Land cover data bases are basically digital maps created in a Geographic Information System (GIS), often produced by photo-interpretation of images on the screen according to a land cover legend defined in advance, in which each class (or label) represents a land cover type (Carfagna and Marzialetti, 2008). During the photo-interpretation process, the photo-interpreter outlines polygons and can make mistakes concerning the borders of polygons as well as the land cover type. In order to guarantee a good quality land cover data base, the photo-interpretation process has to be controlled. Due to cost and time, quality control of the photo-interpretation can be performed only on the basis of a sample of polygons in the methodological framework of statistical inference. Thus, another much more experienced photo-interpreter (the controller) performs a quality control on a sample of polygons in order to test if some mistakes have been made by the photo-interpreter. The result of the two photo-interpretations is a contingency table named confusion matrix with the classes used in the photo-interpretation by the photo-interpreter (rows - i) and by the controller (columns - j). In Carfagna and Marzialetti (2007) we have proposed a quality control method for continuously improving the data base production process. We have proved that the proposed procedure has the advantages of adaptive sequential procedures and does not present their disadvantages. Our procedure is always more efficient than stratified sampling with proportional allocation and is also more efficient than the procedure proposed by Thompson and Seber (1996, pages 189-191), who suggested stratified random sampling in two or, more generally, k phases. As stated above, we have obtained very good results with the described adaptive sequential procedure (ASPRN), when estimating the percentage of area correctly photo-interpreted. However, a sample selection which optimizes the estimate of a parameter can generate very inefficient estimates for other ones. We have considered Cohen's Kappa and Weithed Cohen's Kappa and we have obtained good results
Quality improvement of land cover databases: a sequential approach via agreement measures
CARFAGNA, ELISABETTA;MARZIALETTI, JOHNNY
2008
Abstract
Land cover data bases are basically digital maps created in a Geographic Information System (GIS), often produced by photo-interpretation of images on the screen according to a land cover legend defined in advance, in which each class (or label) represents a land cover type (Carfagna and Marzialetti, 2008). During the photo-interpretation process, the photo-interpreter outlines polygons and can make mistakes concerning the borders of polygons as well as the land cover type. In order to guarantee a good quality land cover data base, the photo-interpretation process has to be controlled. Due to cost and time, quality control of the photo-interpretation can be performed only on the basis of a sample of polygons in the methodological framework of statistical inference. Thus, another much more experienced photo-interpreter (the controller) performs a quality control on a sample of polygons in order to test if some mistakes have been made by the photo-interpreter. The result of the two photo-interpretations is a contingency table named confusion matrix with the classes used in the photo-interpretation by the photo-interpreter (rows - i) and by the controller (columns - j). In Carfagna and Marzialetti (2007) we have proposed a quality control method for continuously improving the data base production process. We have proved that the proposed procedure has the advantages of adaptive sequential procedures and does not present their disadvantages. Our procedure is always more efficient than stratified sampling with proportional allocation and is also more efficient than the procedure proposed by Thompson and Seber (1996, pages 189-191), who suggested stratified random sampling in two or, more generally, k phases. As stated above, we have obtained very good results with the described adaptive sequential procedure (ASPRN), when estimating the percentage of area correctly photo-interpreted. However, a sample selection which optimizes the estimate of a parameter can generate very inefficient estimates for other ones. We have considered Cohen's Kappa and Weithed Cohen's Kappa and we have obtained good resultsI documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
