR	−0.007	0.026	0.181	0.127	−0.004	0.057
CRED	0.109	0.150	0.241	0.166	0.021	0.075

* p < 0.1; ** p < 0.05; *** p < 0.01.

in the Appendix present the results obtained with the k-nearest neighbors (knn) matrix. According to this weighting scheme, which ensures an equal number of neighbors for each spatial unit, the elements of W were set to 1 if the province j was among the knn of the province i, and 0 otherwise. In this study, k was set to 5, corresponding to each province’s median number of immediate neighbors under a contiguity-based criterion.

The spatial autoregressive coefficients are significant and consistently positive across all six estimated models, highlighting a direct and persistent spatial dependence in the variability of land sensitivity to degradation among Italian provinces. Moreover, the magnitude of the spatial autoregressive coefficients exhibits a general upward trend over the decades, reflecting an intensification of spatial influence over time.

The pseudo-R-squared values follow a U-shaped trajectory over the observed period, starting at 0.42 in 1960, decreasing to a low of 0.29 in 1990 and then rising to 0.45 in 2010. This trend suggests that the explanatory variables were more effective in capturing the spatial heterogeneity of land sensitivity to degradation in both the earlier and later years of the analysis.

The models for each decade show distinct patterns. In 1960, significant relationships are identified between the variability of the ESAI and the coefficient of variation of altitude (CVAL), the share of industrial added value (VAIN) and the aging index (ELDE). These findings highlight that the spatial heterogeneity of desertification vulnerability is strongly linked to topographical heterogeneity. In Italy, and more broadly in the Mediterranean region, desertification follows an altitudinal gradient. Lowland areas are particularly vulnerable due to arid climatic conditions and increased anthropogenic pressure, while mountainous regions exhibit lower vulnerability, characterized by milder climates and lower human impact [41]. The negative relationship with the share of industrial added value suggests that areas with less industrial development may be more susceptible to experiencing greater heterogeneity in desertification vulnerability. Based on the negative relationship between desertification vulnerability and the aging index, provinces with younger populations tend to show greater variability in desertification risk.

The analysis for 1970 confirms the relationship between desertification vulnerability and the coefficient of variation in altimetry (CVAL) first observed in 1960, with significant direct and spillover effects. Therefore, provincial CVAL influences ESAI variability both on site, within the same province, and off site, in neighboring areas through spillover effects. Additionally, greater disparities in population density (DIFF) led to a decrease in the heterogeneity of desertification vulnerability. Although with varying significance levels throughout the study period, its direct effects have been consistent from 1960 to 2010. Moreover, significant spillover effects emerged in 1970, 2000 and 2010, indicating the broader spatial implications of settlement dynamics for land sensitivity to degradation for those years. Similarly, the negative relationship between agricultural area per farm (AGRF) and heterogeneity of land vulnerability to degradation remains consistent throughout the study period.

The positive relationship between the share of employees in the credit sector (CRED) and the spatial heterogeneity of the ESAI suggests that areas with a stronger presence of financial services experience heterogeneous vulnerability of land to degradation.

During the 1980s, the factors driving spatial heterogeneity in land vulnerability to desertification changed, reflecting broader socio-economic transformations in Italy [6]. The influence of biophysical factors, such as altitude variability (CVAL), disappeared, while socio-economic factors gained prominence. The land productivity index (PROL) showed a negative relationship with the variability of vulnerability to desertification, indicating that higher land productivity was associated with environmentally healthy territories that had lower and more homogeneous desertification levels. Meanwhile, the significant influence of industrial value-added (VAIN) and economic disparities (VAPR) across provinces—positive and negative, respectively—highlights the increasing role of economic factors in explaining the heterogeneity of desertification risk.

The model shows its weakest performance in 1990, as indicated by the lowest pseudo-R-squared value across the study period. The results for population density range (DIFF), land productivity (PROL) and total agricultural area (AGRF) on farms in 1990 were consistent with the trends observed in the previous decade.

In 2000, the altimetric gradient once again influences the variability of desertification vulnerability with significant direct effects, although it is no longer linked to altimetric heterogeneity but rather to its median provincial values (ALTI). The effects of settlement dynamics and population density (DIFF) are significant, confirming to be key factors in shaping the heterogeneity of desertification risk [31,32]. The provincial agricultural area (AGRP) and agricultural area on farms (AGRF), which reflect the average farm size of agricultural enterprises, negatively impact ESAI variability. The relationship between agricultural activity and land sensitivity to degradation is strengthened by the significant negative spillover effects of AGRP and AGRF, which align with their on-site impacts observed within the province itself.

