Although data integration can be very useful, there are also some requirements that have to be fulfilled for it to be effective:
• geospatial data have to be accurately co-registered in a common grid;
• time gaps between the various data layers have to be known and accounted for;
• systematic effects due to the atmosphere, the viewing angle, the Sun angle, etc., must be corrected for or taken into account.
In particular, if data from multiple sensor systems are integrated, one has to be aware of differences in their spectral sensitivities, wavelength bands, viewing angles, spatial resolutions, etc. Radiative transfer modelling can be applied to bridge the differences in spectral characteristics and viewing geometry of the various sensors. Other forms of modelling (e.g. 3D object modelling) are sometimes required to aid in the analysis of multi-angular data, for instance to differentiate true changes from apparent changes (e.g. shadows) due to a different viewing direction.
Data integration also comprises the incorporation of non-spatial information or point data from field measurements. These data have to be associated with precise moments in time and with precise geographic locations, or with some time interval and fuzzy-defined regions. Thus, here the important issue of the representativeness of this information for the associated time interval and geographic area comes into play.
In general, data integration forces us to consider the uncertainties or inaccuracies of the various data sources available. In some cases, meta-data may contain information about this. When integrating data for some purpose, one has to apply weights to each of them, so that the final result is a balanced compromise in which inaccurate data receive less weight than those with a high degree of certainty.