Land-Use Change Effects on Soil Erosion: The Case of Roman “Via Herculia” (Southern Italy)—Combining Historical Maps, Aerial Images and Soil Erosion Model

Abstract

Land use and land cover (LULC) strongly influence soil erosion/sediment yield, and predicting changes in soil erosion is an important management strategy. Tracing the Earth’s past also helps us better understand the future evolution of the landscape, but research using modern mapping capabilities is hampered by the scarcity of historical landscape information. To fill the data gap and provide an example of how historical maps might be used in land-use change research, we combined an old paper map based on the IT Military Topographical Institute (ITM), aerial photos, and orthophotos to derive land-use history and landscape dynamics at fine scales for a segment of the Roman route “Via Herculia” located in the north-western sector of the Basilicata Region, Italy. Three LULC scenarios were then analysed to represent land use in 1870, 1974, and 2013. Starting from such scenarios, we applied a soil erosion model (Unit Stream Power Eosion and Deposition—USPED) to understand how land-use change over time has modified the areas subject to erosion and deposition. The results show an increase in erosion (from 17% to 20% of the total area) and sediment deposition (from 15% to 19%) over the period 1870–1974. In contrast, over the period of 1974–2013, the results show a decrease in gross erosion (from 20% to 14% of the total area) and sediment deposition (from 19% to 13%).

1. Introduction

Soil erosion is an increasing factor in land degradation and soil loss represents a huge environmental risk everywhere in the world [1,2]. Two major obstacles to the sustainable management of soil and water resources are sediment yield and soil erosion [3]. Any scientifically based soil and water conservation plan and integrated land management require the quantification of these processes [4,5,6,7,8].

Changes in land use are a regular process all around the world, regardless of whether they are caused by natural or anthropogenic processes [9]. They are always determined by the magnitude of the driving force that impacted on the (historical) landscape-land use in a previous period. The increase in soil erosion caused by global human activities has led to increased sediment flow in many places of the world [10,11]. Negative supplementary impacts of soil erosion, such as soil fertility loss, poor water quality, hydrological system changes, and environmental contaminations, have been identified as a severe challenge to human sustainability [12,13,14]. Changes in land cover and land use are important factors for determining the hydrological response at a catchment scale. Many studies have demonstrated that there is a link between land-use change and soil erosion [15,16,17].

Finding solutions for landscape sustainability requires an understanding of the patterns, causes, and effects of landscape change [18,19]. Earth observation (EO) technologies that can precisely identify where, when, and how lands have changed are essential to these activities [20]. The availability of remotely sensed images is a fundamental tool to catch the Earth’s changing landscapes at different scales, especially where climate, growing urbanization, and natural events cause fast and widespread land-use changes [21,22,23,24,25]. The application of EO technologies is helpful for obtaining insights into the processes of land-use changes in relation to socioeconomic causes and related climatic implications, and it can help support effective decisions [26]. The availability of open archives such as Landsat and Sentinel allows measuring forest loss and gain for each 30 m land pixel of the globe [27]; furthermore, such information can be stored quickly using webGIS portals or digital Earth platforms (e.g., Google Earth). Continuous advancement in data analytics is accompanying the growth of geographical data. For example, the set of methods or algorithms for the semi-automatic identification of land cover and vegetation type has evolved in recent years, and it now includes several machine learning techniques such as Random Forests, Gaussian processes, and Support Vector Machines [28,29].

Despite the well-established techniques for mapping present land status, the short-term temporal coverage of satellite images is a strong limitation for the investigation of historical land-use/land cover changes. As a matter of fact, there are few or no maps of the composition of the landscape older than the 80s, although historical data on the landscape can be crucial to better retrace the long-term trends of land-use changes induced by natural factors and human pressure. The only data sources that adequately represent past landscapes of a region are historical drawings or paper maps, such as topographic, cadastral, and military maps. These maps (which can be older than many centuries old) are useful for unravelling long-term land-use history and vegetation dynamics. Their usefulness has become more widely recognised, and community interest in collecting ancient maps is expanding. Currently, old maps are mostly examined by traditional and manual approaches [30].

