The sustainable development of remote and ecologically sensitive regions, such as mountain landscapes, depends also upon land use management. This study examines land use dynamics in the Alpine region of Lombardy, Italy, as delineated by the eco-regions defined by the Italian Statistical Institute (ISTAT). By analyzing land use data from Regional Agency for Services to Agriculture and Forestry (ERSAF), we assess the 2013 and the 2018 status for 19 relevant land use categories, along with the identification of net changes observed during this period. This exploratory work, based on the use of high volume data, lays the foundations for a wider project that aims at modeling spatial autocorrelation, outlining the challenges this estimation would lead to, when managing such a huge volume of data via AI algorithms.
Sustainable land use has become a key issue in mitigating the efects and in adapting to the climate
change. In particular, this issue is even more crucial in the case of remote and ecologically sensitive
areas, such as mountainous regions. These regions, characterized by fragile ecosystems and complex
socio-economic dynamics, are particularly vulnerable to the increasing pressure of urbanization and
soil loss caused by hydro-geological instability. Consequently, the preservation of biodiversity, the
mitigation of land degradation and the promotion of sustainable development require eficient planning
for the conservation and management of human activity [
Developing such plans and formulating public policies that concurrently support ecological integrity
and socio-economic resilience present significant challenges. This underscores the need for targeted
research into land use dynamics, a task that often exceeds the capacities of local administrations due
to resource constraints. Efectively addressing these challenges requires adopting a spatially nuanced
approach to land use management. For this purpose, we need a shift from administrative borders to the
concept of eco-regions, also referred to as ecological regions. These are defined as relatively extensive
units of land or water that contain a distinct composition of species and natural communities [
The objective of this work is to explore the possible integration of the Italian eco-regions system and the land use data available for Lombardy. We focus on the specific case of the Alpine province for 2013 and 2018. In addition, the challenges associated with the modeling of this type of data are discussed, with a particular focus on AI algorithms implementation.
Following a brief description of the study area, of the data, and of the adopted methodology, we discuss our early results. Finally, future research directions will be outlined.
The study area we analyze encompasses 175 municipalities within the Italian Lombardy region, which have been classified by ISTAT into the Alpine Province. Specifically, at the fourth level of the hierarchy, 26 municipalities are classified as “Sottosezione Alpi Nord-Occidentali” (Northwestern Alps Subsection), while the remaining 149 municipalities belong to the “Sottosezione Alpi Nord-Orientali” (Northeastern Alps Subsection). To comprehend the magnitude of this area, it is necessary to consider that the Lombardy region encompasses an area of approximately 24,000 km2, and the region covered by our study (the Alpine region) constitutes approximately one-fourth of that size. The geographical location of the study area is illustrated in panel (a) of Figure 1. This depicts the Lombardy region by municipal boundaries, with colors representing the respective subsections. The two Alpine subsections analyzed, the Northwestern (identified by the code 1A1b and color orange) and the Northeastern (identified by the code 1A2c and color green) correspond to the northernmost. Given the recognized role of altitude in delineating alpine provinces, the mean altitude for each municipality is reported in panel (b) of Figure 1. The darker shades of color in the figure correspond to higher mean altitudes. In particular, norther and northeastern municipalities, which represent the Italian border with Switzerland, appear to exhibit higher mean altitudes. We expect that this feature of the area will afect the spatial patterns of the phenomena we are going to study. 46.5°N 46.0°N 45.5°N 45.0°N 9.0°E 9.5°E 10.0°E 10.5°E 8.5°E 9.0°E 9.5°E 10.0°E 10.5°E 11.0°E
11.5°E (a) Municipalities of Lombardy by ecological subsec- (b) Mean altitude per municipality of the Alpine subsections. tions.
The land use data studied have been extracted from the Regional Agency for Services to Agriculture and Forestry (ERSAF) dataset. This dataset is obtained by combining land use and land cover data with information coming from the Lombardy Region’s Agricultural Information System (SIARL) database 46.6°N Sub−sections 46.4°N 1A1b 11AA22ca 46.2°N 1B1b 1C1a 46.0°N 45.8°N
Our exploratory analysis involves the estimation of two types of statistics, one focused on the stock and the other on the flow: status and net change, respectively.
The status share is used to describe the proportion of total area that has been classified as a particular land use category. The status share for a specific category at a certain time point can be defined as: where , represents the single -ℎ unit classified in the category at time , and is the number of cells belonging to the total area .
The net change is defined as the diference in the status share for a given land use category
time points and − . This can be computed as:
∆ ,, = , − ,− .
between
(2)
These two measures are commonly adopted to estimate land use and land cover [
The results pertaining to the status and the net change can be found in Table 1. Based on the status share (expressed as a percentage of the total cells) for 2013 and 2018, the four most prevalent categories, namely, Woods and tree crops (35.6 and 40.5 for the two years), Forage (27.0 and 23.2), Natural sterile areas (13.9 and 11.7) and Natural vegetation (12.8 and 9.8), collectively account for more than the 80% of the study area.
