Sequential Design in Quality Control and Validation of Land Cover Data Bases

The scale of remote sensing data used in the photo-interpretation represents only the level of detail of the basic material and cannot be considered as the quality of the land cover data base. The quality should be considered from two viewpoints: the quality of the photo-interpretation and the level of agreement between the data base and reality (validation). Due to cost and time, the quality control of the photo-interpretation as well as the validation can be performed only on the basis of a sample of polygons in the methodological framework of statistical inference. Quality control should be made by repeating the production process with the same basic material and the same procedure, for evaluating the quality of the product. In our case, a very expert photo-interpreter repeats the photo-interpretation on a sample of polygons. Validation is a comparison of the land cover data base with another representation of reality which is considered more reliable. We compare a sample of polygons with the corresponding ground truth, in case the scale of remote sensing data is compatible with the ground truth; otherwise, the comparison is made with other remote sensing data with more detailed scale. Generally, in photo-interpretation projects, a very small amount of resources is devoted to quality control; thus, a very cost effective sample design should be adopted. Moreover, since we want to use quality control for continuously improving the data base production process, the quality control has to be done in a very short time, during the photo-interpretation process. In this paper, we propose to adopt a sequential sample design for quality control as well as for validation, because sequential sample designs generally allow reaching high precision of estimates with the smallest sample size and in the shortest time, particularly when a small amount of information is available. Finally, the desire of homogeneity often suggests using the same kind of satellite data and the same legend adopted in other projects, although data and legend can be inappropriate to the specific area or can produce a data base which does not fit the client’s needs. A sequential sample design during the photo-interpretation process allows quality control and validation, for improving the legend adopted and for evaluating the correspondence of the data base to reality, in order to satisfy the client’s needs.