In 2010, some relationships observed in 1960 re-emerge, both for biophysical factors, such as the altimetric gradient (CVAL), and socio-economic indicators, such as the population density range (DIFF), which show both on-site and off-site effects. Additionally, the aging index (ELDE) and the share of employees in the credit sector (CRED) are confirmed to be key drivers of ESAI variability within the reference province. Although the average ESAI significantly changed from 1960 to 2010, as discussed in the literature [17], its variability, particularly the factors influencing it, followed a cyclical pattern, with no substantial changes observed between 1960 and 2010.

5. Discussion

The analysis of land sensitivity to degradation through ESAI variability shows patterns typical of complex adaptive systems. Between 1960 and 2010, the system followed a cyclical trajectory, eventually returning to a configuration similar to its initial state after multiple intermediate phases. The cyclical pattern is evident in the persistence of spatial heterogeneity and the stability of key drivers. Similar trends are reflected in the model’s goodness-of-fit measures, which show comparable values in both the early and late years of the analysis period. Regarding spatial dependencies, the magnitude of the spatial autoregressive coefficients shows a general upward trend over the decades, indicating an intensification of spatial influence over time. This trend suggests that land degradation processes are increasingly interconnected, with areas exhibiting similar characteristics more likely to cluster geographically [42].

Each decade reflects distinct phases of the Italian economic cycle, providing context for the observed relationships. In the 1960s, negative correlations are found between the variability of desertification vulnerability, the share of industrial added value and the aging index. The first can be attributed to the industrial landscape of the time. The industry was largely concentrated in the northern regions of Italy, where spatial heterogeneity was lower, and the levels of degradation and desertification were less severe. In contrast, the southern regions, where industrialization was less developed, showed higher levels of desertification vulnerability and greater spatial heterogeneity. The relationship with the aging index mirrors how younger populations are often associated with more dynamic economic and social contexts characterized by increased anthropogenic pressures. As a result, these areas become more vulnerable to land-use changes, which in turn intensify both the overall land vulnerability to degradation and its spatial heterogeneity.

Throughout the study period, increasing disparities in population density are associated with less variability in desertification vulnerability [43]: areas tend to polarize, with some highly degraded regions and others almost free of risk. This pattern aligns with monocentric urban systems, where a central city with high anthropogenic pressure is surrounded by low-density areas with lower human pressure [6]. In these contexts, the overall level of desertification tends to be relatively homogeneous. On the other hand, provinces with more evenly distributed population density show greater heterogeneity in desertification risk. In these areas, anthropogenic pressure is more widespread across an entirely urbanized territory, with a higher level of desertification and, consequently, greater variability in desertification risk between provinces [43].

Similarly, another significant and consistent relationship, except for the first and last years of the analysis, is observed between the response variable and the provincial agricultural area of farms, which expresses the average farm size. In Italy, large-scale, extensive farms, often characterized by latifundia systems, typically have a severe environmental impact. Latifundia systems can lead to soil depletion and environmental damage due to the overuse of chemical inputs and the lack of crop rotation as production expands, placing excessive stress on the land and increasing the risk of desertification [44]. This degradation tends to be evenly distributed throughout the area, as large-scale farms generally manage extensive tracts of land with similar practices, leading to uniform environmental pressures. In contrast, areas with lower agricultural land per farm are often characterized by fragmented production systems. This fragmentation leads to more varied land-use and management practices, resulting in diverse levels of land degradation. Consequently, regions with smaller agricultural units tend to exhibit greater heterogeneity in desertification risk [45]. This spatial variability highlights the dual role of agriculture as a key economic sector that supports food production and security, provides raw materials for industry and energy while also contributing to land degradation if not sustainably managed.

As for the positive effect exerted by CRED, this is due to the urban configuration typical of areas where financial sectors are more developed. A high share of employees in the credit sector reflects mature, polycentric urban systems characterized by advanced development, wealth and a predominance of tertiary and quaternary sectors. These urban environments often exhibit a mix of areas with both high and low risk of desertification. The highlighted relationship also extends to neighboring areas, as evidenced by the significant spillover effects. This finding should be considered alongside the relationship with population density range, which shows the opposite effect. A higher population density range is typical of monocentric systems, where densely populated urban centers contrast with sparsely populated rural areas, leading to a more uniform desertification risk. In contrast, a high share of credit sector employment reflects polycentric and economically advanced urban systems where different levels of desertification risk coexist [46]. Thus, a decrease in population density disparities or an increase in credit sector employment signals the emergence of complex urban systems with greater spatial heterogeneity in desertification vulnerability [47].

