CC BY
14
УДК 528.48
КЛАССИФИКАЦИЯ РАСТИТЕЛЬНОГО ПОКРОВА И ВИДОВ ЗЕМЛЕПОЛЬЗОВАНИЯ СЮННИКСКОГО МАРЗА ПО СПУТНИКОВЫМ СНИМКАМ RAPIDEYE
Артак Пилоян
Ереванский государственный университет, 0025, Республика Армения, г. Ереван, ул. Алекса Манугяна, 1, магистр географии, е-mail: artakpiloyan@ysu.am
Арам Геворгян
Ереванский государственный университет, 0025, Республика Армения, г. Ереван, ул. Алекса Манугяна, 1, доктор биологических наук, магистр по наукам об окружающей среде и природоохранной политике, е-mail: aram_gevorgyan@yahoo.com
Дэвид Варданян
USAID (Агентство США по международному развитию), Программа «Чистая вода и энергия», 0009, Республика Армения, г. Ереван, Терян, 105/1, бизнес центр «Цитадель», магистр по геологии, е-mail: dvardanyan01@gmail.com
Рассматривается теоретическая и практическая деятельность по созданию классификации растительного покрова и видов землепользования территории бассейна южных рек Армении на основании Европейской системы Корине и спутниковых снимков RapidEye. Использовались полуавтоматические методы изокластерной контролируемой и неконтролируемой классификации для создания финальной карты LCLU (земного покрова и землепользования). Работа проводилась в течение десяти месяцев и завершилась в октябре 1914 года. В результате проведенной работы карта LCLU Южного бассейна была занесена во 2-й уровень классификации по категориям КОРИНЕ; были составлены обновленные карты земного покрова и землепользования по бассейнам рек Воротан, Вохчи и Мегри. Кроме того, была определена точная площадь по каждому классу LCLU, что дало возможность предоставить обновленную информацию по земельному балансу на территории Южного речного бассейна.
Ключевые слова: классификация, растительный покров, землепользование, дистанционное зондирование, RapidEye, CORINE, Армения.
LAND COVER/LAND USE CLASSIFICATION FOR SYUNIK MARZ USING RAPIDEYE SATELLITE IMAGERY
Artak Piloyan
Yerevan State University, 0025, Republic of Armenia, Yerevan, 1 Alex Manoogian, M.Sc in Geography, е-mail: artakpiloyan@ysu.am
Aram Gevorgyan
Yerevan State University, 0025, Republic of Armenia, Yerevan, 1 Alex Manoogian, D. Sc., M.Sc in Environmental Sciences and Policy, e-mail: aram_gevorgyan@yahoo.com
David Vardanyan
USAID Clean Water and Energy Program, 0009, Republic of Armenia, Yerevan, Teryan 105/1, "Citadel" Business Centre, M.Sc in Geology, e-mail: dvardanyan01@gmail.com
This paper presents theoretical and practical activities devoted to land cover/land use classification (LCLUC) in the Southern Basin Management Area of Armenia based on EU CORINE system using RapidEye satellite imagery. The semi-automatic iso-claster unsupervised
and supervised classification methods were used to create final LCLU map. The activity lasted for 10 months and was completed in October 2014. As a result of the activity, LCLU for the Southern Basin was classified into Level 2 CORINE categories and updated land cover/use maps for Vorotan, Voghji and Meghriget river basins were constructed. In addition, precise areas for each LCLU class were computed, which provided updated information on the land balance in the Southern Basin Management Area.
Key words: classification, land cover, land use, remote sensing, RapidEye, CORINE, Armenia.
Introduction. The activity was performed within the USAID Clean Energy and Water Program (CEWP) which was a 4-year initiative to assist the Government of Armenia in the development of river basin management plans for selected river basins, as well as to strengthen the decision-making process in the field of water management through application of contemporary technologies. One of the tasks of the Program was devoted to land cover/land use classification for the Southern Basin Management Area of Armenia based on EU CORINE system using RapidEye satellite imagery.
CORINE Land Cover (Coordination of Information on the Environment Land Cover) refers to a European program establishing a computerized inventory on land cover of the EU member states and other European countries, at an original scale of 1 : 100,000, using 3-level nomenclature. Minimum mapping unit is 25 hectares and minimum width of linear elements is 100 meters. Five main categories at CORINE Level 1 are "artificial surfaces", "agricultural areas", "forest and semi-natural areas", "wetlands", and "water bodies". There are 15 classes in Level 2 and 44 classes in Level 3 [1, 2].
Study Area. The Southern Basin Management Area of Armenia completely coincides with the Syunik province of Armenia and covers an area of 4506 km2 (15% of total area of the country). The Southern Basin includes the basins of the rivers of Vorotan, Voghji and Meghriget (fig. 1).