This study attempts to provide an example of how historical maps can be used to describe the evolution of land use and the dynamics of the landscape. By using a historical (year 1870) map and modern aerial imagery, we conducted a case study in a sector of Southern Italy crossed by an intermountain Roman road, the Via Herculia, dated between the III and IV centuries AD. We selected the “Via Herculia” because of its importance in long-term economic exchange, information transfer, army mobility, territory control, and settlement network growth. The ancient road system is a crucial topic for investigation both for the connection with aspects related to the topography of a territory and for the implications related to the social and economic development of the populations. To this aim, archived data need to be employed to extrapolate the changes in land use, as has been carried out for the Via Herculia case study. With regard to this latter aspect, these data were fed into the USPED (Unit Stream Power Erosion and Deposition) model to understand how the land-use change affected the spatial distribution and rates of soil erosion in the study area.

2. Materials and Methods

2.1. Study Area

The study area includes a buffer area of 2 km drawn around a segment of the Via Herculia route located in the north-western sector of the Basilicata Region. It extends for a length of approximately 46.3 km and covers a total area of approximately 18,350 hectares (Figure 1). It crosses nine municipal territories of Basilicata, from north to south, including San Fele, Filiano, Avigliano, Pietragalla, Potenza, Pignola, Tito, Abriola and Sasso di Castalda.

From a geological viewpoint, the study area crosses along the N-S direction, which comprises a large sector of the axial belt of the chain, and extends northward to the Bradano foredeep areas. The southernmost parts of the site cut the tectonically-controlled intermountain depression of the high valley of the Agri River, whilst the Roman road crossed to the north the morphostructural ridges with high tectonic slopes of the chain axial zone [31].

The studied area is characterised primarily by significantly deformed geological units of the Lagonegro basin and flysch deposits of Miocene syntectonic basins. The Middle Triassic to Miocene Lagonegro units are identified by shallow water, shelf-margin and basinal facies. Middle Cretaceous to Oligocene grey-reddish clays and marls (Argille Varicolori and Corleto Perticara Fms), as well as upper Oligocene to lower Miocene marls and volcaniclastic sandstones (Tufiti di Tusa Fm), outcrop widely in the research area. Upper Miocene deep-sea deposits represented by conglomerates, sandstones, and pelites characterised the Gorgoglione Flysch Fm., which unconformably overlies the Lagonegro units. Marine to continental clay, sandstone, conglomerate, and Quaternary continental deposits make up a thick Pliocene to Pleistocene clastic succession that represents the youngest deposits of the study area.

A modified version of [31] lithological map (Figure 2) was used to determine which lithologies were most significantly impacted by the Via Herculia passage and useful for road construction.

From a lithological basis, the Via Herculia mostly traversed a terrain comprising terrigenous rocks of the heterogeneous complex, with a predominance of the clay and stone component (cag and clap, Figure 2), as well as unconsolidated slopes and colluvial deposits (cg and dc, Figure 2).

Between the III and IV centuries AD., Via Herculia was constructed, extending from Grumentum (Grumento Nova, PZ) to the Ofanto River (between Candela, AV, and Melfi, PZ) [31]. It connects the Appia in the north to the ab Regio ad Capuam in the south and was designed primarily with economic utility in mind.

2.2. Historical Map, Aerial Image and Orthophoto

With respect to the dataset, we used the following: (1) a paper topographic map made in 1870, (2) aerial images of IGM acquired in 1974; (3) and a 2013 orthophoto acquired by the geodata service of the Basilicata Region (RSDI) (example dataset in Figure 3).

Specifically, the old paper map used is a military map made in 1870 by the Military Topographical Institute (ITM). It is among the most accurate maps of that time, depicting details about houses, fences, names of residents, roads, vegetation, drainage, rivers, and fords. The paper version has a scale of 1:50,000, and the digital copy has a resolution of 300 dpi.

The 1974 aerial images are in TIFF format and have a resolution of 2400 dpi and a scale of 1:16,000. The acquisitions were carried out at an altitude of 3000 m using a WILD machine with a focal length of 152.36.

Finally, the 2013 orthophoto obtained from the RSDI (main channel for disseminating spatial information from the Regional Spatial Data Infrastructure of the Basilicata Region) is characterised by a spatial resolution of 20 cm.

2.3. USPED Model

Utilizing the structure of the empirical Revised Universal Soil Loss Equation (RUSLE), the Unit Stream Power Erosion and Deposition (USPED) model calculates the average soil loss (A). USPED is a 2-dimensional model of erosion and its modelling was carried out based on the assumption that erosion and deposition primarily depend on the sediment transport capacity of the surface runoff, in contrast to the 1-dimensional RUSLE model.

