DECLINING LANDSCAPE LEAKINESS INDICATES IMPROVED RANGELAND CONDITION

December 6, 2024

---

Gary Bastin, Adam Liedloff and John Ludwig. Email: gary-bastin@bigpond.com

 

Almost two decades ago we published a ‘leakiness index’ including demonstrating how the index suitably monitored change in landscape function at paddock-scale over time (Ludwig et al. 2007). The Leakiness Index (LI) is based on the potential for landscapes to lose (i.e. leak) litter, plant nutrients and their seeds in soil sediments as landscapes become dysfunctional through inappropriate grazing practices. Index values are calculated from remotely-sensed vegetation patchiness data and a digital elevation model (DEM) with LI being strongly linked to landscape function by algorithms that reflect the way in which spatial configuration of vegetation cover and terrain affect soil loss.

The index generated interest after publication with six direct requests for the LI Calculator and a recent literature search showing six papers that used the method and a further 143 citations (excluding ourselves) referencing the method. Apart from its application at two locations in north Queensland (Bastin et al. 2008, Dunwoody et al. 2013), we are not aware of its wider use in the Australian rangelands.

In this article, we further illustrate the application of LI by documenting decreasing leakiness over time in a central Australian grazed landscape where other remote sensing-based methods had previously shown similar improvement in rangeland condition (Bastin et al. 2023). Moreover, we show how LI values in years of below-average rainfall trended similarly to ‘ground-cover deficit’ values using a method that specifically separates grazing effects on remotely-sensed ground cover from that due to seasonal variation (Bastin et al. 2012).

 

Background

Owen Springs pastoral lease south west of Alice Springs, Northern Territory (NT), was first stocked in 1873. Grazing by cattle, rabbits and, at times, feral horses, combined with recurrent drought, had severely degraded the most preferred land type, calcareous shrubby grasslands (Figure 1). Bastin et al. (2023) estimated cattle density in 1988 at ~12 animal equivalents (AE) / km² and Foran et al. (1985) had previously recorded rabbit densities as high as 20 / spotlight km.

The lease was resumed by the NT Government in 2002 and all cattle and feral horses removed. The rabbit population was substantially reduced from about 1996 through rabbit haemorrhagic disease (Edwards et al. 2002). Exceptional rainfall in 2000-2001 combined with destocking promoted vegetation growth (Bastin et al. 2023). Part of the lease became the Old Man Plains (OMP) research station (522 km² in area) which was subdivided into smaller paddocks, waterpoint locations altered and progressively restocked from 2005 at between three and six AE / km² (Figure 2, see Bastin et al. 2023 for further detail).

Here, we report change in LI values for calcareous shrubby grasslands in two paddocks, Crows Nest and No. 1, between 1990 and 2023. Crows Nest is 25.9 km² with 10.6 km² (41%) of calcareous shrubby grassland. The corresponding figures for No. 1 are 28.3 km² and 18.6 km² (66%). Both paddocks were restocked in mid-2007 as part of a four-paddock rotation (grazed for three months then spelled for nine) with an average annual stocking rate (2007 – 2021) of 4.4 AE / km² in Crows Nest and 4.7 AE / km² in No. 1.

 

Figure 1. Aerial photo of part of Owen Springs from 1984 showing mostly bare ground and erosion on calcareous shrubby grassland in the foreground grading to red-earth mulga shrubland in the background. Photo: Graham Pearce, CSIRO.

 

Figure 2. Current paddocks and waterpoints on Old Man Plains (OMP) research station. The OMP area was originally one large, and a smaller, paddock as part of the Owen Springs lease (see lower right inset map).

 

Methods

The hydrologically enforced 1-second (~30 m) Digital Elevation Model (DEM) for Australia, derived from the Shuttle Radar Topography Mission (Gallant et al. 2011), was downloaded from Geoscience Australia (https://www.ga.gov.au/scientific-topics/national-location-information/digital-elevation-data, accessed 28 August 2024), subset to the area of OMP and reprojected to the Australian Map Grid (GDA94, MGA53) at 30 m resolution.