Fig. 1. The area of study
Methods. Digital image data from RapidEye AG were acquired over the whole of Armenia in 2013 (July-August). The RapidEye is multi-spectral, high-resolution, orthorectified imagery that is reasonably priced compared with other high-resolution satellite image products and the earth's surface area can be revisited daily. The spatial resolution of the images is 5-m with a depth of 12 bits per channel. Spatial resolution is defined as the minimum size of terrain features that can be distinguished from the background in an image, or the ability to differentiate between two closely spaced features in an image. A nearly fully automated pre-processing system generates orthorectified and atmospherically corrected image tiles within 48 hours after acquisition. The multi-spectral imagery of RapidEye includes commonly used RGB and and near infrared (NIR) spectral bands, plus a red-edge spectral band between the red and the NIR that measures variances in vegetation, allowing for species separation and monitoring vegetation health. A total of 100 image scenes (captured over a period of July and August, 2013) were provided by RapidEye AG to cover the footprint of Armenia. As for the study area of the Southern Basin Management Area a total of 16 image tiles were available.
The land cover/land use classification was based on a guideline document [3] and included the following phases:
(1) Data preprocessing: Most of the RapidEye images for Armenia were captured over a two-month period under different environmental conditions with different color intensities. The colors of the images were not balanced among the 96 tiles covering the whole territory of the country. Thus, the initial step was a mosaicking satellite images and grouping the tiles with the same color intensity into larger tiles.
(2) Composing image bands to visually examine the general land cover/use patterns: The RapidEye images were already converted into .tiff file format with five bands (blue, green, red, red-edge and NIR). After some image adjustments (brightness, contrast, and gamma), it was possible to visually examine the general land cover/use pattern and to compare the results with other ancillary data, such as existing GIS layers on settlements, forestry, river network, road network, lakes, irrigation canals, etc.
(3) Performing the Unsupervised Classification of the RapidEye images to identify 50 land cover/use classes for 5-band imagery. During this phase the following tasks were implemented: (a) mosaicking satellite images; (b) comparing the obtained mosaic with existing GIS layers (settlements, forests, road network, river network, etc.); (c) adjusting the mentioned above GIS layers under ArcGIS 10.1 environment; (d) gradual reclassification via grouping the classes into 13 classes of CORINE Level 2; and (e) identification and selection of problematic LCLU areas for groundtruthing.
(4) Conducting field surveys (groundtruthing) including identification of LCUC types at pre-selected sites in accordance with the 15 CORINE classes and marking their locations with the GPS receiver (waypoints or tracks). After the field surveys, the collected data (tracks/waypoints) were downloaded from GPS to DNRGPS software and checked by overlaying the waypoint/track shapefiles on the Landsat 8 imagery, using ArcMap 10.1 software.
(5) Performing the Supervised Classification (based on the results of ground-truthing). This phase included (a) carrying out field surveys at the pre-defined problematic areas; (b) taking representative photos of the problematic areas; (c) downloading the collected tracks/waypoints with identified CORINE classes; and (d) checking the collected data using auxiliary imagery (Google Earth, Landsat 8).
Results and Discussions. The described methodology for classification of land cover/use was applied for three separate river basins of Syunik province, namely the Vorotan, Voghji and Meghriget basins. The preprocessing step included viewing different compositions of the 5-band imagery to identify distinctly visible land cover categories. In particular "natural colors" (band 3-2-1) and "infrared colors" (bands 53-2) were used to identify continental water bodies and areas under vegetation.
Performing the unsupervised classification resulted to 50-class mosaic, which was then reclassified into 15 CORINE Level 2 thematic classes (i.e., urban fabric, arable land, forests, permanent crops, pastures, open spaces, scrub land, water bodies, etc.). Moreover, the spatial coordinates for problematic areas where the class needed further clarifications were identified (fig. 2).
Fig. 2: a) RapidEye natural color image, b) the result of unsupervised classification for the same area
The field surveys (groundtruthing) took place during the summer months of 2014. The following issues were resolved during the groundtruthing: (1) distinguishing between pasture/grasslands and open spaces; (2) distinguishing between pasture/grasslands and arable lands; (3) distinguishing between forests and shrubs; (4) distinguishing between permanent crops and heterogeneous agricultural areas;
(5) distinguishing between mining areas and open spaces; (6) identifying mining and industrial areas and (7) identifying heterogeneous agricultural areas.
Total 540 waypoints/tracks were taken during the groundtruthing activity in the Syunik province of Armenia, out of which 244 - in the Vorotan, 129 - in the Voghji and 167 - in the Meghriget river basin. Fig. 3 shows spatial distribution of the way-points within the Vorotan and Voghji river basins.
Fig. 3. Waypoints/tracks taken during groundtruthing in the a) Vorotan and b) Voghji
river basins
Fig. 4. Sample results of supervised classification into CORINE Level 2 classes
The performed supervised classification was based on the outcomes of the groundtruthing activity. The results include updated land cover/use maps for the Vo-rotan, Voghji and Meghriget river basins, as well as precise computed areas for each land cover/use class in square meters or hectares. The latter served for updating information on land balance in the Southern Basin Management Area of Armenia (fig. 4).
REFERENCES
1. CORINE land cover. Technical guide. Luxembourg, 1994.
2. CLC2006 technical guidelines. EEA Technical Report No 17/2007. Copenhagen, 2007.
3. Stepwise Procedures for Land Cover/use classification using satellite imagery. Guideline document prepared by Tom S. Sheng, Computer Assisted Development, Inc., 2013.
© А. Пилоян, А. Геворгян, Д. Варданян, 2017
