What Happens in the City When Long-Term Urban Expansion and (Un)Sustainable Fringe Development Occur: The Case Study of Rome

This study investigates long-term landscape transformations (1949-2016) in urban Rome, Central Italy, through a spatial distribution of seven metrics (core, islet, perforation, edge, loop, bridge, branch) derived from a Morphological Spatial Pattern Analysis (MSPA) analyzed separately for seven land-use classes (built-up areas, arable land, crop mosaic, vineyards, olive groves, forests, pastures). A Principal Component Analysis (PCA) has been finally adopted to characterize landscape structure at 1949 and 2016. Results of the MSPA demonstrate how both natural and agricultural land-uses have decreased following urban expansion. Moreover, the percent 'core' area of each class declined substantially, although with different intensity. These results clearly indicate ‘winners’ and ‘losers’ after long-term landscape transformations: urban settlements and forests belong to the former category, the remaining land-use classes (mostly agricultural) belong to the latter category. Descriptive statistics and multivariate exploratory techniques finally documented the intrinsic complexity characteristic of actual landscapes. The findings of this study also demonstrate how settlements have expanded chaotically over the study area, reflecting a progressive 'fractalization' and inhomogeneity of fringe landscapes, with negative implications for metropolitan sustainability at large. These transformations were unable to leverage processes of settlement and economic reagglomeration around sub-centers typical of polycentric development in the most advanced socio-


Introduction
Moving further away from inner cities, vastly different landscapes have been more intensively shaped by settlement expansion. Urban growth and the intimate interrelations with land-use represents a complex issue involving multiple planning and ecological dimensions [1][2][3]. Landscape transformations in metropolitan regions are often the result of interactions between (apparent and latent) socioeconomic factors of change [4][5][6]. The linkage between land-use changes and demographic dynamics in economically advanced countries is becoming increasingly multifaceted because of the mutual interplay of environmental and planning spheres that influence the sustainable development of regions and local communities [7][8][9]. In this sense, exurban development has been strongly influenced by land prices' variations, altering the equilibrium of forces, which has profoundly shaped the actual organization of economic activities within cities [1,[10][11][12]. The monocentric model, intended as a characteristic and simplified urban spatial structure, has for a long time represented the basis for economic analysis and evaluation [13][14][15][16]. According to the monocentric model, central cities hold a core area concentrating upper socioeconomic functions (e.g., population density, land value, house prices, dwelling characteristics, and capital-land ratio) reflected in a specific land-use profile [5,17,18]. As we move outward, these functions gradually decrease in intensity and/or spatial extent [19]. However, given the progressive and irregular expansion of cities with different sizes (from large to small extent), the monocentric model does not provide an accurate and realistic depiction of the land use structure [20][21][22][23]. Recently, the polycentric model has been considered more appropriate in capturing the current spatial structure of urban areas [24][25][26][27]. The third, even more frequent way of urban growth is now the exurban development (i.e., the characteristic pattern of urban expansion following low-density settlements scattered on a natural or agricultural matrix [1,28,29] resulting in urban morphologies very different from the traditional monocentric and polycentric structures [30][31][32]).
Exurban expansion usually tends to produce fragmented landscapes without distinctive features of the polycentric model (such as production poles or sub-centers [33]). Landuse changes in peri-urban areas have been the subject of a series of studies indicating also possible knock-on effects on environmental matrices [34][35][36][37][38][39][40][41]; on the contrary, the relationship between landscape changes and the three different ways of urban enlargement (monocentric, polycentric and dispersed, see Figure 1) has not yet been adequately studied, especially in areas where anthropogenic pressure has increased more rapidly in recent years [42][43][44]. Fortunately, land use databases are increasing in size and populate the web more frequently with respect to a recent past, involving all the geographic scales [45][46][47]. Therefore, the most suitable tools to validate working assumptions on the different urban development modalities and to verify the reliability of various indicators rely on land-use datasets [48]. In this context, exploratory analysis is a possible approach to shed light on the complex links between structures, functions and change dynamics of (rapidly expanding) urban settlements [49].
Our work introduces an original perspective summarizing today's knowledge and scientific directions in the field of long-term urban expansion in light of different models of urban growth. To our knowledge, we present here the first study documenting (and quantitatively analyzing) long-term (nearly 70 years) landscape transformations in a Mediterranean peri-urban landscape, bringing substantial implications for diachronic analysis of urban expansion in developed (but peripheral) countries in Europe. As a novel contribution to landscape studies, the present work focuses on the prevalent modes of urban expansion occurred in Rome over a relatively long-time span (1949-2016) by performing a comprehensive, spatially explicit analysis of landscape transformations based on a mathematical spatial morphology approach. In the last half century, the metropolitan area of Rome, a typical semi-compact and dense Mediterranean city, has undergone different, sometimes conflicting, waves of settlement expansion, representing a paradigmatic example of urban transformations in the whole southern Europe [42,[50][51][52]. The empirical results of this study allow discussing the (supposed) unsustainability of current urban expansion compared with the past settlement structures as far as land fragmentation and loss of relict habitats and traditional crops at the fringe are concerned. At the same time, our study presents polycentric urban growth as a candidate for a feasible and more sustainable means of metropolitan development and, possibly, urban containment, in compact Mediterranean cities.

