6 Plant Protection Quarterly Vol.18(1) 2003 The focus is primarily on the policy and management implications. First, the im- Using aerial mapping to analyse factors affecting the pacts of Scotch broom in the Park are de- spread of Scotch broom scribed briefly and those factors that pro- mote the spread of broom are reviewed. Doreen OdomA, G.R. GriffithAB, Mellesa SchroderC and J.A. SindenA Next the aerial maps that form the basis A Graduate School of Agricultural and Resource Economics, University of of the analysis are described. These maps provide data on spread and density of New England, Armidale, New South Wales 2351, Australia. broom at different times, and provide an B NSW Agriculture, Armidale, New South Wales 2351, Australia. opportunity to evaluate whether the use C NSW National Parks and Wildlife Service, Nelson Bay, New South Wales of maps like these is a viable alternative to 2315, Australia. extensive fieldwork. Following, the data calculations and the empirical methods used for the analysis are discussed. Fi- nally, the results are presented and some Summary New South Wales and Victoria, and in conclusions are drawn. Scotch broom is an invasive weed in western Tasmania. It has also been record- many subalpine ecosystems. It often has ed around Perth, Western Australia. The Impacts of Scotch broom in the substantial negative effects on ecosystem total area infested in Australia is estimated Barrington Tops National Park structure and functions. Decisions on to be over 200 000 ha and still spreading The Barrington Tops National Park is par- optimal management strategies require (Hosking et al. 1998). ticularly important ecologically because it predictions of the rates and patterns of A study to investigate the economics of comprises one of the fifteen areas of rain- Scotch broom spread. This paper ex- management strategies for Scotch broom forest within New South Wales which are plores the environmental and manage- in the Barrington Tops National Park has included in the World Heritage Listing of ment factors that influence the rate of involved the construction of a dynamic the Sub- tropical and Temperate Rainfor- spread of Scotch broom in Barrington programming model of the Park (Odom est Parks of Eastern Australia. The rain- Tops National Park. et al. 2001). This model requires as input a forest of Barrington Tops National Park is The NSW National Parks and Wildlife biological model of Scotch broom spread. the core of one of the five major regions of Service (NPWS) and the then neighbour- Rees and Paynter (1997) developed a rainforest originally present in New South ing NSW State Forests prepared aerial spatial model for Scotch broom, which Wales before European settlement (Trudg- maps showing Scotch broom infestation allowed them to explore changes in popu- eon and Williams 1989). in the Park for 1989, 1993 and 1999. These lation size and the proportion of ground It is an area where cold adapted spe- maps were used to generate data for the covered by the weed. Two versions of this cies have survived climatic warming and current analysis. Map reference points 1 model were presented. The first model contains sub-alpine swamps as well as km apart along the southern boundary of was a complex simulation model which grassland, woodland, open forest and the the 1989 area of infestation were exam- was spatially explicit and incorporated rainforest mentioned earlier. ined and 1993 and 1999 differences from spatially local density-dependent compe- Extensive tracts of the Barrington Tops the reference points were measured. Also tition, asymmetric competition between National Park, bordering Polblue Con- measured were environmental factors seedlings and established plants, a seed servation Reserve, Stewarts Brook State including natural vegetation type, natu- bank, local seed dispersal and an aged- Forest, Barrington Tops State Forests and ral vegetation density, soil type, slope, structured established plant population. adjoining private land on the Barrington altitude and the presence of private This model incorporated much of the Tops plateau, have been invaded by dense property or crown land. The incidence of known population biology of broom. stands of Scotch broom. An almost im- natural disaster, feral animal activity and The second model was an analytical ap- penetrable shrubby understorey of Scotch NPWS management activities were also proximation of the simulation model. broom currently occupies large areas that included in the analysis. Analytical approximations were devel- prior to the 1960s was relatively open veg- The regression analysis shows that oped in order to allow the simulation etation. A major concern is that the inva- the three variables contributing most to model to be interpreted and to illuminate sion by Scotch broom will have long-term an expansion of broom are sparse vegeta- inter-relationships between parameters. effects on the native vegetation of the pla- tion (62 m pa), flat slopes (23 m pa) and Data from many different regions includ- teau. A number of plant and animal spe- State Forest land (23 m pa). The three ing Australia, New Zealand, the United cies listed as rare or threatened are found variables contributing most to a