The degree in which the "raster-vector" conversion affects the accuracy of final area estimations depends on the algorithm implemented in the software you were using. From experience, I know that the specific algorithms of ERDAS, IDRISI and GRASS have good performance (i.e., they don't introduce high errors). However, they smooth

the stepwise look (corresponding to a conversion with "zero errors") to produce geographic features with more realistic shapes. On the other hand, I recommend you to take into account all steps in producing the data sets you intend to compare. In addition, I would start with two questions: what is the accuracy of the land cover area estimation ? (it depends on the image spatial resolution); what is the accuracy of the land use map with respect to areas (it depends on the type of projection - does it preserve areas, angles or shapes ?- and, of course, on the scale).

Dear Ruxandra Vintila

The spatial resolution of the image is 20m (Spot-XS), while the land use map is produced from large scale aerial photographs (1/10000), and so the accuracy is less than 1m. The projection is Cassini (cylindrical projection), it is suitable for the study area and provides high area estimation accuracy.

In this case, the loss of accuracy from the "raster-vector" conversion does not contribute significantly to the difference in area estimation, regardless of the algorithm used. This difference is caused by the scales, namely 1:10,000 versus 1:50,000 (for the latter, some experts would say 1:100,000, see http://www.isprs.org/proceedings/XXXV/congress/comm2/papers/197.pdf, Table 2).

Ruxandra,

Thanks for your answer and the paper you sent.

Ahmed, another variable to take in count that could be affect the accuracy from the raster-vector conversion is the pixel size of your raster format. A grater pixel size less accuracy.