DEM Fusion using a modified k-means clustering algorithm

Digital elevation models (DEMs) are a necessary dataset for modelling the Earth’s surface; however, all DEMs contain error. Researchers can reduce this error using DEM fusion techniques since numerous DEMs can be available for a region. However, the use of a clustering algorithm in DEM fusion has not been previously reported. In this study a new DEM fusion algorithm based on a clustering approach that works on multiple DEMs to exploit consistency in the estimates as indicators of accuracy and precision is presented. The fusion approach includes slope and elevation thresholding, k-means clustering of the elevation estimates at each cell location, as well as filtering and smoothing of the fusion product. Corroboration of the input DEMs, and the products of each step of the fusion algorithm, with a higher accuracy reference DEM enabled a detailed analysis of the effectiveness of the DEM fusion algorithm. The main findings of the research were: the k-means clustering of the elevations reduced the precision which also impacted the overall accuracy of the estimates; the number of final cluster members and the standard deviation of elevations before clustering both had a strong relationship to the error in the k-means estimates.