reduc- States, the United Kingdom, France, Eire only in the areas now infested by broom tion in broom spread are NPWS general and Spain were used in this model. (Heinrich and Dowling 2000). As a conse- treatment activities (-14 m pa), NPWS This existing biological model for quence, the value of vegetation conserved specific wetlands treatment (-40 m pa) Scotch broom has not been validated by in the Park may decline and the stock car- and steeper slopes (-59 m pa). data explicitly from Barrington Tops Na- rying capacity of local grazing lands may Key words: aerial mapping, Scotch tional Park. Thus, the aim of this paper is be reduced (Waterhouse 1988). Prior to the broom, spread, park management, re- to explore the environmental and manage- hand over of NSW State Forest land to the gression models. ment factors that have influenced the rate NSW National Parks and Wildlife Service, of spread of Scotch broom on Barrington the economic potential of stands of com- Introduction Tops1. An understanding of how Scotch mercial timber in the State Forests may Scotch broom (Cytisus scoparius (L.) broom spreads here will help decide have been jeopardized. Link) has become a major weed in parts whether there is a need to modify the re- Dense thickets of Scotch broom have of south-eastern Australia (Smith 1994, lationships between the parameters used blocked disused tracks and have hin- Hosking et al. 1998, Schroder and Howard in the Rees and Paynter biological model dered access to some watercourses on 2000). It now poses a serious threat in to fit the case of Barrington Tops. In doing the plateau. Many recreational users of many regions including the Barrington this, we look broadly at average rates of the plateau regard the infestation as a Tops National Park in New South Wales, spread across the whole Park rather than considerable hindrance to access and the Australian Alps National Parks in look in detail at individual plants or sites. aesthetically less pleasing than the natural Plant Protection Quarterly Vol.18(1) 2003 7 open vegetation. The thickets are thought Additional aerial surveys of the bound- Table 1. Summary of broom spread, to provide harbour for feral pigs, which ary of the broom infestation were under- 1989–1999. are causing considerable local damage to taken in 1995/1996 using a helicopter- Description of Number Average the ground surface and vegetation in the mounted Global Positioning System ‘live spread of sites spread native forests (Waterhouse 1986, Hosk- linked’ to a geographic information sys- ing and Smith 1999). Attempts by private tem on a lap-top computer (Carter and Si- Expansion 24 34 m pa landholders, the NPWS and State Forests gnor 2000). This exercise was done for the No movement 8 0 to control the infestation have been costly purpose of locating isolated infestations Contraction 23 35 m pa and largely unsuccessful due to large soil for on-ground treatment. A survey using Total 55 2 m pa seed banks and limited access to most of this technology was carried out again in the infestation (Wapshere and Corey 1999, late 1998, and maps were produced in Smith 1994, Waterhouse 1988). 1999, so as to compare with the 1989 map area were given a value of two, and (Schroder and Howard 2000). iii. how far the broom has spread (in kilo- Factors that promote the spread of Limited fieldwork was undertaken metres). The spread distance from 1989 Scotch broom after each aerial survey to validate the to 1993, 1993 to 1999, and 1989 to 1999 Scotch broom produces 30–360 seeds per outer edge of the broom infestation and was measured in centimetres on the square metre per annum below a eucalypt to check the density classes (McDonald map and then converted to kilometres. canopy, and 7700–8900 seeds per square 1999). Most of the work was undertaken A summary of the spread of broom at the metre per annum in the open (Hosking et around Polblue Swamp and other easily measured sites over the 10-year period is al. 1998). In Australia seedbanks in the soil accessible areas. shown in Table 1. have been measured to contain from 190– Hence, there are 1989 and 1999 maps 50 000 seeds per square metre (Smith and available from the NPWS with similar Vegetation types and density Harlen 1991). Scotch broom plants live to information, and these can be compared, The broad types of natural vegetation a maximum of about 20–25 years and all together with information for 1993, to communities for each of the observation regeneration is by seed rather than by veg- show changes in the spread of the broom points were obtained from NPWS staff. etative means (Hosking et al. 1998). Long- infestation. The types identified were open eucalypt term control of the plant is complicated by forest, rainforest, riparian, and subalpine. the presence of this large, long-lived soil Data collection The density of the natural vegetation was seedbank (Smith and Harlen 1991). The southern sections of the two existing recorded from the 1985 Barrington Tops Scotch broom is restricted to mainly maps showing Scotch broom infestation topographic map. The densities were cool temperate areas of Australia. In in Barrington Tops National Park for 1989 dense, medium, and scattered. New South Wales most of these areas are