preliminary spatial model using fire scar challenge report - april... · 2014. 10. 17. ·...
TRANSCRIPT
Preliminary spatial model using fire scar
data to monitor Carpentarian Grasswrens
Steve Murphy1, Graham Harrington2 and Leasie Felderhof3
March 2011
1 Map IT PO Box 500 Malanda Q 4885 [email protected] 2 Birds Australia North Queensland PO Box 680 Malanda Q 4885 [email protected] 3 Firescape Science PO Box 158, Atherton Q 4883 [email protected]
2
Contents Acknowledgements ............................................................................................................................. 3
1.0 Summary ....................................................................................................................................... 4
2.0 Background ................................................................................................................................... 4
3.0 Methods ........................................................................................................................................ 5
3.1 Identifying sample areas for analyses ....................................................................................... 5
3.2 Fire data .................................................................................................................................... 8
3.3 Estimating fire-induced displacement ...................................................................................... 9
4.0 Results & Discussion ................................................................................................................... 12
4.1 Test for association with geology ........................................................................................... 12
4.2 Description of fire patterns ..................................................................................................... 12
4.3 Displacement analysis ............................................................................................................. 13
4.3 Fire challenge index ................................................................................................................ 15
4.4 General points for discussion .................................................................................................. 16
5.0 References .................................................................................................................................. 18
Appendix 1 – Geology in each population ........................................................................................ 19
Cite this report as: Murphy, S., Harrington, G. and Felderhof, L. (2011) Preliminary spatial model using fire scar data to
monitor Carpentarian Grasswrens. Report by Map IT and Birds Australia North Queensland.
3
Acknowledgements
The field surveys and cost of GIS work to produce this report was funded by grants from Xstrata
Mount Isa Mines and Mount Isa Water. We received essential logistic support from Southern Gulf
Catchments NRM. We would like to thank officers of the Northern Land Council for permission to
access Waanyi/Garawa Aboriginal Land Trust country. We would also like to acknowledge the efforts
of 37 volunteers from Birds Australia who carried out the field surveys at their own expense.
Colleagues at the CSIRO Ecosystem Sciences, Atherton have been supportive with time and
expertise.
4
1.0 Summary
Carpentarian Grasswrens (CGW) have been found in 4 main populations between Mount Isa
and Borroloola, where they are thought to be susceptible to unsuitable fire regimes.
The species survives in only three of these areas and Birds Australia has declared an
“Important Bird Area” over each.
The three northern populations appear to have declined (one to extinction) whilst the
southern population appears to be stable.
There are significant differences in fire patterns among these populations, driven mainly by a
north-south cline of decreasing fire size and frequency.
These differences translate into significant differences in the mean dispersal distances likely
to have been required by individual CGWs to cope with fire in each of the main populations.
Although relatively simple, the spatial and statistical models presented here provide a
preliminary framework to monitor the potential impacts of fires on CGW populations.
These models should be seen as the first steps towards an improved monitoring system,
which ideally would incorporate better fire scar data (i.e. based on Landsat imagery) and a
more detailed understanding of certain aspects of CGW biology. Together, these
improvements will lead to a more complete picture of the effects of fire on CGWs and more
accurate monitoring based on remotely sensed data.
2.0 Background
The Carpentarian Grasswren (CGW; Amytornis dorotheae) is a small, insectivorous passerine
confined to north-western Queensland and north-eastern Northern Territory (Fig 1; Higgins et al.
2001). Broadly speaking, the species occupies rocky savanna systems (Higgins et al. 2001; Rowley
and Russell 1997), typically dominated by the highly resiniferous and therefore flammable grass
spinifex (Triodia spp.). The region has a monsoonal climate, leading to profuse herbaceous growth in
the wet season (roughly Dec-Mar), followed by drying and curing during the remainder of the year.
As such, fire is a regular feature of the landscape.