In particular, the following was used:

A = R·K·LS·C·P

(1)

where

A is the annual average soil loss;

R is the rainfall intensity factor;

K is the soil erodibility factor;

L is the slope length factor;

S is the slope steepness factor;

C is the land cover factor;

P is the soil conservation or prevention practice factors.

The authors of [32] developed the equation based on the upslope contributing area that takes into account both the profile (in the downhill direction) and the tangential (perpendicular to the downhill direction) curvature [33]. Using the following equation, net erosion or deposition (ED) inside a grid cell is calculated as the divergence of sediment flow (change in sediment transport capacity):

ED = ∂ [(A·cosα)/∂x] + ∂ [(A·sinα)/∂y]

(2)

where α is the terrain’s aspect (in degrees).

The fundamental physical assumption of the model is that net erosion pixels correspond to areas of profile convexity and tangential concavity (flow acceleration and convergence), whereas net deposition areas correspond to areas of profile concavity (decreasing flow velocity).

The R factor is determined by rainfall–runoff characteristics, which are influenced by geography and altitude. This factor was calculated using the empirical equation created [34] and was used previously in [35]:

EI30 = 0.1087 (P₂₄)^1.36

(3)

where

EI30 is the rainfall erosivity (MJ mm h⁻¹ ha⁻¹ yr⁻¹);

P24 is the daily rainfall amount in mm.

Equation (3) was calculated using daily precipitation data for the period from January 2010 to December 2020.

The soil erodibility factor (K) was calculated as the rate of soil loss per rainfall erosion index unit (Mg h MJ⁻¹ mm⁻¹) by using a standard Wischmeier erosion plot [36].

The K factor represents the soil profile response to the erosive power of rainfall events and was calculated using the following equation:

K = 0.001317 [2.1·10⁻⁴ (12−M) [(Si + f_S)(100 − c)] 1.14 + 3.25(a − 2) + 2.5(P − 3)]

(4)

where

M is the organic matter content (%);

Si is the silt fraction in % (2 to 50 μm);

f_S is the fine sand content in % (50 to 100 μm);

c is the clay content in % (less than 2 μm);

a is the structure;

P is the permeability class (within the top 0.60–0.70 m).

Factor 0.001317 is derived via division by 100 of the conversion value (0.1317) to the SI.

Texture, structure, and permeability data were obtained from previous papers dealing with the physical and chemical characterization of deposit outcropping in the foredeep area of the southern Apennine chain [37,38]. The K factor map was drawn by combining land-use maps (above-mentioned literature data) and lithological data about the texture and permeability of the main lithological units of the study area.

The topographic factor LS (dimensionless) was obtained using the following equation:

LS = (m + 1)·(A_c/α₀)^m·sin(β₀/0.0896)ⁿ

(5)

where

A_c is the upslope contributing area per unit of width (m);

α₀ is the length (72.6 ft, equal to 22.13 m) of the standard terrain;

β₀ is the angle (9%, equal to a 5.16 slope degree) of the standard terrain;

m and n (0.6 and 1) are empirical exponents [32].

The topographic parameters, slope, and particular catchment area were calculated using a 5m DEM collected using the RSDI service.

According to literature data [39,40], the C factor, which indicates the effects of crops and management techniques on soil erosion rates, was calculated using values allocated based on vegetation cover, density, and monthly rainfall runoff erosivity. According to the C values proposed in the literature, C values were assigned to the resulting 13 land-use classes (for example [41]).

For the P factor, there is no significant supporting practice in the study area, so its value is considered equal to 1.

3. Results

3.1. Land-Use Change (1870–2013)

The study area was classified into six land-use classes for each period: (1) urban, industrial, and commercial area (

Table 1. The results of land-use change in 1870, 1974, and 2013 in “Via Herculia”.

