Naslov
Accuracy Analysis of the Inland Waters Detection

Autori
Gudelj, Marina ; Gašparović, Mateo ; Zrinjski, Mladen

Vrsta, podvrsta i kategorija rada
Radovi u zbornicima skupova, cjeloviti rad (in extenso), znanstveni

Izvornik
Conference Proceedings, Volume 18, Issue 1.5. - 18th International Multidisciplinary Scientific GeoConference SGEM 2018 / International Multidisciplinary Scientific GeoConference SGEM - Sofija : Stef92 Technology, 2018, 203-210

ISBN
978-619-7408-72-0

Skup
18th International Multidisciplinary Scientific GeoConference (SGEM 2018)

Mjesto i datum
Beč, Austrija, 03.12.2018. - 06.12.2018

Vrsta sudjelovanja
Predavanje

Vrsta recenzije
Međunarodna recenzija

Ključne riječi
inland water monitoring ; inland water detect ; unsupervised classification ; supervised classification ; spectral indices

Sažetak
Climate changes and human activities on Earth's surface effect on natural Earth's resources as well as on inland waters. Any changes in the Earth's atmosphere, such as changes in temperature, humidity, precipitations and their intensity affect the inland waters area and level. In order to carry out the monitoring of the areas under the inland waters economically, the satellite imageries collected by remote sensing sensors are used. Satellite imagery used for this research were cloud-free Landsat-8 imagery at study area Zagreb. The classification of satellite imagery gives an insight into the state of the Earth's surface at a given moment. Today, many classification methods of satellite imagery have been developed, but in order to find the most accurate classification method for the extraction of inland waters, this research compare the accuracy assessment of the classification methods. For purposes of this research, the satellite imageries are classified into two classes, inland waters and others. Firstly, the DOS1 atmospheric correction was performed in QGIS software. Then study area Zagreb was classified using supervised classification methods Maximum Likelihood Classification (MLC) and Random Forests (RF). Both supervised classification methods were performed on six Landsat-8 30 m spatial resolution bands and same training polygons. For this research, normalised difference water index (NDWI), modified normalised difference water index (MNDWI) and automated water extraction index (AWEI) were created and used for classifying satellite imagery scene in two classes. Spectral indices, supervised classification, as well as unsupervised classification, were made in open software SAGA GIS. In order to compare all the methods of extraction inland waters, visual inspection and objective analysis were carried out. The objective analysis was performed by determining a figure of merit, overall agreement, omission and commission. By objective and subjective analysis, the best method for extraction of inland waters is the Random Forests method of supervised classification whose overall agreement 99.78%. The second method, whose accuracy is slightly lower than the RF method, is a method based on MNDWI and unsupervised classification whose overall agreement 99.71%. Accuracy assessment of the MLC method is lower than the previous two methods whose overall agreement 99.41%. The accuracy of methods based on NDWI and AWEI are worse. Overall agreement, for a method based on AWEI and k-means unsupervised classification, is 83.23%, while overall agreement for a method based on NDWI and k-means unsupervised classification is 79.05%.

Izvorni jezik
Engleski

Znanstvena područja
Geodezija

POVEZANOST RADA