Seamless merging of DEMs

Project Aims

This project aims to develop, investigate and compare methods for merging different resolution (and accuracy) Digital Elevation Models (DEMs) and identify the best implementable method to merge all available 1 to 5m resolution LiDAR datasets with coarser national DEMs (25m resolution). While the Murray Darling Basin (MDB) in south eastern Australia will be used as a case study, the project will develop generic methods and a workflow usable in other spatial applications.  

The boundaries between the high resolution and coarse DEMs, such as the SRTM DEM for Australia, can run for thousands of kilometres (Figure 1). The elevation differences between the DEMs (e.g., LiDAR and SRTM) can be substantial (>10 m) and non-systematic along the boundaries. Previously CSIRO has merged SRTM DEM data with LiDAR collected along the river channels (extent was limited due to prohibitive costs for collecting LiDAR for entire floodplains), for relatively small areas in Northern Australia (e.g. the Fitzroy and Roper catchments). However, techniques used over small areas are not necessarily appropriate for larger scale regions. The digital innovation in this project will be to develop, implement and compare methods for merging high and low resolution DEMs using the MDB as a case-study (1 million km² extent, 851 high resolution DEMs within this area). The project will address the problem of seamlessly merging multiple high resolution DEMs with a coarser base DEM (e.g. SRTM) to ensure the best possible DEM data are available as one dataset in a seamless form and that water flows correctly across internal DEM edges when the merged DEM is used in flood modelling.

Methods evaluated will include traditional merging techniques such as kriging and gaussian smoothing of elevation differences (Environment team members), as well as emerging machine learning techniques (Data61 team members). We will compare the outputs and select and implement the best method for merging various DEMs for on-ground applications. The work will enhance the capacity of CSIRO Environment staff who will develop new skills in the application of machine learning methods and use of the CSIRO Earth Analytics and Science Innovation platform for parallel processing.

The above image shows the composite input data on the left, the the initial output on the right.

For more information contact jenet.austin@csiro.au