EVALUATION OF THE POSITIONAL ACCURACY OF RAPIDEYE ORTHOIMAGERY BASED ON GEOREFERENCED RURAL PROPERTIES IN BRAZIL
ISBN 978-85-88783-11-9
Authors
1Ferreira, R.; 2Soares, R.; 3Maranhao, M.; 4Oliveira, L.
1IBGE Email: rafael.damiati@ibge.gov.br
2IBGE Email: renan.soares@ibge.gov.br
3IBGE Email: marcelo.maranhao@ibge.gov.br
4IBGE Email: leila.oliveira@ibge.gov.br
Abstract
An orthorectified image is one that has been corrected for the distortions due to imaging geometry and topographic relief, and has the properties of a planimetric map. Like any cartographic product, orthoimages are also uncertain with respect to the reality they represent. The verification of the positional accuracy of images is performed trough the establishment of control points in the field that are also easily recognized in the image, followed by statistical analysis of the differences between these points. Thus, it is possible to classify these images according to their cartographic accuracy standards. A batch of RapidEye orthoimages (year of 2011) bought by the Brazilian government was previously evaluated for positional accuracy, indicating a critical area located in the central part of the country. A concentration of high planimetric errors was observed in this region, exceeding the tolerance specified in the Brazilian cartographic accuracy standard for the 1:50,000 scale (25 meters of tolerance for higher class). The evaluation was performed using control points collected in the field (PTCON, which gathers information with excellent reliability), and orthophotos generated by IBGE (Brazillian Institute of Geography and Statistics). In an attempt to increase the number of control points and test a different checkpoint database a new test is proposed, using georeferenced rural properties polygons provided by INCRA (Brazilian National Institute for Colonization and Agrarian Reform) and distributed throughout the national territory. In theory, these data have good reliability in defining polygon vertices and thus can serve as checkpoints. The objective of this study is to evaluate the use of georeferenced vertices of rural properties as checkpoints to verify the positional accuracy of RapidEye images. A set of points was obtained directly in vector layers of INCRA’s georeferenced properties using QGIS Valmiera (2.2.0). The process consisted in the selection of vertices of INCRA’s polygons, which served as checkpoints for positional accuracy evaluation, and the corresponding features in RapidEye orthoimagery. To ensure that the selection was made exactly at the vertex, the snapping function was enabled. Finally, the statistical discrepancies of the sampled points were analyzed in the GeoPEC 3.0.1 software. A total of 318 checkpoints were collected spread across 58 image tiles. Although the number of sample points of INCRA (318 points) is three times higher than the selected PTCON samples (98 points), the results are statistically consistent in both databases. For PTCON, the average value was 40.78 meters, standard deviation was 9.08 meters and root mean square error was 41.99 meters. On the other hand, INCRA’s average value was 42.37 meters, standard deviation was 10.51 meters and root mean square error was 43.72 meters. Thus, the planimetric errors were in the same range regardless of the database used, and both exceeded the tolerance specified in the Brazilian cartographic accuracy standard. These results indicate that the georeferenced rural properties database provided by INCRA can be trusted to assess the positional accuracy of images, considering the similarity of the results to a reliable method. For the sample set used, the values of average, standard deviation and root mean square error of the discrepancies were compatible both with the results based on INCRA’s polygons and tests previously conducted with PTCON’s field points. The method of collecting checkpoints tested here stands out for its use of data already produced by the private sector, thus not generating expenses in field missions. Furthermore, the use of INCRA’s georeferenced polygons increases the number of control points, improving the possibility of collecting points in regions where reference data are scarce. Finally, new tests must be carried out in other regions to evaluate the behavior of the results with different reference data.