Modelling the accuracy of image data reprojection

Abstract

Data transformation and reprojection are required for integrating remotely sensed data and geographical information system data into one map projection. At global or continental scales, the reprojection procedure may bring significant distortion of original pixel values. This research investigated a scale factor model to explain the amount of distortion. Specifically, model results were compared with experimental results. Six possible reprojections among the Equal-area Cylindrical projection, the Mollweide projection, and the Sinusoidal projection were tested. Results showed that reprojection accuracy could be explained using the ratios of scale factor changes along vertical and horizontal axes between source and target projections. The reprojection accuracy was the reciprocal of the maximum scale factor change either along the vertical or horizontal axis. However, the model accuracy was very sensitive to the skew effect that leads to significant increase of accuracy. Spatial autocorrelation and the number of unique pixels were found to affect the accuracy of reprojection. Considering the increasing use of global and continental datasets, this research suggests that reprojection of raster image data should be performed very carefully.