Landsat TM images of fractional cover between 1990 and 2023 were downloaded from http://qld.auscover.org.au/public/data/landsat/seasonal_fractional_cover/fractional_cover/nt/ (accessed 24 June 2024), subset and reprojected as for the DEM . The photosynthetic (PV) and non-photosynthetic (NPV) components were summed to provide percentage vegetation cover. This was treated as ground cover for areas of calcareous shrubby grassland because of generally sparse woody cover. Where possible, the September – November seasonal composite was accessed as it was likely to have the lowest ground cover throughout the year. Temporally adjacent images were used where cloud or other factors made the ‘spring’ composite unsuitable.

We briefly describe how landscape-scale leakiness is calculated (see Ludwig et al. 2007 for further detail). Leakiness is one minus the potential capacity of landscapes to retain resources, the following component in square brackets raised to the power k:

The parameter k defines the shape and steepness of the S-shaped decay curve formed when LI is plotted against vegetation cover. Here, we used the ‘perennial short tussock grasses’ function to estimate pixel-level leakiness from ground cover (Ludwig et al. 2007). This equation is:

LI = 0.993 / [ 1+ (GC/20.08)^3.467]
where LI is the pixel-level Leakiness Index value (0 – 1) and GC is percent ground cover.

L_calc is the sum of all the progressive flow values out of all the lowest pixels located along the paddock boundaries (Fig. 2).

L_max is the value obtained for L_calc when all pixel covers are zero (i.e. maximum potential leakiness). It also equals the total number of pixels analysed which, for a large area, reduces the ability to detect small changes in potential leakiness. To avoid this issue, we set L_max, by consistently using the earliest ground cover image in comparing LI change over time (i.e. LI_1990-1991, LI_1990,1992, ……, LI_1990-2023). This resulted in an L_max of 3,000 and 2,300 for Crows Nest and No. 1 paddocks respectively.

For convenience, L_min was set to zero. This assumed that calcareous shrubby grassland pixels in both paddocks had high ground cover that effectively trapped any soil sediments flowing (or blowing) about the landscape. We know this was not the case but setting L_min to a value greater than zero was problematic because we could not specify the expected seasonally-variable ground cover for the landscape in excellent condition. In practice, assigning L_min is not that critical; the aim is to have any changes in potential landscape leakiness with time be due to changes in vegetation cover and its spatial configuration as reflected in L_calc, not in L_min.

Annual LI values were computed with the LI Calculator (Leakiness Version 5, Liedloff et al. 2024).

 

Results

 

Terrain characteristics

Calcareous shrubby grasslands are higher in elevation in Crows Nest compared with No. 1 paddock (Table 1 and Figure 3). Some areas close to the southern foothills of the MacDonnell Range in Crows Nest paddock have relatively steep slopes but, across the whole land type, 64% of the area has slopes of ≤2%. The corresponding proportion for No. 1 paddock is 84%. These data indicate sufficiently defined hillslopes and drainage suitable for applying the LI Calculator.

 

Table 1. Descriptive statistics of elevation and slope data for calcareous shrubby grasslands in Crows Nest and No. 1 paddocks, Old Man Plains.

Statistic	Crows Nest paddock		No. 1 paddock
	Elevation (m)	Slope (%)	Elevation (m)	Slope (%)
Mean	615	1.77	584	1.01
Median	613	1.67	583	1.11
5th percentile	594	<0.001	567	<0.001
95th percentile	697	40.13	620	6.71

 

 

Figure 3. Digital elevation model for areas of calcareous shrubby grasslands in Crows Nest and No. 1 paddocks, Old Man Plains. Data made available by Geoscience Australia under the Creative Commons Attribution 3.0 Australia Licence.

 

Paddock-level leakiness

LI values were relatively high and variable in the 1990s following mostly below-median annual rainfall (Figure 4). At that time, the area of the current paddocks was grazed under extensive rangeland conditions as part of a much larger paddock on the Owen Springs lease. Leakiness substantially declined in the early 2000s following high rainfall, destocking and decline in the rabbit population (Edwards et al. 2002). Thereafter, leakiness increased in drier years to again approximate that of a decade earlier. Calcareous shrubby grasslands in both paddocks have been more conserving of rain water and nutrients in recent time apart from a small increase in leakiness following the dry years of 2018 and 2019. This recent stability is demonstrated by comparing the data for 1991 and 2018: 80 mm less rainfall in the 12 months to June 1991 increased LI from the previous year by 0.20 in No. 1 paddock. The corresponding values for 2018 were 105 mm less rainfall and a 0.02 increase in LI.