and 1999 were the basis of data for the over 600 m above sea level. Broom occurs analysis. Some 55 map reference points Slope on soils derived from a wide variety of one kilometre apart along the boundary The slope of each of the map reference substrates, particularly basalt. It grows of the 1989 area of infestation were plot- points was calculated from the topograph- best on moist, fertile soils and is rarely ted onto the 1999 map. Changes in the ic map, and recorded as a fraction. found on undisturbed skeletal sandy boundary between 1989, 1993 and 1999 soils. It spreads rapidly in open spaces were recorded at each point. These obser- Altitude and more slowly in dense vegetation such vation points were then transferred to the The altitude of each of the map reference as rainforest. Apart from normal seed relevant 1:25 000 topographic maps. points was recorded from the topographic drop, Scotch broom has been shown to be For each of the points, both environ- map, and recorded in metres. spread over longer distances in four major mental and management factors were ob- ways: by ants (Smith and Harlen 1991); by tained. Environmental factors measured Soil type animals (Smith and Harlen 1991, Harlen include natural vegetation types, natural The soil type at each of the observation 1989, Moodie 1985, Waterhouse 1988); by vegetation densities, soil types, slope and points was taken from the Heinrich and human beings (Smith and Harlen 1991, altitude and whether the point was on Dowling (2000) mapping of rare and Harlen 1989, Moodie 1985, Waterhouse private property or crown land. Areas of threatened plants in the Barrington Tops 1988, Hobbs 1991, Smith 2000, Robertson natural disaster, feral animal activity and National Park. The soil types identified et al. 1999); and by natural disasters (Wil- NPWS management activities were also in that study were peaty, moist loam, me- liams 1981, Harlen 1989, Smith and Harlen included in the analysis. These data were dium brown loam, and deep brown loam. 1991). estimated as follows. Presence of crown land The aerial maps Broom spread If an observation point was located in NPWS and State Forests undertook map- Broom spread between 1989, 1993 and crown land (the old State Forests area) ping of the Scotch broom infestation in 1999 was measured in three different rather than NPWS land, it was given a Barrington Tops National Park on several ways: value of one, otherwise zero. occasions during the 1980s and 1990s. In i. whether the broom had spread or not 1988, data from 1982 aerial photographs (yes/no). The observation points that Natural disaster were mapped to identify the boundaries showed a positive spread at 1993 and/ Due to high rainfall on the Barrington of the infestation. This was updated and or 1999 were given a value of one, while Tops the most likely natural disaster is enhanced in 1989 by the production of a those that had no positive spread were floods. The observation points considered more detailed map showing the density given a value of zero, to be most affected are those close to the of broom from 1987 aerial photographs. ii. whether the spread was zero, positive rivers and water catchments. These points State Forests again mapped the perimeter or negative. The observation points were given the value of one, and all other in 1992/1993, and this map indicated a that showed no change at 1993 and/or points were given a value of zero. slight increase in the area of the main 1999 were given a value of zero, those infestations from 1989 (Schroder and that showed a positive change at 1993 Animal activities Howard 2000). The 1993 information was and/or 1999 were given a value of one, On Barrington Tops the animals that cause overlaid onto the 1989 map. while those that showed a reduction in most disturbance are feral pigs. Based on 8 Plant Protection Quarterly Vol.18(1) 2003 discussions with the Park staff, those Table 2. F-tests for groups of environmental variables. observation points that were thought to Period Vegetation density Vegetation type Soil type have feral pig activity were given the value of one and observation points with 1989–1999 no records on feral pigs were given a value F value 2.93* 0.96 1.22 of zero. * significant at the 10% level. NPWS management activities Information on broom management due to perfect correlation with the yes or Table 3. Preferred OLS results. activities at the observation points was no choice. The overall level of explanation Variable 1989–1999 obtained from NPWS staff. Some of the was also very low at 23% for the Probit points were in areas where broom had model and 10% for the Logit model. These Constant 0.377 been treated, and these were given a value results are not reported here, but interest- (0.123) of one. Points in areas where no treatment ed readers may consult Odom et al. (2002). 3.057 was applied were given a value of zero. A Similarly, the OLS results for the two sub- 0.002 second treatment variable was created for periods, 1989–1993 and 1993–1999, had NPWS treatment -0.198 a particular small area of wet peat where low R2 values and relatively few signifi- (0.122) intensive hand pulling of young broom cant coefficients so they are not reported -1.624* plants has been done. here either. 