Increasing the accuracy of tree height estimation in a tropical forest from Landsat ETM+ data through using super-resolution analysis

Authors: Boyd, D., Anuar, M., Foody, G.M., Hill, R. and Hopkinson, C.

Conference: ForestSAT 2010: Operational Tools in Forestry Using Remote Sensing Techniques

Dates: 7-9 September 2010

Abstract:

The focus of study was a field site situated along the Rio Tambopata in the Department of Madre de Dios in south east Peru. In this region, the foothills of the Andes meet the western Amazon Basin creating a hotspot of biodiversity (Myers et al. 2000). The biological significance of the forests in this region is reflected in their status as one of the most protected regions in Amazonia. The forests of this area are predominantly undisturbed and mature. At this site airborne LiDAR data were acquired on June 7, 2009 using a Leica ALS50 II system flown at an average altitude of 1600 m. This system has a 200 kHz repetition rate and uses a 1064 nm wavelength laser. The along/across track sampling rate was one hit per m2, resulting in an average density of 2.1 laser hits per m2 on the ground with flightline overlap. The nominal footprint size was 24 cm. The Leica ALS50 II system records multiple (i.e. up to four) returns per laser pulse. These were separated into ground and non-ground (i.e. vegetation) hits based on the algorithm of Axelsson (2000) which is implemented in TerraScan software (Terrasolid, Finland). The ground returns were subsequently interpolated into a Digital Terrain Model (DTM) at 2m spatial resolution, applying an Inverse Distance Weighting algorithm. A Digital Surface Model (DSM) was created of the maximum return height per 2m grid cell and this was normalised using the DTM to create a Canopy Height Model (CHM). These modelled canopy height data were related to a Landsat ETM+ image acquired on July 28, 2009 and downloaded from the USGS Earth Explorer website. The ETM+ data were processed using a super-resolution technique, namely bicubic resampling, resulting in an image with a spatial resolution of 2m (i.e., to match that of the lidar-derived CHM). Correlations between the spatially enhanced ETM+ data and the corresponding canopy heights were computed for an extract of the data covering a range of canopy heights. All these correlations are significant and by referring to the transect plots shown in Figure 1 it is evident that more improved information on forest height can be retrieved from the Landsat ETM+ data. These show an improved relationship between ETM+ data and canopy height in some wavebands and not others over those obtained using the original spatial resolution ETM+ image. Further investigation is required to fully understand why this is the case and whether these super –resolution techniques can be used reliably in this way to overcome errors associated with spatial sampling conducted by Earth observation sensors. This paper demonstrates the potential of using super-resolution analysis for enhancing the information content of medium resolution imagery, in this case for estimation of tropical forest height.

Source: Manual

Preferred by: Ross Hill