Estimating RUSLE C-Factor values for Great Barrier Reef Catchments using satellite derived ground cover estimates

Abstract

Ground cover is a primary contributing factor in preventing hillslope erosion. Accurate modelling of this erosion is therefore dependent on accurate estimates of ground cover. Ground cover is temporally and spatially variable and, as such, remote sensing is an ideal source of ground cover estimates over large geographic areas. For the Great Barrier Reef, this information has generally been derived from the Remote Sensing Centre (RSC) (within the Queensland Department of Science, Information Technology, Innovation and the Arts) Bare Ground Index/Ground Cover Index (BGI/GCI). The BGI/GCI has been readily adopted by the modelling community as a preferred source of ground cover information at both the paddock and catchment scale. While the BGI/GCI has been well received by the modelling community, it has been reported to overestimate ground cover by various authors, when compared with visual assessments. The BGI/GCI is calibrated and validated against point intercept ground cover data. Visual assessments of ground cover have been shown to estimate less ground cover, particularly in the middle ranges, when compared with point intercept/BGI/GCI estimates. In general point intercept methods are regarded as more reliable than visual estimates. Typically, modelling frameworks utilise ground cover data using the RUSLE model. In Australia the RUSLE C-factor is typically determined using as: from visual assessments, it is reasonable to expect an underestimate of the RUSLE C-factor when using satellite based estimates. Therefore, to obtain accurate C-factor estimates from satellite data, it may be necessary to adjust for the source of the ground cover data. In addition, the BGI/GCI has recently been replaced by the Fractional Cover Index (fCI). It is not known what the effect the fCI will have on ground cover estimates. This paper looks to investigate both the performance of the new fCI product on ground cover and C-factor estimates and explore the possibility of adjusting C-factor estimates to account for source data that is obtained from satellite, rather than visual field estimates. Two long-term average dry-season bare ground indexes were compared (BGI/GCI and fCI). Both products were identically masked for cloud, cloud shadow, water and foliage projective cover (FPC) greater than 15%. Total ground cover rasters for each catchment were calculated by taking the inverse of the bare fraction for each raster and were clipped to catchment boundaries. Difference rasters for both total ground cover and derived C-factors were calculated. Additionally, satellite ground cover estimates were adjusted to an equivalent visual estimate by a non-linear conversion function. The adjusted visual estimates were then converted into C-factors (adjusted-fCI). The density distributions and medians for total ground cover, C- factors and adjusted C-factors for both indexes were also calculated. There was very little predicted difference in erosion predictions between the new fCI product and the BGI/GCI. A slightly lower median ground cover value for the fCI did not translate into any appreciable difference in the RUSLE C-factor. The median C-factors for both indexes were typically low for all catchments. The lack of change in the C-factor is attributed to the high levels of ground cover predicted for both indexes, for all catchments. Changes at high levels of ground cover have minimal effect with the Rosewell conversion function. In comparison, the distribution of differences between the BGI/GCI and the adjusted-fCI, shows considerable areas of very high decreases in ground cover and large increases in C-factors. For example, the effect of this adjustment in the Fitzroy was to reduce median ground cover from 75% (BGI/GCI) to 52% (adjusted-fCI) and increase the median C-factor from 0.01 to 0.05. The spatial distribution of these increases also varies, which in itself would affect modelling outcomes. All C-factor estimates were reasonable for rangelands. The conversion function between satellite and visual estimates relies on an accurate determination of the relationship between visual and objective estimates and there is limited research done in this area. Given that satellite estimates will certainly remain the primary source of ground cover data for modelling purposes, developing a new conversion function with this understanding should be a priority. Further research is required and suggested for determining a more appropriate conversion function.