The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part BI. Beijing 2008 
1320 
than parcels boundaries. We called it the ligneous 
overestimation effect. 
Figure 3. Example of area measurement on orthophoto and 
Cartosatl images 
4.2.3 Reproducibility and critical difference 
The final results of buffer determination are presented in table 4. 
The repeatability limits gave the maximum difference between 
two operators on the same image at 95% confidence, considered 
as tolerance in this experiment. For orthophoto, this value was 
equal to 2.86m and was equal to 5.17m and 8.76m respectively 
for Cartosat-1 Aft and Fore. Even if Fore had higher value of 
tolerance than Aft, it didn’t signify that Fore was worse than 
Aft. Effectively, Fore having been processed before Aft, we 
could assume that a “training” effect would have influenced the 
operators’ recognition and memorisation capacities and limit 
the relevance of the comparison. 
The critical difference (CD) value gave us the maximum 
difference between the reference area and the measured areas 
(again at 95% confidence). The lowest value was equal to 
0.96m and was obtained from orthophoto. On Cartosat-1, CD 
was equal to 1.65m and 3.21m respectively with Aft and Fore. 
Regarding the two last results, Cartosat-1 images were both less 
accurate than orthophoto; and we consequently do not 
recommend using Cartosat-1 images as the main tool to perform 
CwRS under the actual CAP regulatory framework. 
To illustrate the major limitations observed during this survey, 
figure 3 presents problems met with parcel border identification. 
On the orthophoto, long shadows representing vegetal hedges, 
which should not be considered as parcel boundaries, are visible; 
however, these shadows have been regularly considered as 
boundaries by operators in Cartosat-1 images and not included 
in the parcel area measurement. In addition, changes in texture 
for Cartosat-1 provoked the disappearance of narrow paths and 
resulted in overestimation. 
5. FINAL DISCUSSION 
Compared to the orthophoto, the majority of parcels were 
correctly identified on both Cartosat-1 images; only 62 
observations out of 3330 were found to be outliers. 
Nevertheless, the main problem with parcel area measurement 
was the correct border identification due to a loss of 
information as a result of shadows, small and narrow objects, 
and texture changes. Overall, changes resulted in a less accurate 
delineation of the parcel boundaries and very often to an 
overestimation of the parcel area. With regard to reproducibility 
limits and critical difference, neither of the Cartosat-1 images 
tested can be considered as a primary solution for Control with 
Remote Sensing in accordance with European CAP 
requirements. 
Comparison of images and evaluation of factors highlighted the 
need to consider the CAPI process as organised around a 
tripartite relationship between (i) global image quality, (ii) 
operator’s recognition capacity and (iii) operator’s object 
memorisation. As proposed in figure 4, these three components 
should be considered as integrated and dependent inside the 
CAPI system. Image interpretation and agricultural parcel 
boundary delineation appeared closely related to the operator’s 
personal experience. CAPI experience means the capacity to 
recognise an object whatever the source of the information (the 
image) and the capacity to compare this information to a pool 
of personal reference obtained from regular training and/or 
previous experiences. Both initially depend on the global 
quality of the image; quality can be perceived as the 
effectiveness of the image to provide to the operator properties 
of each single object (e.g. precision) as well as difference 
between objects (e.g. contrast). The fact that “visibility” was 
one main problem when identifying parcel boundaries may 
confirm this assumption. Further, the operator’s interpretation 
was often biased due to intrinsic parcel characteristics such as 
shape, size or land cover. This suggests that despite using 
images of high quality, CAPI efficiency remains dependent on 
the operator’s references and his adaptability. To perform CAPI 
efficiently, any contractor responsible for CwRS should use a 
variety of the image sources when training its staff and 
regularly test the accuracy (deviation) of each individual 
involved within the process. By evaluating the relation between 
operator and images, land cover, physical characteristics of 
parcels, the contractor would efficiently assess the quality of 
the whole workflow, excel in measuring agricultural parcel area 
and quickly meet the diverse CwRS regulatory requirements. 
Before that, several potential effects identified during this 
survey should be apprehended: the magnification or 
compensation effect, the smoothing effect and the ligneous 
over-estimation effect.