The CGW is listed as Endangered in the Northern Territory, Near Threatened in QLD and Birds
Australia has submitted a case for this species to be declared Vulnerable at the federal level. The
main threatening process is thought to be the contemporary fire pattern, which is dominated by
frequent, large fires that often burn through the latter part of the dry season.
Birds Australia has accepted three “Important Bird Areas” (IBA), for which the CGW is the primary
focus and they are monitored by Birds Australia North Queensland.:
1. Buckley River IBA covers an area of 483,710 ha stretching from Mount Isa north to
Gunpowder;
2. Boodjamulla IBA is synonymous with Boodjamulla (Lawn Hill) National Park and extends over
372,271 ha.
3. The 960,000ha Wollogorang IBA spans the Queensland/Northern Territory border.
5
This objective of this study was to develop a framework for monitoring populations of CGWs by
analysing annual fire patterns. It is based on a spatial model that simulates the displacement of
individual birds following real fires that occurred between 2004-2010, given the actual
distributions of unsuitable and suitable habitat. It also considers the proportion of pairs likely to
have been affected by fires in each population each year.
3.0 Methods
All spatial manipulations and analyses were performed using ArcView 10.0 (Environmental System Research Institute Inc., Redlands, CA, USA) and ET Geowizards (www.ian-ko.com, verified March 2011).
3.1 Identifying sample areas for analyses
The areas chosen for this fire mapping study are based on the current and past known locations for
CGWs and were selected to be comparable to Birds Australia’s IBA boundaries. The northern area
near Borroloola is not an IBA because the species no longer occurs there but the fire characteristics
may help to explain why the species is locally extinct. The boundaries of the other three areas are
not identical to the boundaries of the IBAs because they are based on historical as well as current
distribution of the species. Point locations for CGW sightings came from two main sources: (1)
miscellaneous historical records, including those published in the two atlases published by Birds
Australia (Barrett et al. 2003; Blakers et al. 1984) and (2) systematic surveys (Harrington et al. 2009).
These records (199 in total) fall into four clusters: Borroloola, Wollogorang, Boodjamulla (Lawn Hill)
and Buckley River (Mount Isa). Minimum convex hulls with a 10km buffer were created about each
of these clusters (Fig. 1).
6
Fig. 1. CGW locations used to define minimum convex hulls with a 10km buffer, for each population.
Historical and current CGW records were edited to remove multiple sightings at single localities. The
remaining records were overlayed onto 1 million scale geology data (Geoscience Australia,
www.ga.gov.au) to determine the geological types important to CGWs (Table 1; Appendix 1). These
geological types were buffered by 1km (to allow for minor errors in the locations of records) and
then clipped using the above-mentioned convex hulls to define each sample area (Fig. 2). A test for
association was done using a contingency analysis, with the area of each geology type expressed as a
proportion of the total area used to calculate expected values for bird censuses.
7
Table 1. Occurrence of CGWs on major geological substrates.
Geology (Lithgroup1) Hectares
(proportion*) No. CGW sightings
(proportion) meta-igneous mafic volcanic
(metamorphosed basalt) 251,841 (0.060)
7 (0.04)
metasedimentary siliciclastic (metamorphosed sandstone)
409,710 (0.097)
16 (0.09)
regolith (outwash gravel and sandy alluvium)
1,377,499 (0.326)
28 (0.15)
sedimentary carbonate (limestone)
622,318 (0.147)
5 (0.03)
sedimentary siliciclastic (sandstone)
1,570,383 (0.371)
125 (0.69)
* Proportion of minimum convex hull; excludes substrates within convex hull where no CGW were
observed.
Fig. 2. Samples areas used in this analysis defined by suitable geological types within the minimum
convex hulls of all historical locations of CGWs.
8
3.2 Fire data
The majority of the fire data used in this analysis were sourced from the North Australia Fire
Information (NAFI) website (www.firenorth.org). These data are derived from imagery captured by
MODIS sensors aboard the two satellites, Terra and Aqua. As mapped by NAFI, MODIS data has a
spatial resolution of approximately 250m (NAFI). As with any remote sensing, errors were
encountered using the NAFI fire dataset. Most errors are omissions, and a good example is
presented in Fig. 3.
