| چکیده انگلیسی مقاله |
Remote sensing is referred to as a process of observing, monitoring and collecting information about the physical characteristics of a specific area, mostly obtained by satellites or special aircrafts. Such information has found application in environmental sciences, and their availability in recent years has increased, regarding spatial resolution. Due to their use, remote sensing products have become essential in the process known as image classification, i.e. mapping the land use/land cover. Depending on classification technique and using geographic information system (GIS), it is possible to create thematic maps for further spatial and statistical analyses, based on multi-channel mono or multitemporal satellite images. Land use information, obtained from maps may differ significantly, and later use of generated layers may lead to significant differences in further assessments in environmental sciences, as well as in decision making and measure application. The accuracy of the classified data (classes) depends primarily on the quality of satellite images, the appropriate method and technique for classification, as well as the user’s expertise in choosing reference data, whereby the error in the process of classification is a common factor. In this paper, the results of soil erosion assessment were analyzed, in the Zagrža river catchment in central Serbia, which is characterized by different land use classes. Important segment of the paper is given through C factor, obtained by three different methodologies: supervised pixel based and object-based classification of a monotemporal multi-channel Sentinel 2A satellite images (10 m resolution) and by using CORINE Land Cover database. Generated layers were transformed to C factor maps, which were used in soil erosion assessment by USLE model. After classification, an accuracy assessment was performed, with estimated errors of omission and commission. This process involves evaluation and comparison of the image classification to reference data that are assumed to be true. In addition, stratified estimation was applied, in order to determine the unbiased accuracy assessment. Supervised pixel-based classification and object-based classification gave the accuracy of 76.62 % and 52.54 %, respectively. On the other hand, unbiased accuracy gave the results of 74.93 % and 63.59 %, respectively. These estimations showed the difference in soil loss between 6.61 and 8.5 ton.ha-1yr-1. These results suggest that land use maps should be accompanied by the assessment of land use classification accuracy, which would include sample size, an error matrix, details of stratified estimator, and errors of omission and commission. Accuracy assessment represents an important uncertainty parameter of applied technique and a key point in decision making for further data processing and analysis. |