The models estimated for the central years of the analysis, the 1980s and 1990s, show a reduction in explanatory power. This decline can be attributed to the socio-economic transformations of the period, including urbanization, industrial restructuring and shifts in land-use policies, all of which influenced land sensitivity to desertification. Notably, land-use policies began to focus on limiting urban sprawl, promoting sustainable agriculture and protecting areas of environmental interest. A key example is the Legge Galasso (1985), which introduced strict regulations to protect landscapes and natural areas, imposing constraints on urban expansion and preserving large portions of agricultural land to prevent their conversion into built-up areas. In 1990, desertification risk was geographically polarized, with a marked contrast between areas facing extreme degradation risk and those with minimal risk. This polarization peaked in 1990, following significant deterioration in physical and geophysical conditions driven by soil erosion, compaction and the loss of organic matter, along with changes in hydrological and geomorphological processes.

Interesting relationships emerge in the 2000s between variability in land sensitivity, the share of agricultural area at the provincial level and the average farm size. Provinces with a strong agricultural vocation and larger agricultural businesses exhibit greater homogeneity in sensitivity to land degradation. Large farms, typically associated with extensive farming practices and uniform land management, reduce the variability of desertification risk, thus fostering a more homogeneous environment in terms of vulnerability.

Overall, while the average ESAI showed significant changes throughout the study period, its variability and the factors influencing it followed a cyclical pattern, with no noteworthy changes observed in 2010 compared to 1960.

Based on the above discussion of the results, several considerations can be drawn to inform public policies to combat desertification. To effectively address this issue, the strategies must go beyond merely tackling the average level of degradation; they should also focus on managing its distribution and variability. While generalized interventions may be sufficient for managing overall degradation, targeted policies are required to reduce heterogeneity, specifically tailored to both the most vulnerable and the least vulnerable areas. In essence, effective desertification mitigation should prioritize restoring the most degraded regions while safeguarding the least degraded ones, which represent the poles of spatial variability in ESAI.

This approach may be particularly critical at the provincial level, where ESAI heterogeneity is more pronounced than at larger scales. The success of these efforts depends on balancing local and national policies while aligning them with long-term strategic goals, as land degradation is a gradual and persistent process.

6. Conclusions

This study provides a novel perspective on land sensitivity to degradation by focusing on the spatial variability of desertification vulnerability, measured by the coefficient of variation of the ESAI, across Italian provinces over five decades. This innovative approach captures within-region disparities in degradation processes, addressing a key gap in the existing literature. The study advances theoretical understanding of the interactions between local economic structures and environmental vulnerability by integrating socio-economic and biophysical variables within a spatial econometric framework.

The growing spatial dependence over time highlighted the importance of policy interventions that considered local effects rather than only addressing land degradation at the national level. The results showed contrasting profiles of economically advanced and urbanized provinces compared to predominantly agricultural areas. Large urban centers, characterized by a high concentration of financial and industrial activity, such as those with a significant share of employees in the credit sector, are characterized by higher spatial heterogeneity in desertification vulnerability. This increased variability can be attributed to the different land-use patterns and human pressures in these urban systems. In contrast, rural areas with extensive agricultural enterprises show less variability, with larger agricultural areas associated with more homogeneous desertification risks.

The study also identified a cyclical pattern in the variability of desertification risk from 1960 to 2010. While the average ESAI showed significant changes over time, the underlying variability remained relatively stable. Therefore, despite the evolving socio-economic context and the impacts of land degradation, the drivers of ESAI variability have not undergone substantial shifts. This research provided an original assessment of land sensitivity to degradation by introducing alternative desertification indicators for decision support systems. It also highlighted the need for integrated, multi-target and multisectoral policies tailored to the socio-economic characteristics of different territorial contexts. While much of the research on desertification focused on environmental factors, this study emphasized the importance of considering socio-economic dimensions, which have received less attention in recent decades within the European context.

In this paper, we specifically focused on spatial interactions and the direct and spillover effects that each predictor tested exerts on the variability of land vulnerability to desertification. Future research could explore the use of geographically weighted models to assess the heterogeneity of the relationships identified across different territories within the study area. These models could contribute to the literature by evaluating the extent to which local characteristics influence territories’ vulnerability and allow for tailored policy recommendations.