Naslov
Rule-based object-oriented land cover classification of RapidEye multispectral satellite images for dasymetric mapping

Autori
Valožić, Luka ; Cvitanović, Marin

Vrsta, podvrsta i kategorija rada
Sažeci sa skupova, sažetak, znanstveni

Izvornik
36th EARSeL Symposium “Frontiers in Earth Observation” 20-24 June, 2016 Bonn, Germany Programme & Abstract Book / Menz, Gunter ; reve, Klaus ; Balthesen, Oliver ; Rienow, Andreas ; Dubovyk, Olena ; Thonfeld, Frank - Bon : Rheinische Friedrich-Wilhelms - Universität Bonn, 2016, 237-237

Skup
36th EARSeL Symposium “Frontiers in Earth Observation”

Mjesto i datum
Bonn, Njemačka, 20.06.2016. - 24.06.2016

Vrsta sudjelovanja
Poster

Vrsta recenzije
Međunarodna recenzija

Ključne riječi
land cover classification; object oriented analysis; ruleset; dasymetric map; Zagreb

Sažetak
Land cover data obtained by means of remote sensing were used for the production of a thematic map of the population density of the City of Zagreb in Croatia. A binary method of dasymetric mapping was selected for this task. An object-oriented land cover classification was conducted on RapidEye multispectral satellite imagery for the entire administrative area of the City of Zagreb with the total area of 641 sq.km. Trimble’s eCognition 9.0 and ESRI’s ArcGIS 10 computer software have been utilized for the research. Minimal pre-processing of satellite imagery was performed in order to avoid any unnecessary data transformation. Coordinate systems had to be transformed for the five RapidEye satellite image tiles used in research, and had to be co-registered as well to match the vector data layers of administrative units and reference images in GIS software. Rule-based method used for the land cover classification was based on range of statistical data on image segments, arithmetic features, vegetation indices, datasets acquired by principle component analysis, and spatial relations of image segments. Ruleset applied for the classification comprised of threshold values for the selected features and has resulted with crisp classification of image segments. Ridd’s V-I-S (vegetation–impervious surface-soil) model was used as a reference for land cover classification. Dataset created by land cover classification in this research consists of four final classes: vegetation, impervious or built-up surfaces, bare soil, and water. Such classification allows further geographical research of apparent dichotomies of the urban versus rural, or artificial versus natural surfaces. Classification accuracy was assessed by using methods of error matrix and kappa statistic. Aerial imagery and other satellite imagery of higher spatial resolution were used as reference data for the creation of the error matrix. Accuracy assessment resulted with the overall accuracy of classification at 92.06% and kappa statistic of 0.89. Data acquired by land cover classification was used for the production of dasymetric thematic map of population density and distribution for the City of Zagreb. Population data has been obtained from the Republic of Croatia’s Census of Population, Households and Dwellings 2011 at the administrative level of the Local committees. The City of Zagreb is comprised of 218 Local committees. RapidEye satellite images from 2011 were intentionally used for the land cover classification in order to optimally match remote sensing data with official census data. Contributions made by this research involve development of land cover classification ruleset, production of land cover map and land cover GIS layers for the area of the City of Zagreb, and production of dasymetric map.

Izvorni jezik
Engleski

Znanstvena područja
Informacijske i komunikacijske znanosti, Geografija

POVEZANOST RADA