 

Figure 4. Top: Leakiness Index values for calcareous shrubby grasslands in two Old Man Plains paddocks between 1990 and 2023. Values were not calculated for 2012 because of malfunctioning of the Landsat 5 Thematic Mapper sensor. Bottom: July-June and median annual rainfall recorded at Alice Springs Airport (80 years of rainfall data).

 

Trend indicators align

Ground cover varied with rainfall (see Figures 1 and 11 in Bastin et al. 2023) and so, in turn, did calculated LI values (Figure 4). So how do we objectively determine whether altered grazing management has produced a real change in landscape leakiness? Bastin et al. (2023) proposed that retrospectively examining change in remotely-sensed ground cover following sequences of very low and exceptionally high rainfall removes vegetation dynamics generated by lesser rainfall events and can increase the ability to separate grazing effects from seasonal variability. Retained ground cover in dry times should reduce erosion risk and enhance landscape stability. It can also be reasonably inferred that where palatable species are present, persistent ground cover provides important forage for livestock. Thus, change in LI between successive dry years probably provides the most important signal against which to judge grazing management.

LI values following years of ≤ decile-2 rainfall had a similar trajectory of change to published ground-cover deficits for calcareous shrubby grasslands in Crows Nest and No. 1 paddocks (Figure 5, see Bastin et al. 2023 for detail) [see Note below]. Both metrics indicated that the landscape type was in a poor state in the 1990s. The absence of cattle, a depleted rabbit population and residual ground cover from the large growth event of 2000-2001 presumably contributed to minimal leakiness and reduced ground-cover deficit in 2005. Index values retreated from that improved state in 2008 with grazing having recommenced the previous year. Site-based monitoring data for 2008 showed ~10% ground cover and a pasture comprised of ~45% buffel grass (Cenchrus ciliaris) and the remainder, native grasses and forbs (Fig. 11 in Bastin et al. 2023). Buffel grass continued to increase comprising 70% of pasture biomass in 2020, thereby assisting landscape stability in that very dry year.

 

Figure 5. Cross-plot of ground-cover deficit () and Leakiness Index values for calcareous shrubby grasslands in Crows Nest and No. 1 paddocks in selected dry years. Symbols: triangles, 1991; asterisks, 1996; squares, 2005; diamonds, 2008; circles, 2020.

 

[Note: Spatially-averaged ground cover deficit  was derived using the Dynamic Reference Cover Method (Bastin et al. 2012)  This method calculates an expected (reference) level of ground cover for each pixel. The difference between actual and reference cover (i.e. deficit) indicates the extent to which an area has been modified by past grazing provided recent fire has not reduced ground cover.]

 

Discussion

The national availability of a hydrologically enforced DEM (Gallant et al. 2011) and multi-temporal fractional cover at ~30 m resolution means that the Leakiness Index can be more readily applied to suitably sloping areas of the Australian rangelands than was the case in 2007. We used the PV and NPV components of fractional cover to indicate ground cover on the basis that there was minimal tree and shrub cover on the calcareous shrubby grasslands of interest. Modelled ground cover based on the remotely-sensed fractional cover product is also available from the AusCover portal (e.g. https://portal.tern.org.au/, accessed 3 October 2024). We did not use these data with LI here because we were uncertain of the rigour with which this product has been validated in the Northern Territory. However, where validated, the modelled data would be useful for LI applications.

There were differences in elevation and slope between the two paddocks (Table 1) which meant that yearly LI values were not directly comparable. Our interest was more in evaluating change in leakiness of the landscape type in each paddock over time, as proposed by Ludwig et al. (2007). Nevertheless, both paddocks had a similar trajectory of change throughout and minimal differences in LI from 2000 onwards (Figure 4). This is mostly explained by their adjacency, therefore similar rainfall, and grazing management (similar stocking rate and spelling regime). The gradual increase, until 2019, of buffel grass as a component of the pasture (Bastin et al. 2023) likely contributed to the buffering of LI against variable rainfall in the last decade.