Land-Use Type	Land Use 1870		Land Use 1974		Land Use 2013
	Area (ha)	Percent	Area (ha)	Percent	Area (ha)	Percent
Urban, industrial, and commercial area	286.0	2.1	796.8	5.9	2517.5	18.5
Arable land	5000.4	36.8	9806.3	72.7	6987.7	51.5
Fruit trees, vineyards, and olive groves	752.1	5.5	179.6	1.3	44.5	0.3
Pastures and natural grasslands	5700.2	42.0	1337.8	9.9	1434.6	10.6
Forest area	1525.4	11.2	1070.0	7.9	2357.6	17.4
Water	322.6	2.4	300.8	2.2	235.3	1.7

); (2) arable land; (3) fruit tree, vineyard, and olive groves; (4) pastures and natural grasslands; (5) forest area; (6) water.

The comparison of land-use change shows how urban, commercial, and industrial areas increased from 286 hectares (1870) to 797 hectares (1974) and up to 2517 hectares in 2013. In this way, urban areas have increased by 16.4% from 1870 to 2013.

Arable land had an increase in coverage from 5000 hectares in 1870 to 9806 hectares in 1974 (an increase of 35.9%), while from 1974 to 2013, there was a decrease of 2819 hectares, which corresponds to 21.2%.

Fruit trees, vineyards, and olive groves have a general decreasing trend from 1870 to 2013 with respect to land cover. The trend decreased from 752 hectares (1870) to 180 hectares (1974) and then decreased to 44 hectares in 2013; it went from 5.5% to 0.3% of the total occupied area.

Pasture and natural grassland areas decreased from 5700 hectares in 1870 to 1338 hectares in 1974; then, they were almost stable at 1435 hectares in 2013. This went from 42% of the occupied area in 1870 to 10.6% in 2013.

The forest area decreased by 455 hectares between 1870 and 1974, finally increasing to 2358 hectares in 2013. Thus, the total area occupied changed from 11.2% in 1870 and 7.9% in 1974 to 17.4% in 2013.

Finally, the area occupied by water is almost stable over the considered period: 2.4% in 1870, 2.2% in 1974, and 1.7% in 2013.

3.2. USPED Model

For the period of 2000–2020, the rainfall intensity factor (R factor) calculated for the study area using available hourly rainfall data is 860 MJ mm ha⁻¹ h⁻¹ yr⁻¹.

The soil erodibility factor (K factor) (Figure 4a) was derived from lithological information in the geological map and shows a range between 0 and 0.06. A value of 0 was attributed to the presence of a lake, while a value of 0.06 was attributed to unstable areas predisposed to erosion, such as landslide areas.

The slope length and steepness factor map (LS factor, Figure 4b) show a value between 0 and 100, an average value of 0.07, and a standard deviation value of 1.65.

In the study area, the land use and land management factor (C factor) (Figure 5) range from a maximum of 0.4 assigned to bare or degraded areas to a minimum of zero attributed to urban, industrial, commercial, and transport areas. The C factor map was generated for 1870 (Figure 5a), 1974 (Figure 5b), and 2013 (Figure 5c) by considering land use.

Figure 6 represents the results of the USPED model’s implementation. Each pixel has been categorised using the eleven erosion/deposition classes listed below:

• Extreme erosion (<−40 Mg ha⁻¹ yr⁻¹);
• High erosion (−40/−20 Mg ha⁻¹ yr⁻¹);
• Moderate erosion (−20/−10 Mg ha⁻¹ yr⁻¹);
• Low erosion (−10/−5 Mg ha⁻¹ yr⁻¹);
• Very low erosion (−5/−2 Mg ha⁻¹ yr⁻¹);
• Stable (−2/2 Mg ha⁻¹ yr⁻¹);
• Very low deposition (2/5 Mg ha⁻¹ yr⁻¹);
• Low deposition (5/10 Mg ha⁻¹ yr⁻¹);
• Moderate deposition (10/20 Mg ha⁻¹ yr⁻¹);
• High deposition (20/40 Mg ha⁻¹ yr⁻¹);
• Extreme deposition (>40 Mg ha⁻¹ yr⁻¹).

Figure 6. Mean annual soil erosion of the study area as inferred by the USPED model: (a) 1870; (b) 1974; (c) 2013.

Figure 6. Mean annual soil erosion of the study area as inferred by the USPED model: (a) 1870; (b) 1974; (c) 2013.

The distribution of classes in the USPED model (Figure 7) shows that most of the area analysed is stable, particularly in 1870 (Figure 7a) at 68%, 1974 (Figure 7b) at 61%, and 2013 (Figure 7c) at 72%. Overall, classes showing erosion (extreme, high, moderate, low, and very low erosion) increased from 17% to 20% for the period of 1870–1974 and decreased from 20% to 14% for the period of 1974–2013. In contrast, the classes that represent deposition (extreme, high, moderate, low, and very low deposition) assume values of 15% for 1870, 19% for 1974, and 13% for 2013.

