Terrain illumination correction is an important step in the normalisation of remotely sensed data for the inversion of land surface parameters, and for applications that aim to detect land surface change through time series analysis. An appropriate resolution of the Digital Elevation Model (DEM) data with sufficient quality is critical for effective correction of remotely sensed data over mountainous areas. Conversely, using terrain illumination correction and scale-based analysis, such as filter bank analysis, the quality of DEM data can be evaluated relative to the scale of the target data. In this study, TanDEM-X Intermediate DEM (IDEM) data at 12 m and 30 m resolutions, and the 1-arc second Shuttle Radar Topography Mission (SRTM) data (~ 30 m resolution) were used to evaluate their absolute and relative effectiveness in terrain illumination correction for Landsat satellite optical data.

The island of Tasmania in Australia has significant local terrain detail as well as high regional relief. This, together with its high latitude and wide variation in terrain illumination throughout the year, makes it an ideal study site to test correction methods and assess different resolution candidate DEM data. A set of images from both Landsat 7 and 8 multispectral bands (MS) and panchromatic (Pan) band were collected. These images were put through standard atmospheric and BRDF processing as well as terrain illumination correction using different sources of DEM. Comparisons were made by undertaking terrain correction with the various input DEMs and using visual and difference image methods to evaluate them. Quantitative Filter Bank analysis is also used to evaluate performance as a function of spatial scale. In total, five DEMs were used for the study, which include three at Landsat MS scale (derived from 12m and 30m IDEM and 1 sec SRTM data) and two at Landsat Pan scale (derived from 12m IDEM and 1 sec SRTM data).

Results from the terrain illumination correction and filter bank analysis show that (provided the analysis is confined to areas without some specific data issues), although all data sets can carry out the task well, the IDEM 12 m resolution based datasets can resolve finer details of terrain shading than the SRTM based DEM. This indicates that IDEM can deliver better results in areas with detail-rich terrain monitored with Landsat data. However, since the data available for this study is a sample from an intermediate product, spikes and other noise artefacts not expected in operational data were prevalent. Noise artefacts occurred especially over areas covered by water. Operational use of the IDEM will require the removal of such noise artefacts, but from the present study we can say if that were fully achieved, the 12m resolution data could form the basis for better terrain correction of Landsat data. The filter bank analysis also found that both Landsat panchromatic data and IDEM 12 m data seem to be over-sampled; studies using even finer DEM data would be required to examine this further.

It should be possible to correct for noise issues, as similar processing was carried out with SRTM data at an early stage to good effect. More detailed evaluation of the relative merits of the TanDEM-X based DEM compared with the SRTM based DEM data for terrain illumination correction may be possible when the WorldDEM product based on TanDEM-X data becomes routinely available with the water areas noise issues resolved.