Geometrical Accuracy Assessment of ALOS PRISM Products
Carlos Perez-Gutierrez(2), Nilda Sanchez(1), Jose Martinez-Fernandez(1), Laura Cejuela-Broncano(2) and Jesus Alvarez-Mozos(3)
(1) University of Salamanca, C/ Río Duero, 12. Parque Científico, 37185 Villamayor (Salamanca), Spain
(2) University of Salamanca, Avda. Hornos Caleros, 50, 05003 Avila, Spain
(3) Public University of Navarre, Los Tejos, Arrosadia s/n, 31006 Pamplona, Spain
This work uses ALOS data to create topographical cartography and to evaluate its global accuracy as an alternative to traditional photogrammetric products.
Some PRISM triplets from catalogue were acquired. Collinearity techniques from analytical photogrammetry adapted to pushbroom technology were used for the calibration process. Digital Surface Model and Orthoimagery products were created.
Ground Control Points were collected from 1:1000 cadastrial cartography and a bundle adjustment orientation for the altogether CCDs was solved. Tie Points were used to fix the different views (forward, nadir and backward). The accuracy assessment using Independent Check Point reports a very good geometrical accuracy for orthoimagery. No radiometrical balance was necessary for merging the different CCDs into the same acquisition but some artifacts were detected between different images in the same triplet.
In general, metadata for geometrical correction are not confiable. Nadir acquisition is preferred for orthoproduction because of a better detail. Some resamples are being analyzed merging the different scenes into the triplet as an enhanced fusion. No final results have been obtained yet but we the study is on going.
Several Digital Surface Models were built using an improved approach where a triplet is more confiable than a stereoscopy pair. Several analyses were carried out using forward, backward and nadir views in stereopair set and triplet set. Correlation based triplet stereo images matching let to build the digital elevation surface and to control the quality itself.
The comparison with the groundtruth surface yielded a standard desviation close to 5 meter in elevation (double of the pixel’s footprint). Digital Surface Models created without Ground Control Points were rejected because matching was impossible to fix. Position metadata were found not adequate for a fine detail.