The influence of remotely sensed thematic maps on landscape ecology studies
遥感主体图的准确度对景观生态学研究的影响

This paper systematically explains the nonlinear effects of the classification errors of remotely sensed data on the errors of landscape indices. The explanations are made mainly through hypothetical examples and case studies, including the global and regional land use data. On one hand, remote sensing technology meet the needs of landscape ecology by providing necessary land use and land cover data; on the other hand, remote sensing technology varies so sophistically that all the land use and land cover data derived from remotely sensed data are different. From users' points of view, there is almost no choice. Users simply use whatever is available with little knowledge about how bad or good the choice is. Both the hypothetical examples and case studies indicate that the variations of landscape indices are much greater than the variations the classification accuracy can explain. Under the existing levels of classification accuracy, the uncertainties or errors of landscape indices may be too high to help trigger sound findings or conclusions from landscape ecology studies. The error propagation processes become even more serious when change detections are performed with inaccurate land use and land cover data. It is undoubted that some past landscape ecology work must have made misleading conclusions due to the blind use of inaccurate land use and land cover data. This paper also explains the principles on how to correct the areas of individual land cover types. Up to date, nearly all the landscape indices cannot be assessed nor corrected. Almost the only thing that can practically be done is to try to increase the accuracy of remotely sensed thematic maps. The sample algorithms of image data classification provide good potential for increasing the accuracy of landscape indices because their classification units are defined in a similar way as patches are defined in landscape ecology. This paper introduces a case study that proves the superiority of the sample algorithms over pixel algorithms. It is clear that the increase of image data classification accuracy is necessary to obtain more reliable estimates of landscape indices but it is unclear about the required magnitude of the increase under various circumstances. This represents a new question for both landscape ecologists and remote sensing scientists.