Fig. 3. Example of MODIS-NAFI omission error in 2006 from Boodjamulla (Lawn Hill) National Park.
The underlying raster is a Landsat 5TM image (2 Dec 2006) with fire scars highlighted.
A second MODIS-derived dataset (MCD45) was used to fill gaps left by MODIS-NAFI (Boschetti et al.
2009). This product uses a different approach to identifying fire scars and has a spatial resolution of
500m (Roy et al. 2008; Roy et al. 2005; Roy et al. 2002). Fig 4 shows an example of where MODIS-
MCD45 was used to fill a gap that the MODIS-NAFI dataset had missed.
9
Fig. 4. Part of the fire scar omitted by MODIS-NAFI 2006, filled by the MODIS-MCD45 product.
3.3 Estimating fire-induced displacement
There are 3 key fire-related attributes in any given year that are important to CGWs:
(1) The distribution of suitable habitat at the start of each year. This is defined as habitat that
has escaped burning for 3+ consecutive years in the north (Borroloola, Wollogorang and
Boodjamulla) and 4+ consecutive fire-free years at Mount Isa. This temporal difference is
related to lower rainfall, and therefore slower regeneration at Mount Isa (Harrington et al.
2009).
(2) Fire scars that occur within suitable habitat by the end of each year. This assumes fires in
unsuitable habitat do not affect CGWs in that year.
10
(3) The distribution of suitable habitat at the end of each year. Any fires that occur within
suitable habitat throughout the year affect the proximity of suitable habitat at the end of
the year.
Fig. 5 shows an example of the key stages in the transition from suitable to unsuitable habitat.
Fig. 5. Example area for key stages in the changes of suitable to unsuitable habitat during a year.
11
The underlying assumption of this analysis is that fires in suitable habitat displace a number of CGW
pairs to the nearest unburnt suitable habitat. This assumption relates to fire being identified as the
main process threatening CGWs (see Background). The number of pairs affected by a particular fire
depends on (a) variation in habitat quality and (b) home range size. For this analysis, we have
assumed homogeneity in habitat quality. We have also assumed that pairs occupy 25ha. As such, the
number of pairs affected by each fire can be calculated. The number of pairs affected by fire in any
particular year expressed as a proportion of the total number of pairs in each population (calculated
using the total area of suitable habitat at the beginning of each year) is a critical parameter.
A number of randomly located points were generated within each relevant fire scar (related to the
size of that scar). These represent CGW pairs. The perpendicular distance from each point to the
nearest suitable habitat edge (i.e. a pair’s displacement distance) was calculated, and means and
standard errors calculated for each year, for each sample area. A graphical example of displacement
calculation is shown in Fig. 6.
Fig. 6. Example of how displacement data was generated for CGWs following fire in suitable habitat.
12
4.0 Results & Discussion
4.1 Test for association with geology
The contingency table used to test for a non-random association with geology is shown in Table 3.
The association between CGWs and geology was significantly non-random (χ42 = 85.03 p<0.01).
CGWs showed a positive selection for sandstone (especially sedimentary siliclastic) and a negative
association with limestone (sedimentary carbonate). The term "regolith" included both clay and
gravel sediments and we had no measure of their relative contribution to our mapped areas. Thus
we cannot draw conclusions as to the preference or otherwise for this class of substrate. CGWs
were not recorded on clay soils and we presume that these records are on or close to outwash fans
of gravel from rocky hills.
Table 3: Contingency table to test for association between CGWs and geology
MIMV MSS R SC SS
Observed 7 16 28 5 125
Expected 10.8 17.5 58.9 26.6 67.2 MIMV: meta-igneous mafic volcanic; MSS: metasedimentary siliciclastic; R: regolith; SC: sedimentary carbonate; SS: sedimentary siliciclastic
4.2 Description of fire patterns
Basic statistical descriptions of annual fire patterns for each CGW population are shown in Table 4.
