Geodeziya, kartografiya, geografiya

Do I need to visit the area I'm mapping?
Although land cover maps are often made without visiting the field, there are good reasons why field visits should be made. The two primary reasons for visiting the area that is being mapped are to collect data that can be used to train the algorithm or the interpreter and to collect data that can be used to evaluate the land cover map and estimate the accuracy of the individual classes (a process called validation). At a minimum, these data can be collected in one trip but often two or more trips are preferred so that validation information can be systematically collected using a sampling design based on the classification results.
Data collected in the field must be georeferenced so that the point where the data were collected can be located on the imagery. GPS receivers are commonly used to record this location information. The type of information collected can range from detailed notes describing a site to a photograph of the site. Some of the detailed information that can be recorded includes: type of vegetation, crown closure slope, aspect, soil type, and other bio-physical characteristics that are important to identify the land cover type. If photographs are taken it is a good idea to record the direction the camera was pointed and to make notes about the area to supplement the content in the photograph. For example, you could add information about species composition, tree height, and possibly land use.
When land cover maps are created without using field data from the region of interest it is difficult to predict the accuracy of the final land cover map. An analyst with significant experience may be able to produce a land cover map of high quality but without validation information the true accuracy of the image classification quality is not known.
How do I know if my map is any good?
In the needs assessment some thought should be given to the required accuracy of the final map. The accuracy can refer to either spatial accuracy or thematic accuracy. Spatial accuracy is directly related to the base information that is used. If you are using satellite imagery the spatial accuracy of the final map will be dictated by the spatial accuracy of the satellite image that was used to create the map. Thematic accuracy specifies how well individual mapped classes match what is on the ground. For example, if the map indicates that there is forest at a particular location, what is the actual land cover type? Error is inherent to the mapping process and determining an acceptable level of error requires careful thought. Factors affecting accuracy include: quality and type of the base data, experience of the analyst, the analyst's familiarity with the area being classified, and the level of detail of the classification scheme. The accuracy of a land cover map can be assessed using statistical sampling procedures that are outlined in many remote sensing textbooks. The basic idea is to select a number of sample sites from each of the cover types in the final image and then go into the field to see what type of cover is actually on the ground. This information is then compiled in a contingency table (Table 2) so that the accuracy of each class can be determined.
The sampling statistics for this process can be fairly straightforward but the practical issues such as limited access to the study area, insufficient funds to visit all of the sites, and a lack of time tend to impose limitations that must be accommodated in the sampling design. Dealing with this less than ideal situation may necessitate some creative solutions that require an understanding of statistics beyond the expertise of the image analyst. When this is the case it is good to have the sampling design reviewed by someone with a solid understanding of statistics. More information about accuracy assessment can be found in the Overview of accuracy assessment of land cover products guide.