Study Area
The investigated area covers a part of Rome's province including the local municipalities of Rome and Fiumicino, for a total surface area of 1,500 km 2 ( Figure 2). The lowlands (the so called 'Agro Romano', a cultivated area with traditional rural landmarks, biodiversity and cultural heritage) surrounded the inner city of Rome expanded through the alluvial plain of the Tiber river [53]. Although the increasing human pressure threatened the original forest vegetation, relict forests are preserved along the coastal rim. Industrial areas are located in the eastern part of the city while traditional cropland intermixed with pasture and shrub land still occur in the western part of the study area. Rome's climate is typically Mediterranean with rainfalls concentrated in autumn and spring and relatively mild temperatures in winter. The average long-term (1961-1990) annual rainfall and mean daily temperature in Rome were 700 mm and 16°C respectively. However, decreased precipitation rate and increasing average temperatures have been recorded in recent decades.

Land-Use Maps
Land-use data were obtained from the elaboration of two compatible spatial geo-databases with a digital land classification based on a simplified Corine Land Cover classification nomenclature [50][51][52][53][54]: (i) the Italian Istituto Geografico Militare topographic map (1:25,000 scale) produced in 1949, and (ii) a land-use map (1:25,000) produced by the Cartographical Service of the Latium regional authority derived from photo-interpretation of digital ortho-images released from the Italian National Geoportal (this map was originally produced in 1999-2000 and completely updated in 2016 with new images referring to the same year). They were processed and reprojected to the common WGS84 UTM zone 32 in GIS environment (QGIS 3.14.1, see http://qgis.osgeo.org). Seven homogeneous classes with a minimum mapping unit of 1 hectare were selected as follows: (i) arable land, (ii) mixed cropland, (iii) vineyards, (iv) olive groves, (v) forests, (vi) pastures, (vii) built-up areas, and (viii) wetlands and other (minor) use of land (e.g., beaches, dunes, rocks). The related figures were qualitatively checked for consistency with independent official data derived from statistical sources, e.g., agricultural and building censuses, rural surveys, land price maps, cadastral maps. These comparisons were preliminary to further analysis and were carried out with the purpose only of verifying that different data sources were coherent in addressing the same landscape transformations (i.e., expansion of urban settlements, decline of arable land, moderate increase of forestland) over such a time interval (1949-2016).

