The vertical accuracy of the collected data strongly depends on the characteristics of the terrain, image scale, illumination conditions, and average quality of ground control points. In order to assess the achieved accuracy of the Stromboli dataset, an analysis of the elevation residuals between the data acquired in 2003 and those produced from the 2001 survey, taken as a reference, was performed. The analysis was carried out for the area not involved in the landslide phenomena and lava flow. A visual inspection of residual maps allowed the detection of other small sectors showing persistent variations compared to the reference model, eventually caused by either surface deformations or by errors of the 2001 processing. Therefore, these areas were not included in the analysis. After a preliminary estimation of an overall standard deviation (ri3 m), it was decided to eliminate the data with residuals exceeding 10 m, thus confining the residual range within about three times the estimated r. The remaining data, about 95% of the total, are approximately normally distributed (Fig. 5), and were used for the estimation of the statistical parameters listed in Table 2. The observed differences in the height coordinates show mean values of less than 1 m and a standard deviation varying from 1.7 to 3.3 m; these values may be considered an upper limit for the accuracy due to the presence of shadows, vegetation, and rough terrain. As a matter of fact, the same analysis, repeated on an under formed area of the Sciara slope characterized by good illumination and visibility conditions, smooth terrain, and absence of vegetation, showed an improvement in the quality of the photogrammetric DEMs. The significant decrease of the standard deviations of the differences to less than 1 m is shown in Table 2(values in brackets). These new values are, as expected, greater than those of the absolute orientation residuals of the control points (Table 3).
A further, but not significant, reduction of the standard deviation and the correction of the systematic height difference are obtained by applying the least squares surface matching procedure. The estimatedrototranslation parameters are applied to the whole DEM before computing the surface variations.