4. Discussion and Conclusions

A lack of reliable, spatially accurate historical land-use/land-cover data continues to be a limitation to the study of land change and global environmental change sciences. We proved that by using old maps, this gap could be partially filled. The integration of the military map, aerial photos, and the modern orthophoto enabled deriving long-term landscape change and land-use history from 1870 to 2013 using high spatial resolutions.

In general, an increase and decrease in land use resulted in changes in soil loss. The study area (Via Herculia roman road and surrounding sectors) has undergone severe land-use changes over time. This is due to its important role in economic exchange, information transfer, and the growth of settlement networks in historical and recent times. As demonstrated by [35], the topic of viability is crucial because it is frequently connected to the topographical features of territory (and consequently to land use), similarly to how the social and economic developments of people are closely linked to a road network. Additionally, Sabia [42] points out that the presence of roads and routes greatly influenced the use of land, generating productive activities related to pastoralism and agriculture that were perpetuated for centuries.

From a methodological point of view, the use of historical maps for LUC analysis can involve tedious manual processing. In fact, there are several steps that need to be followed with respect to using the historical paper map [35] or the digital and georeferenced map in a GIS environment. Among the main disadvantages of such an approach, there are the errors that are introduced in the transition from paper to digital maps. Furthermore, most of the time, the resolution of the data obtained from these map data is very coarse compared to recent data.

In this study, the land-use dynamic (LUD) analysis starts from the land-use situation in 1870; it can be observed that the land use of the study area at that time was mainly characterised by two classes: (1) pasture and natural grassland; (2) arable land. The same classes were present for the 1974 year, but a large increase (about 35%) in arable land can be observed. The increase in arable land was made possible by new technologies using motorised vehicles, which made it possible to cultivate substantially larger areas using less time and physical effort. Finally, in 2013, there was an increase in the “Urban, industrial and commercial area” class (about 12% more than in 1974), with a decrease in “Arable land” and an increase in the “Forest area” class. The latter change reflects the phenomenon of the abandonment of rural areas, which has been occurring over the last 20 years (as already shown by [35,43,44]).

The main steps of the research study can be summarized as follows: (i) mapping of the LULC using different input maps; (ii) map generation for USPED parameters; (iii) calculation and comparison of soil erosion value for the different years considered.

The USPED model was applied using land uses for the following three periods: 1870, 1974, and 2013. We ran the USPED model, taking into account that the support practice factor was not significant in the study area. Each raster cell of the USPED model was divided into eleven classes of erosion and deposition. This classification was derived by adopting (i) the erosion risk classes used for Italy by [35,45] and (ii) the concepts formulated for Mediterranean environments [46]. The results obtained by applying the USPED model show that the value obtained for the “Stable areas” is higher in 2013, and this is due to two factors: (1) increase in “Forest Area” (consequence of land abandonment compared to 1974) and (2) increase in “Urban area” due to overbuilding that has characterised the territory since the 1990s. While higher erosion and deposition values are observed in 1974 (compared to 1870 and 2013), this is essentially due to the large agricultural areas present in that year, which makes the territory fragile to erosion in the absence of vegetation cover.

The results show that, in general, land use is one of the factors that most influence the behavior of the USPED model, resulting in variations in soil erosion. In Figure 8, a small region of the study area characterised by land-use change over time was examined. Specifically, in 1870, there was a large area of woodland that was removed in 1974 in favor of pasture and arable land, and then it was reforested in 2013. This change in land use over time is reflected in the variance of erosion/deposition, especially in 1974 when there was a lack of protection provided by vegetation cover, as shown by the histogram in Figure 8.

The results of this research study emphasized the importance of monitoring processes in scenarios of land-use change. Climate and land-use change have the greatest impact on all USPED factors. Climate change is a natural and unpredictable phenomenon that is beyond human control. Thus, if land-use change can be controlled, despite future climatic change, soil loss can be controlled and even reduced significantly.

Finally, it is important that we preserve and geotag historical data because this will help future generations document long-term landscape changes and plan the future use of the land in a better way in response to its orographic and productive vocation.