Orthorectification of a large dataset of historical aerial images: procedure and precision assessment in an open source environment

The availability of data time series spanning a long period is crucial for landscape change analysis. A suitable dataset, both in terms of time span and information content, must be available for the use with a GIS. In Italy, one of the most important historical source of land cover analysis is the GAI (Gruppo Aereo Italiano) photogrammetric survey (”Volo GAI”) commissioned in 1954 by the Italian national mapping agency, Istituto Geografico Militare Italiano (IGMI). The survey covers the whole Italy, but so far only some Regions, namely Lombardia and Veneto, have carried out the image rectification and the successive analyses to map land cover and use. This work describes the process of image orthorectification of the Volo GAI images for the Province of Trento (Provincia Autonoma di Trento). Image orthorectification must be performed to transform the images in maps available for analysis. This procedure corrects the geometry according to the terrain surface described by a Digital Terrain Model (DTM) to create an image compatible with the cartographic projection in use. To this end, the orthorectification modules available in GRASS GIS have been used, with the advantage of using the same GIS environment which will be used for the landscape analysis. The dataset covering the whole Province contains almost 100 images, this paper presents the preliminary results of the orthorectification of a quarter of the images. A reduced dataset has been used to test the results obtained using different settings with respect to: digital image resolution, DTM resolution and number of Ground Control Points (GCPs) used for the external orientation. These preliminary tests show that for the average quality of the Volo GAI images scan resolution beyond 600 DPI and DTM resolution above 10m do not provide significant improvements for orthorectification images. The minimum number of GCPs to guarantee the requested accuracy can vary from image to image, depending on the image quality and recognizable features position, but it is usually in the 15-20 points range