Fig. 7 shows the mean proportion of each population area burnt for all years. There is a significant
difference in the proportion of area burnt between the populations (F3, 24 = 4.80, p = 0.0093), driven
by a strong north-south cline, with larger frequent fires in the north compared to the south.
Table 4. Basic annual descriptive statistics of fire patterns within each CGW population boundary.
Areas shown are in hectares. (Prop. is the proportion of total area of each convex hull burnt).
2010 (prop) 2009 (prop) 2008 (prop) 2007 (prop) 2006 (prop) 2005 (prop) 2004 (prop)
Borroloola
Wollogor.
Boodja.
Mount Isa
169,235 (0.2)
25,646 (0.04)
11,787 (0.04)
321,595 (.11)
282,060 (0.33)
220,332 (0.32)
24,418 (0.09)
69,031 (0.09)
154,664 (0.18)
107,239 (0.16)
0 (0)
65,209 (0.02)
314,934 (0.37)
95,877 (0.14)
0 (0)
80,360 (0.03)
323,375 (0.38)
268,188 (0.39)
134,468 (0.51)
285,684 (0.1)
97,035 (0.11)
107,558 (0.16)
8,015 (0.03)
41,133 (0.01)
346,717 (0.41)
325,762 (0.48)
15,562 (0.06)
66,836 (0.02)
13
Fig 7. Mean proportion of area burnt for all seven years, grouped by CGW population.
Standard error bars are 1 standard error from the mean.
4.3 Displacement analysis
The proportion of pairs affected by fire for all years differed significantly among the populations
(F3,12=4.6, p=0.023), which is expected given the differences in extent and frequency of fire among
the populations. Visual inspection of the data, when split by year (Fig. 8), shows more consistency in
the proportion of pairs affected by fire in Mount Isa (and to some extent Boodjamulla) compared to
the northern populations.
14
Fig. 8. The proportion of all pairs affected by fire in each population, split by year.
The significant differences in annual fire statistics for each population were also reflected in
significant differences in mean displacement distances of fire affected birds generated from the
simulated spatial model. An analysis of variance of displacement distance among populations
(pooled by year) showed signficant differences (F3, 78994 = 74.45, p<0.0001), with Boodjamulla having
the shortest mean distance compared to the other populations. This is largely driven by there being
no detected fires in 2007 and 2008, and therefore no displaced birds in those years. In a least
squares regression, the interaction year•population explained a significant amount of variation in
displacement (F11,76150 = 479.7, p<0.0001; Fig. 9). Note that Boodjamulla was excluded from this
model as it lost degrees of freedom because of two years of no displacements (i.e. no fires).
15
Fig. 9. Mean displacement distance for CGW populations for each year. Standard error bars are 1 standard error from the mean. Numbers on bars are sample sizes.
4.3 Fire challenge index
The objective of this study was to develop a framework for monitoring CGW populations using
annual fire scar data. There are significant differences in the proportion of area burnt annually in
each population (Fig. 7) and these differences are broadly consistent with what is understood to be
the relative population size and stability of each population. That is, the three northernmost
populations are extinct (Borroloola) or very small (Wollogorang and Boodjamulla) compared to the
southern, more stable one at Mount Isa (Harrington et al. 2009). Given this relationship, it is
313 255 266 698
731 75 602 318
299 378 728 63
0 0 96 45
16
tempting to think that simply analysing the extent of annual fires in each population is sufficient for
monitoring. However, it is important to consider the spatial distribution of burnt and unburnt
patches beyond simply considering area burnt – two years may have identical areas burnt, but
present very different challenges to CGWs because of the spatial distribution of suitable and
unsuitable habitat. Our displacement model is designed to capture these potential differences. One
must also consider the proportion of pairs affected by fire in order to interpret displacement data in
context.
