A method of highly accurate geometric correction for MODIS images based on SRTM3.

The features of terrain and ground surface are highly integrated and there are only some characteristic features with a super or large scale on the ground. Therefore, the potential obvious ground features that can be recognized and chosen as GCP are very limited. It is the biggest problem for lower and moderate resolution images to handle highly accurate geometric correction. In this paper,we present a method that creates the georeferencing data sets by using SRTM3 for GCP collecting for geometric correction,which can keep scale of the geometric elements consistent between satellite images and DEM. Firstly,the terrain synthesis needs to be done from SRTM3 with 90 m resolution to new DEM with 250 m pixels. Layers of terrain features extracted from 250 m DEM and the gorge features are merged into data sets of the borders of large lakes and big rivers from the 1:25 million map. And then,the visual negative relief color maps with 250 m are created by the terrain illumination and lighting model (Hillshade). Thus, the georeferencing data sets that include three kinds of geographic information elements are formed. Then multispectral bands from MODIS data sets with 250 m and 500 m resolution are selected and composited as pseudo color images with sub-pixel by decomposition and enhancement technology. Finally,maps of the georeferencing data sets and pseudo color images are overlaid and displayed on the same window for GCP collection by manual and visual comparison. All MODIS bands are subjected to highly accurate geometric correction by using a model of adaptive triangulation nets. As an example, it has 431 m of average square error for all GCP set. The results show that the geometric dimensions of 250 m of DEM are similar to features from MODIS sub-pixel images. The features that can be recognized and located on MODIS images will be found out and identified on the georeferencing data sets from SRTM3 or on the big scale maps of hydrology. The numbers of land features in the georeferencing data sets are more than in MODIS images. Compared with the traditional method that selects GCP from base maps, the method we propose is more suitable to interpret,identify and capture the same objects. And the operation and data processing is easy. It has higher accuracy of coordinates for capturing GCP and high working efficiency.