Fine scale mapping of fractional tree canopy cover to support river
basin management
Abstract
Management of water, regionally, nationally and globally will continue
to be a priority and complex undertaking. In riverine systems, biotic
components like flora and fauna, play critical roles in filtering water
so it is available for human use and consumption. Preservation of
ecosystems and associated ecosystem functions is therefore vital. In
highly regulated large river basins, natural ecosystems are often
supported through provision of environmental flows. Flow delivery,
however, should be underpinned by rigorous monitoring to identify and
prioritise biotic water requirements. Broadscale monitoring solutions
are thus integral and for woody tree vegetation species, this is can be
via measurement of field evapotranspiration, regionally scaled using
remote sensing. However, as there is generally a mismatch between field
data collection area and remote sensing pixel size, new methods are
required to proportion tree evapotranspiration based on tree fractional
canopy area per pixel. Within, we present a novel method to derive tree
fractional canopy cover (FTCC) at 20 m resolution, in semi-arid and arid
floodplain areas. The method employs LiDAR as a canopy area field
measurement proxy (10 m resolution). Sentinel-1 and Sentinel-2, radar
and multispectral imagery, were used in Random forest analysis,
undertaken to develop a predictive FTCC model trained using LiDAR for
two regions in the Murray-Darling Basin. A predictor model, combing the
results of both regions, was able to explain between 85-91% of FTCC
variation when compared to LiDAR FTCC, output in 10% increments.
Development of this method underpins the advancement of woody vegetation
monitoring to inform environmental flow management in the Murray-Darling
Basin. The method and fine scale outputs will also be of value to other
catchment management concerns such as altered catchment water yields
related to bushfires and as such, has application to water management
worldwide.