Until we have a better understanding of the relative effects of displacement and proportion of pairs
affected on population viability we remain cautious about combining these two parameters into an
overall index. Moreover, unless the relationship between the two parameters is relatively straight-
forward such an index could become meaningless in a biological sense. Alternatively, for the time
being at least, annual monitoring of CGW populations should state mean displacement (and
standard error) and the proportion of pairs affected as independent descriptors.
4.4 General points for discussion
The strength of the approach presented here is that the models (both spatial and statistical) included
actual fire scar data, and the parameters relating to CGW biology were kept to an absolute minimum
(i.e. only home range size, which was used to estimate the number of pairs affected by each fire scar.
Note that the relative effect of fire on each population would not be affected by the home range size
unless this varies between populations). This is appropriate given the paucity of ecological
knowledge about this species.
We acknowledge that the models presented here could be improved and expanded in a few key
areas. First, the fire scar data used here is rather coarse (250m pixel size at best) and as stated
above, is prone to omission errors. An important effect of these issues is that suitable unburnt
habitat patches below a certain size are likely to be overlooked, resulting in over-estimated
measures of displacement. It is highly recommended that future models include fire scar data
derived from Landsat satellite imagery, which has a resolution of 30m (L5-TM).
One of the important benefits from conducting a study such as this is that it highlights and prioritises
key gaps in our knowledge of CGW biology. A better knowledge of home range size will improve our
understanding about the relationship between individual fire sizes and their impact on CGWs.
17
Information is also required about CGW dispersal abilities. Such data will allow the models described
here to be expanded to answer questions relating to the immediate effects of fire on mortality.
Questions relating to longer term post-fire effects will require information about the effects of living
in small unburnt patches with increased population density, especially how this relates to things such
as social system effects, changes to physiology etc. Finally, although the assumptions used here
about when habitat becomes suitable for CGWs are expected to be reasonably accurate (i.e. 3+ years
post-fire in the north, and 4+ in the south), a more detailed understanding about effects of different
ages habitat on critical life history parameters, such as mortality and breeding success needs
investigation.
18
5.0 References
Barrett, G., Silcocks, A., Barry, S., Cunnigham, R., and R., P. (2003) 'The new atlas of Australian birds'. (CSIRO and Royal Australasian Ornithologists Union: Victoria, Australia.)
Blakers, M., Davies, S.J.J.F., and Reilly, P.N. (1984) 'The Atlas of Australian Birds.' (Melbourne University Press and Royal Australasian Ornithologists Union: Carlton, Victoria)
Boschetti, L., Roy, D., and Hoffman, A.A. (2009) MODIS Collection 5 Burned Area Product - MCD45 User's Guide Version 2.0. University of Maryland, South Dakota State University, LM University of Munich.
Harrington, G., Perry, J., Forsyth, R., and Venables, B. (2009) A Tale of Two Grasswrens. Wingspan 19, 23-25.
Higgins, P.J., Peter, J.M., and Steele, W.K. (Eds) (2001) 'Handbook of Australian, New Zealand and Antarctic birds. Volume 5: Tyrant-flycatchers to Chats.' (Oxford University Press: Melbourne)
Rowley, I., and Russell, E. (1997) 'Fairy-wrens and Grasswrens.' (Oxford University Press: Oxford)
Roy, D.P., Boschetti, L., Justice, C.O., and Ju, J. (2008) The Collection 5 MODIS Burned Area Product - Global Evaluation by Comparison with the MODIS Active Fire Product. Remote Sensing of Environment 112, 3690-3707.
Roy, D.P., Jin, Y., Lewis, P.E., and Justice, C.O. (2005) Prototyping a global algorithm for systematic fire-affected area mapping using MODIS time series data. . Remote Sensing of Environment 97, 137-162.
Roy, D.P., Lewis, P.E., and Justice, C.O. (2002) Burned area mapping using multi-temporal moderate spatial resolution data - a bi-directional reflectance model-based expectation approach. Remote Sensing of Environment 83, 263-286.
19
Appendix 1 – Geology in each population Based on 1:1 Million scale data – www.ga.gov.au
20
21
22