Introduction
River basin management continues to challenge humanity worldwide in
finding a balance between anthropogenic hydrological water needs and
those of the environment (Hoekstra et al. , 2012; Wheeler et
al. , 2017; Bouckaert et al. , 2018). Both surface water and
groundwater resources continue to diminish via over abstraction and
declining rainfall across many countries, resulting in poor riverine
ecological function and environmental condition (Vörösmarty et
al. , 2010; Rolls and Bond, 2017; Bouckaert et al. , 2018). To
arrest environmental decline in Australia’s food-bowl, the
Murray-Darling Basin (MDB; Figure 1), the Murray-Darling Basin Plan was
developed to ensure sustainable water use while supporting reliant
industries and the environment (MDBA, 2009).
Recognising the MDB will remain a highly managed system, central to the
Murray-Darling Basin Plan is water for the environment to support and
where possible, restore ecological condition to achieve long-term
environmental outcomes in the absence of natural flows (MDBA, 2009).
However, prioritising when and where environmental water is required
across the MDB is extremely complex and to a large extent, relies onin-situ monitoring of ecological assets such as fish, birds and
both woody and non-woody vegetation. As the MDB covers
~14% of Australia’s land area (or 1,000,000
km2), reliance on in-situ monitoring is not
only costly, it impedes the ability to understand large-scale ecological
water requirements and ecological responses. Remote sensing methods,
however, can provide robust broad scale monitoring options, underpinned
by in-situ observations.
Over the last decade, it has become evident that field measurement of
water loss from floodplain woody vegetation via transpiration and
evapotranspiration (ET), provides a way to observe forest or woodland
ecological condition and water needs (Doody et al. , 2015; Jarchowet al. , 2017; Wallace et al ., 2019). When tree water
requirements are not met, both transpiration and ET decline in response
to tree water saving adaptations such as stomatal closure to prevent
xylem cavitation and tree death (Baird et al. , 2005; Doodyet al. , 2009, 2015). Severity of decline is related to extent of
continued water deficit/drought. Broadscale monitoring of woody
vegetation transpiration and ET during drought can highlight tree
condition decline trajectories and inform prioritisation of
environmental water over both short and longer timeframes. Similarly,
monitoring tree response to increasing water availability is equally
important in a water management context related to decisions around
where and when environmental water should be delivered.
Considerable investigation has demonstrated that remote sensing provides
a broadscale means to monitor woody vegetation water loss via ET
(Guerschman et al. , 2009; Glenn et al. , 2011; Mu et
al. , 2011). In particular, MODIS imagery, providing 8-day composite
greenness products (normalised difference vegetation index (NDVI) and
enhanced vegetation index (EVI), provides excellent temporal resolution
from the year 2000 from which to develop vegetation monitoring solutions
(Glenn et al. , 2011; Nagler et al. , 2016).
Evapotranspiration algorithms, such as those developed by Mu (2013),
Guerschman (2009) and Nagler (2016), lend themselves to broadscale
ecological monitoring. However, the MODIS spatial scale of 250 m can be
limiting when it comes to calibrating remote sensing outputs within-situ data collected over smaller scales, as often happens. For
example, in the MDB, collection of woody vegetation ET typically occurs
using 50 x 50 m plots, with highly heterogenous canopy cover (Doodyet al. , 2015).
To augment development of temporal fine scale remote sensing monitoring
methods which are applicable to ecological vegetation systems worldwide,
a method to downscale low spatial resolution ET estimates based on high
spatial resolution FTCC maps requires development. Improved downscaling
will reduce the risk of over or underestimating vegetation ET during the
field calibration process and improve large-scale remote sensing ET
estimates. While several fractional vegetation cover products exist in
Australia (Guerschman et al. , 2015; Guerschman and Hill, 2018;
DEE, 2019), the information they impart is not suitable to downscale ET,
due to lack of fine scale classification of vegetation cover or
discrimination between groundcover vegetation and trees.
The objective of this study, therefore, was to develop a method to
determine fractional tree canopy cover (FTCC) at a pixel resolution of
20 m. The intent was to provide a fine scale evaluation of the
proportion of tree canopy in each MODIS pixel and field measurement
area. Understanding the amount of remote sensing ET related directly to
tree ET (from field measurement) allows improved calibration accuracy
between the two. This in turn, improves broadscale ET estimates used for
monitoring, by accounting for heterogenous canopy cover in each pixel.
The innovative method reported, combines radar (Sentinel-1) and
multi-spectral (Sentinel-2) imagery for the first time, to estimate
FTCC. Imagery based on LiDAR (Light Detection and Ranging) data was
employed as a surrogate for field canopy cover and used to train and
build a model that can estimate FTCC across large areas of the MDB.
Given the paucity of methods to identify and map canopy cover at fine
scales, the research presented within is likely to be important to many
aspects of environmental management and hydrology, specifically
catchment water management and improving our understanding of the
underlying hydrological processes related to vegetation presence and
absence.