Landscape Analysis
Landscape composition (1949 and 2016) and annual per cent rate of change were been calculated by land-use class. Additionally, we used Guidos software [55] to classify landscapes based on Mathematical Spatial Pattern Analysis (MSPA), a technique investigating shape and form of objects [56][57][58]. MSPA is a customized sequence of mathematical morphological operators aimed at describing geometry and connectivity of the image components. As it is based on geometric principles, MSPA can cover different kinds of applications independently of scale and type of digital images in any application field (for further details, see https://forest.jrc.ec.europa.eu/en/activities/lpa/mspa/). This methodology implements image processing routines to identify hubs, corridors and other spatial elements relevant to landscape analysis [57]. Seven basic elements were identified here: (i) 'core', (ii) 'islet', (iii) 'bridge', (iv) 'loop', (v) 'branch', (vi) 'edge', and (vii) 'perforation' (Figure 3). 'Core' areas are the inner part beyond a certain distance to the boundary, 'islets' are those portions of land that are too small and isolated to contain a core area. Each core area is surrounded by 'edges' and 'perforations'. Perforated areas were defined as transition zones between 'cores' and a different land-use class, while 'edges' represent the transition between 'core' and 'non-core' areas within the same class. Loops, bridges and branches connect "core" areas. Loop areas represent corridors which connect to the same core, bridges connect two (or more) cores and branches connect a core area with a non-core area within the same land-use class. A Principal Component Analysis was run on a matrix composed of the relative proportion of the seven morphological classes (see above) by land-use category and year. The data matrix was constituted of seven columns (morphological classes) and seven land-use classes (rows) and contains the respective value (see above) by column and row as a proxy of landscape structure and composition together. To delineate similarities in the spatial configuration of the studied landscape, a biplot illustrating the statistical distribution of component loadings (morphological categories) and scores (land-use classes) was used. This plot contributed to representing the latent relationships in the composition and structure of the investigated landscape, providing a new perspective (and operational tools) in land-use analysis. Components were extracted on the base of the absolute eigenvalue; components with eigenvalue > 1 were considered significant and analyzed further.  Table 1 reports aggregated and individual class changes in landscape composition over the study period. While built-up areas expanded significantly from 6.6% in 1949 to 28.9% in 2016, the largest decline was observed for agricultural systems including pastures (from 81.5% to 58.3%). Forests increased moderately and other land use classes (including wetlands and water bodies) remained largely stable. Among agricultural systems, arable land experienced the largest negative change (Figure 4). Less intense changes were observed for vineyards (decreasing), olive groves, pastures and crop mosaics (slightly increasing).  Based on MSPA, trends over time in the per cent 'core' area were illustrated in Table  2. Built-up areas contained a relatively stable proportion of 'core' patches with a moderate increase from 34.9% in 1949 to 35.7% in 2016. Arable land, vineyards, pastures and crop mosaics experienced the most intense decline over time. The other classes had a less intense decline in 'core' parches; forests were the class with the lowest negative rate of change in 'core' patches. Specifically focusing on built-up areas, changes over time in the average patch size (ha) is shown in Table 3 by morphological class. While patches classified as 'cores' remain the largest for both years, their size declined significantly during the study period. The same pattern was observed for 'edge' patches. By contrast, 'bridges' increased significantly over time, likely suggesting a sort of landscape 'fractalization' in smaller patches with some intrinsic interconnections. Table 3. Average patch size (ha) of built-up areas by morphological class and year (difference was calculated as annual per cent change over time; statistics is the level of probability associated with a non-parametric Mann Whitney U-test checking for similarity in patch size over time; * indicates significant differences at p < 0.05 after Bonferroni's correction for multiple comparisons).  Figure 5 illustrates the empirical results of a Principal Component Analysis run on the bi-dimensional matrix linking land-use classes and morphological categories with the use of a representative target variable, the average patch size. For 1949, PCA extracted two main components explaining together 83.5% of landscape variability. Component 1 (65.1%) discriminated largely un-fragmented landscapes dominated by arable land (right side) from patchy agricultural systems mostly associated with olive groves (smaller patch size). Core, bridge, edge, loop were the classes most associated with Component 1. Component 2 (18.4%) discriminated pastures and forests from vineyards evidencing the intrinsic role of 'perforation' class. All in all, 1949 landscape structure was rather simplified and oriented towards intensity of land-use. The contribution of built-up areas in landscape variability was quite modest. The biplot (loadings vs. scores) clearly discriminated natural systems, extensive rural systems and intensive crops were discriminated.

Class
For 2016, PCA extracted two main components accounting together for 66.2% of landscape variability. Component 1 (39.1%) discriminated un-fragmented landscapes dominated by built-up areas and arable land together (right side) from fragmented agricultural-natural systems associated with olive groves, pastures and crop mosaics (smaller patch size). This result suggests an increasing fractalization of peri-urban landscapes, mixing residential settlements and arable land (typical association observed at the fringe of Rome). Extensive rural systems further away from the fringe are becoming increasingly fragmented with a latent association between natural covers and productive crops (pastures, olive tree, crop mosaic). Perforation and core patches were mostly associated with built-up areas and arable land. Islet and branch categories are primarily associated with olive groves, pastures and crop mosaic. Component 2 (27.1%) discriminated forests from vineyards evidencing the role of 'edge' and 'loop' categories. Based on these results, the actual landscape structure was intrinsically more complex and less ordered along a landuse intensity gradient, evidencing in turn the important contribution of built-up areas to landscape variability.