The dynamics between 2013 and 2018 can be observed by checking the “Change” section (last three columns of Table 1). In particular, it is reported in net terms, shown by Eq.(2), and, to provide a more comprehensive analysis, in year-to-year relative change (in percentage) and in squared kilometers (km2) terms. By examining the net change values, it is possible to observe how Woods and tree crops and Wasteland are the land use that increased the most, in terms of status share (4.9 and 4.3, respectively), while Forage, Natural vegetation and Natural sterile areas are those that decreased the most (-3.8, -2.9 and -2.1, respectively). By considering the year-to-year percentage change, the land use categories subject to the highest positive changes are Resting land (+3425.0%), Industrial plants and dried legumes (+601.2%) and Wasteland (+207.4%). Negative signs are observed for Seeds (-67.3%), Natural vegetation (-23.2%), and Natural sterile areas (-15.6%). Similar conclusions, to those discussed when focusing on net changes, are obtained by also considering the dynamics in squared kilometers terms between 2018 and 2013, allowing for a more understandable unit of measurement. The calculation of this value can be achieved for each land use category by computing the diferences between the two years number of cells ( # cells) and by subsequently converting the value from squared meters to squared kilometers. Given that each individual cell is 20x20 m, the total number of cells is multiplied by 400 and divided by 106. By employing this method, it is possible to ascertain that moving from 2013 to 2018 the areas dedicated to Forage, Natural vegetation and Natural sterile areas showed a loss of 222.34 km2, 171.99 km2, and 125.59 km2, respectively. Concurrently, the land dedicated to Woods and tree crops and Wasteland increased by 283.31 km2 and 250.96 km2, respectively.
In summary, an examination of the statistics presented in Table 1 reveals a decline in naturally vegetated areas (vegetation and sterile areas) and in the land for forage production. Concurrently, there has been an increase in areas allocated to wood and tree crop production, as well as in wasteland areas.
To obtain a general measure of variability in the land use able to quantify the total variation observed by moving from 2013 to 2018, we compute a weighted mean of the absolute relative changes using as weights the number of cells observed for each category in 2013. This statistic, equal to 0.1874, points out that in 5 years the 18.74% of the cells changed their land use (corresponding to an annual average percentage of 3.75%). This information is useful for future comparisons between other time points or windows in order to understand if the observed changes are more or less pronounced.
To further enhance the findings presented herein, it would be beneficial to consider the estimates of
transitional changes in land use categories. This would permit the tracking of a cell’s transition from
one category to another. This approach would improve the comprehension and reconstruction of the
dynamics that the territory have experienced between 2013 and 2018. However, the implementation
of additional data harmonization techniques is necessary, given that a direct overlap of grids referred
to the two years is not feasible. For instance, a possible strategy could be the establishment of a fixed
common cell structure over the years. This can be achieved through the realignment of the cells with
or without transitioning to a lower level of detail, i.e. aggregating 20x20 m cells to obtain a less refined
grid. For these purposes, AI solutions could be a viable option. An example of developed AI strategy
helpful in this framework can be found in the paper of [
A further research development would focus on distinguishing between structural land use changes
and annual rotation of land use specific of agricultural practices. Finally, as already done in the
literature, the observed categories could be collapsed into a reduced number of broader categories to
better represent the dynamics of land use change [
Such a type of results would permit a general coordinated valorization of the whole area, such as interventions in favor of certain destination use, but also regarding the preservation of the land heterogeneity and of sustainable use of the territory. For these reasons it is necessary to acknowledge the importance of such type of data, that actually are available for a limited time window.
Finally, the next step of this research is the modeling of spatial autocorrelation, an aim that comes
with several challenges. Firstly, it is important to note that classic AI algorithms have been developed on
the basis of intrinsic i.i.d. assumptions. Consequently, these algorithms are not able to explicitly model
dependence structures among the data. A variety of strategies for incorporating spatial information into
AI models have been proposed. They range from more immediate approaches, such as the inclusion
of coordinates as absolute spatial references, to more complex and challenging modeling techniques,
involving the integration of the covariance matrix into AI algorithms [
In conclusion, these eforts may also be oriented towards the development of an AI solution based on past cells occurred changes and spatial patterns. This method will allow to predict the cell land use across time points not only generally, but also referring to each specific cell (or neighborhood of cells). Such a solution would be useful, in a data-driven perspective, to guide policy-makers in setting regulations or in planning resources according to the expected patterns in land use.
The authors have not employed any Generative AI tools in writing this paper or processing the data.