The Leakiness Calculator application (version 4) that provided a full user interface to manage settings, perform calculations, and visualise data and results is no longer compatible with the latest operating system. A command line interface was developed (version 5, Liedloff et al. 2024) to provide access to the leakiness functions using the same configuration and data files. All code is available upon request (https://bitbucket.csiro.au/projects/LEAK). Validation tests indicated the new command line version produced the same results as the previous version, which was expected given they use the same underlying calculation code.

A further issue related to the simple water- and sediment-routing mechanism implemented in the leakiness calculations that requires at least one neighbouring pixel of every pixel in the analysis area to have a lower elevation, with warnings thrown when pits are detected. This approach does not cater for groups of pixels in various parts of the DEM with only identical or higher elevation to neighbouring pixels (i.e. flats and pits). The solution was manual pixel editing to impose drainage across perfectly flat areas; a tedious task where numerous flats occur. This limitation can be overcome in future by (a) defining routing algorithms across pits and flats (i.e. automatic pit filling), or (b) allowing pits to accumulate resources and reduce the leakiness with flats reducing the movement of resources across the landscape. The magnitude of any reduction in flow before continuing across the landscape would need to be assessed as the Leakiness Calculation does not account for volume or flow rates.

The ecohydrological research leading to the Leakiness Index aligns with the latest flow management practices (e.g. leaky weirs) being implemented to retain valuable water and resources. The Leakiness Index is well suited for assessing the outcomes of these interventions and any resulting changes in cover at fine scales across grazed landscapes. These strategies also highlight the need to better understand the influence of pits and sinks across the landscape.

 

Acknowledgements

We thank Chris Materne and Robyn Cowley, NT Government, for providing the stocking density and pasture monitoring data as reported in Bastin et al. (2023).

 

References

Bastin GN, Abbott BN, Chewings VH, Wallace J (2007)  Metrics of landscape health for sustainable grazing in the Burdekin Dry Tropics, Queensland. Project report for the Sustainable Grazing Program, Great Barrier Reef catchments node, Water for a Healthy Country Flagship. CSIRO, 2007–06. 58 pp

Bastin G, Cowley R, Friedel M, Materne C (2023)  Applying two remotely-sensed methods for monitoring grazing impacts in the Australian arid zone. The Rangeland Journal 45(4), 141–159. doi:10.1071/RJ23030

Bastin G, Scarth P, Chewings V, Sparrow A, Denham R, Schmidt M, O’Reagain P, Shepherd R, Abbott B (2012)  Separating grazing and rainfall effects at regional scale using remote sensing imagery: a dynamic reference-cover method. Remote Sensing of Environment 121, 443–457. doi:10.1016/j.rse.2012.02.021

Dunwoody E, Apan A, Liu X (2013)  Effects of spatial resolution on measurement of landscape function using the landscape leakiness calculator. In Proceedings of the Surveying and Spatial Sciences Conference (SSSC 2013). University of Southern Queensland.

Edwards GP, Dobbie W, Berman DM (2002)  Population trends in European rabbits and other wildlife of central Australia in the wake of rabbit haemorrhagic disease. Wildlife Research 29(6), 557–565. doi:10.1071/WR00097

Foran BD, Low WA, Strong BW (1985)  The response of rabbit populations and vegetation to rabbit control on a calcareous shrubby grassland in central Australia. Australian Wildlife Research 12(2), 237–247. doi:10.1071/WR9850237

Gallant JC, Dowling TI, Read AM, Wilson N, Tickle P, Inskeep C (2011)  1 second SRTM Derived Digital Elevation Models User Guide. Geoscience Australia www.ga.gov.au/topographic-mapping/digital-elevation-data.html

Liedloff A, Ludwig J, Bastin G, Chewings V, Eager R (2024)  Leakiness Calculator (Version 5) Console Application using .Net Runtime 8.0.

Ludwig J, Bastin G, Chewings V, Eager R, Liedloff A (2007)  Leakiness: a new index for monitoring the health of arid and semiarid landscapes using remotely sensed vegetation cover and elevation data. Ecological Indicators 7(2), 442-454.