| Abstract | The declining productivity and loss of ecosystem condition of arid and semi-arid lands is a worldwide concern and a major problem in Australia. Ecosystem condition can be assessed with the help of satellite imagery to measure the loss of basic resources (leakiness) from these areas. Leakiness has been shown to depend on the amount, type and position of vegetation cover in the landscape. It is well established that image scale (the observation scale) strongly affects the detection of landscape patterns and that rescaling changes these observed patterns through change in the structure of image features. Determining the relationship between leakiness calculated from images at different scales may assist in comparing results from the newer satellites with data from older long-duration time-series satellites such as Landsat and MODIS.
This research investigated the effect of different image resolutions on the calculation of leakiness (CSIRO Leakiness Calculator) from a savannah grazing catchment in North Queensland, Australia. Temporally and spatially coincident images from SPOT, Landsat and MODIS satellites were analysed for 11 vegetation indices. These were used in the Leakiness Calculator (LC) to calculate catchment leakiness. Catchment and sub-catchments were defined from DEMs at scales matching the imagery. A high resolution DEM matching the SPOT resolution was extracted from an aerial photograph stereo model. The SRTM 1s DEM and the GEODATA 9s DEM were each rescaled to match the Landsat and MODIS image scales. Rescaling was by cubic convolution in ArcGIS and other image adjustments were done using ERDAS Imagine, SAGA and ERMapper software. Image structure was analysed by variogram analysis using FETEX 2 software in an ENVI IDL environment.
This study found that the amount of vegetation cover varied with the type of analysis method and the spatial resolution. There was no clear pattern of cover values, except that the 25m Ground Cover Index (GCI) had the highest values. The usual measure of catchment leakiness, Calculated Leakiness (Lcalc) was nominally higher at higher resolutions. This is because it is influenced by the number of cells in the analysis area. A new measure of leakiness, the Adjusted Average Leakiness (AAL) was formulated to be insensitive to cell number and to cell size.
AAL responded inversely to amount of vegetation cover for a given vegetation index but there was no consistent relationship between AAL and type of vegetation index. AAL from Perpendicular Distance Indices (PDI) correlated negatively with cover (as expected) but AAL from the Soil Adjusted Vegetation Index (SAVI) and the Normalised Difference Vegetation Index (NDVI) correlated positively with amount of cover (unexpected). Other vegetation indices had irregular correlations between amount of cover and AAL.
Leakiness scaling functions for calculating both types of leakiness between 10 – 250m resolutions were developed (Resolution Scalograms). Lcalc scalograms took the form of linear reciprocal squared relationships for leakiness from SAVI and the Stress Related Vegetation Index (STVI) and a cubic reciprocal squared relationship for leakiness from the Perpendicular Distance of red-over-green band index (PDrg). AAL scalograms were simpler and took the form of simple linear relationships for leakiness from SAVI and STVI, but cubic for leakiness from PDrg. The high correlation between sill variance and resolution allowed the development of Variance Leakiness Scalograms (VLS). VLS for SAVI and STVI were positive logarithmic relationships and the PDrg VLS was a positive linear relationship.
Analysis of the structure (variance) of observation scale images of the catchment showed they had bounded natural logarithmic variograms. This structure decayed with progressive upscaling. Both observation scale and upscaled images had higher variances at lower a resolution. This is substantially different from previously reported findings. Three-dimensional (3D) models of the variance surfaces showed the effect of upscaling on image structure for different vegetation indices. The PDrg image variance response was the most complex. These models identified the optimal image resolution at which SAVI, STVI and PDrg features are expressed. Correlation between leakiness and conventional variogram indices and indices developed by the Universidad Politecnica de Valencia (UPV) was used to analyse for relationships between image structure and resolution. DEM variograms behaved differently. They had unbounded quadratic variograms and retained their form when upscaled.
The effect of vegetation cover in different areas of the catchment was tested by increasing SAVI and PDrg vegetation cover at different locations relative to major catchment features such as streamlines, elevation, slope, aspect, topographic feature and amount of pre-existing cover. Leakiness decreased the most when cover was added to zones distant from streams, at higher elevations, on lower slopes, on the crest of rises, on the top of ridge lines and in areas with the lowest amount of pre-existing cover. It is acknowledged that these findings are not entirely consistent with each other. There is mixed support for them in the literature. Smaller amounts of cover reduced leakiness more per unit of added cover than larger amounts of cover in all situations. |
|---|