Land cover / land use classification and its changes over time 

​

The service can be seen as an investment decision-making support tool able to monitor overall urban development processes, and the detailed spatial distribution and evolution of urban land use classes and features. 

​

A recent situation in the four cities was presented in so-called baseline urban classification maps (BUCs), based on Pléiades imagery acquired during 2014 (spatial resolution 0.5 m). These contain information on land use such as artificial surfaces, non-artificial surfaces and other natural and semi-natural areas. The broadest level of categorisation (Level I) distinguishes among broad land cover types: urban, agricultural, forest, water, irrigated lands, etc. For urban land, the second level of categorization (Level II) distinguishes among thematically detailed land uses, e.g. high- and medium-dense urban fabric, discontinuous urban fabric, main and secondary roads, and rail network. The methodology is based on semi-automatic pixel- and object-based image analysis with post-classification, including (especially for the Level II categorisation) analysis of textures, patterns, alignments and arrangements. 

​

The land use change maps reflect a region’s urban land use history, detecting major changes that occur in the urban land use classes, including the urban extent itself and major transportation networks. The change maps were based on a reference period (a combination of SPOT 4 and SPOT 5 imagery at 5 m resolution acquired during 2002–2004) and the BUC maps from 2014. 

​

One possible type of analysis of the land use changes consisted in estimating the geometric city centre, after which the degree of physical continuity of urban settlements can be computed by summing up the area of artificial land use classes within concentric buffers of 100 metres centred on the geometric city centre. This calculation allows the identification of the central continuum of each city and of the artificialised land occupation, and helps to understand its structure. The land occupation change over time can help decision-makers take the appropriate decisions for future development. 

​

The land use maps allow also studying comparatively, in addition to the degree of land consumption, the morphology of the urbanisation of the defined cities. This can be done by studying the variation of entropy (or fragmentation). This indicator allows understanding the similarities and differences between city landscapes in different periods and characterising the typology of the urban sprawl process.

​

Landslide risk assessment 

​

Being a mountainous country with a high occurrence of steep slopes, Armenia is one of the most affected countries in terms of landslides. This service helped characterise and delineate the areas subject to landslides by delivering the following products:

• Landslide Inventory Map (LIM),

• Digital Elevation Model (DEM),

• Landslide Susceptibility Map (LSM). 

​

The LIM was produced for Dilijan using measurement of ground deformation derived from satellite-based SAR (Synthetic Aperture Radar) interferometric data together with visual interpretation of high-resolution optical imagery and morphological analysis of the DEM. The satellite sensors used were Envisat ASAR and ALOS PALSAR. The interferometric approach is exploited to measure, at specific points, ground deformations, and to construct time series of their displacement. The effectiveness of remote sensing techniques is particularly relevant for wide regions and inaccessible places, for which conventional (ground-based) analysis cannot compare in terms of timely update, cost-efficiency, systematic coverage, accuracy and precision, due to the large extent of the investigated area. 

​

For the city of Dilijan, the LIM was produced at basin scale. Detection (identification of features related to topographic surface movements) and mapping are mainly based on visual interpretation of different types of remote sensing products. A large number of instability phenomena were identified (totalling 204), thereby increasing the number and area of most of the already known landslides. The landslides are typically low-motion, with rates from a few mm/year to several cm/year. For each landslide, its boundary was drawn and the on-going motion described. Landslides were classified based on the prevalent type of movement, the estimated depth, and the relative age. Landslide types were also determined based on the local morphological characteristics. Pre-existing landslides are represented by flows, slides, and complex landslides. For a small group of landslides (about 15%) it was not possible to determine the landslide typology. 

​

The most important phenomena were detected on the southern slope of the Dilijan valley, where several large landslides were recognised. These are probably complex landslides affected by partial reactivations. In this area, several other phenomena were mapped in the upper part of the valley (in the right tributaries of the Ahgstev River) where the vegetation cover is absent or minimal. Areas with superficial and channelised erosion affecting the surface deposits were classified with the label "superficial erosion", characterised by lack of soil and vegetation cover, with an ephemeral drainage. 

​

The state of activity of most mapped phenomena can be considered as dormant or inactive by merely evaluating the so-called SqueeSARTM line-ofsight (LOS) velocities. The active and most dangerous landslide is the one located in the Mets-Tala district, south of the Aghstev River. Also west of Little Maymech Mountain, an active complex landslide was observed, characterised by a velocity of –11 mm/year but its location does not threaten urban areas. 

​

The Digital Elevation Model product covered the area of Dilijan and was meant to demonstrate the ability of high-resolution SAR satellite data (in this case COSMO-SkyMed) to enhance the freely-available SRTM DEM product by using the so-called DInSAR (Differential Interferometric SAR) technique. Pairs of acquisitions from satellites at slightly different times are combined resulting in a 15 m resolution elevation product with 7 m vertical accuracy, available wherever interferometric coherence is present. Roads, bare and rocky areas, and urban areas generally yield good results, while snow-covered or vegetated areas (with changes from one acquisition to the following) are not possible to map. 

​

When dealing with landslide investigations, the ultimate aim is the estimation of the risk posed by existing and/or future slope failures to population or infrastructure. Landslide inventory mapping is typically one of the first workings steps toward a further hazard and risk assessment. The Landslide Susceptibility Map (LSM) of the Dilijan area was produced using a simple implementation of the Random Forest (RF) machine-learning algorithm performing a multivariate classification. The method considers numerous parameters in order to avoid the subjectivity in the choice of explanatory variables: elevation, slope, aspect, flow accumulation, land use from Service 1, etc. The analysis was performed on a grid resolution of 81m. 

​

The obtained results highlight a large area, along the right bank of the Ahgstev River, characterised by very high susceptibility. Other very high or highsusceptibility zones are present in the Golovino area and in the lower part of the valleys of the right tributaries of the Ahgstev River.