Discussion
In recent years, comparative analysis aimed at capturing underlying patterns and trends from land-use databases has been fed by a continuous demand for high-resolution data as well as reliable indicators and new methodologies [60][61][62][63][64][65][66][67]. This work focuses on land-use changes occurring in Rome, which has grown very fast over the last seventy years, to infer different means of urban expansion using exploratory data analysis. The role of a continuous monitoring of land use is a fundamental support for sustainable development policies in urban regions [68]. In areas, such as those of southern Europe, characterized by generally small-fragmented cities [69][70][71][72], polycentric patterns have infrequently been observed with the exception of a few limited cases. In comparison with northern and western European regions [14,16,[73][74][75], typified by polycentric modes of urban development, Mediterranean cities have experienced a partial failure of this model due to their specific socio-economic features [18,76,77]. Recently, in fact, significant changes in morphology and socio-economic structures have been found in many Mediterranean cities as a result of incoming sprawling phenomena [78][79][80] following a more dispersed (and not polycentric) pattern [42,81,82]. In the specific case of the city of Rome, thanks to morphological analysis, how the landscape changed its structure following different waves of expansion of various intensities occurring in recent decades was demonstrated.
In particular, local socioeconomic contexts in the study area have sometimes accelerated the emergence of a more scattered and discontinuous agglomeration to the detriment of a denser and more compact asset, with severe consequences for the health status of environmental components. These changes were probably stimulated by socio-economic drivers such as land speculation, second homes, and internal/ foreign migration [12,23]. More recently, in other Mediterranean cities, urban expansion modes have followed a scattered pattern for settlements, infrastructural facilities and industrial areas [2, [83][84][85]. This translates into a sharp reduction of the average size of core areas and a strong patch fragmentation, especially loops and bridge patches that show an increase in size for most of the land uses considered. In general, changes in the morphology of the agricultural lands have been strongly influenced by sprawl phenomena [11,29,36], confirming the inferences of past studies at both local and continental levels [79].
Economy-oriented factors have strongly influenced the urban growth history of Rome in the last seventy years. The diachronic analysis presented in this paper  provides evidence for the formidable transformations occurring during the time period investigated, which can only partially be attributed to cover changes (increase of urban land uses, decrease of natural and agricultural areas). The major upheavals involve the landscape structure: the core areas of all the considered land uses decline with high magnitudes for agricultural and semi-natural classes, indicating the major role assumed by sprawling phenomena (scattered and chaotic development). The approach followed in this paper laid a good foundation for future studies aimed, on the one hand, at continuously updating the adopted monitoring system and, on the other, at refining/improving the detail level of the adopted databases. In both the cases, we were able to make use of remote information.
The extraordinary wealth of data deriving from satellite/airborne sensors can be profitably used at local/regional scale to achieve the easy updating of land cover maps and a better characterization of the investigated areas. Land-use maps obtained from a multitude of classification methods from remote-sensed imagery have a higher spatial resolution compared with those used in this work (based, for example, on the Corine Land Cover having a coarser spatial resolution: Minimum Mapping Unit (MMU) of 25 hectares for areal phenomena and a minimum width of 100 m for linear phenomena). This can be achieved by relying on long-term datasets of satellite images such as Landsat or SPOT [83,[86][87][88][89] or more recent, free and high-resolution generation of satellite data such as Sentinel 2 [90][91][92]. Furthermore, the enhanced possibility of using finer spectral resolutions and more spectral bands (as in the case of hyperspectral data, see e.g., [93][94][95][96][97]) can result in a very high labelling of the present land cover categories. This facilitates the detection of specific classes, especially urban land uses (see, e.g., [98,99]) and improves the sensitivity of morphological analysis to capture subtle phenomena such as attrition processes (i.e., gradual loss of remaining fragments of peculiar land cover classes), which often indicate the last stage of land transformations subject to heavy fragmentation [100]. Starting from these technical improvements, supported by mathematical morphology and multivariate statistics, it could be easier to notice failure or the effectiveness of adopted policy measures in the perspective of a continuous refinement of territorial planning actions.

Conclusions
When evaluating long-term landscape transformations in peri-urban regions, the empirical results of this study indicate the expected "wasteful" character of sprawling development models leading to a more fragmented land-use pattern with unavoidable effects on the capacity of ecosystems to maintain a complete provision of services and goods. In this field, additional studies are needed to refine the spatially explicit analysis of land-use data matrices in a multidimensional space by adopting appropriate monitoring tools. Our study may provide useful insights on sprawling processes and pushes forward the debate on the future development of sustainable plans in Mediterranean cities. These results can be generalized to urban regions in developed (but peripheral) countries in Europe and, possibly, in other socioeconomic contexts throughout the globe.  Data Availability Statement: All data generated or analyzed during this study are included in this article.

Conflicts of Interest:
The authors declare no conflict of interest.