0.056 The full OLS model outlined above State Forest land 0.288 Method of analysis for the dependent variable kilometres (0.163) Dummy variables were constructed in spread was estimated first. Few variables 1.763** the usual way for the yes/no variables showed significance because of the prob- 0.042 defined above, where the presence of the able high correlations between the various factor was denoted by one and the absence environmental factors. Those sets of envi- Slope -0.953 by zero. For those independent variables ronmental factors with multiple dummy (0.287) with more than two categories (vegetation variables, natural vegetation densities, -3.318** type, vegetation density and soil type), a natural vegetation types and soil types, 0.001 base class was chosen and one-zero dum- were then tested separately as groups Sparse vegetation 0.391 my variables were constructed for the using an F test. The results are shown in (0.310) remaining classes. For example, for the Table 2. Only the vegetation density vari- 1.259 vegetation density variable, medium was able showed a significant contribution to 0.107 chosen as the base class and two dummy explaining variance. Vegetation types and variables were constructed for the dense soil types were also tested individually Floods -0.354 and scattered categories respectively. The and none were found to be significant, so (0.151) categorical dependent variables were con- vegetation type and soil type were omitted -2.350** structed as described above. The variables in all successive estimations. 0.012 for spread in kilometres, slope and altitude The possibility of interactions between NPWS wetland treatment -0.496 were treated as continuous variables. the explanatory variables was considered (0.234) Due to the nature of the three depend- but also ignored due to insufficient data -2.114** ent variables, three different methods points. 0.020 were required to estimate the models: The preferred model is reported in Ta- • an Ordinary Least Squares (OLS) ble 3. For each set of values, the estimated R2 0.389 model (Griffiths et al. 1993) was used coefficient is given first, the standard error * statistically significant at the 10% level on for kilometre spread as the dependent second, the t-statistic third, and the prob- a one-tail test. variable, ability value fourth. Based on the review ** statistically significant at the 5% level on • a Probit model (Pindyck and Rubinfeld material referred to earlier, the expected a one-tail test. 1998) was used for yes/no spread as signs for the explanatory variables can be the dependent variable, and predicted with some confidence, so one- • a Multinomial Logit model (Maddala tail tests are used for the t-statistics. 1992) was used for the categorical had the expected negative sign with a co- zero/positive/negative spread as the Kilometres spread 1989–1999 efficient estimate of -0.198 and a P-value dependent variable. In general, the results for the full period of 0.065, statistically significant at the 10% In all three cases, the estimated model was show that the coefficients of all of the in- level on a one-tailed test. This implies that, of the general form: cluded variables have the expected signs, for areas where treatment was done, the Dependent Variable = and all except one are significant at the average spread was 0.198 of a kilometre F(slope, altitude, vegetation-density 10% level. The R2 statistic implies that less than the average over the ten year dummy variables(2), vegetation-type about 39% of the variability in the spread period. dummy variables(3), soil-type dummy of broom is explained by the model. The variable indicating the site was variables(4), treatment dummy vari- At the mean values of the included in a wetland area and received special ables(2), flood dummy variable, pig continuous explanatory variables, the base management activity had the expected activity dummy variable). cases of the included categorical variables negative sign with a coefficient estimate and the zero values of the included yes/ of -0.496 and a P-value of 0.020. This addi- Results no dummy variables, the results show that tional control activity reduced the spread Unfortunately, with the relatively small over the 10 year period 1989–1999, the es- by 0.496 of a kilometre on average in the numbers of observations, the Probit and timated expansion in broom was 0.377 of areas applied. Logit model results were not reliable with a kilometre across all sites. The variable indicating that the site several of the proposed explanatory vari- The variable indicating the presence was on crown land (previously managed ables having to be excluded or modified or absence of NPWS treatment at the site by State Forests) had the expected positive Plant Protection Quarterly Vol.18(1) 2003 9 sign with a coefficient estimate of 0.288 • on lands which were previously man- of other significant variables. It would be and a P-value of 0.042. This implies that, aged by State Forests, spread increased expected also that those factors causing on the crown reserve land, the average by 0.288 km on average during the 10 changes in the density of broom could be spread of broom was 0.288 of a kilometre year period, more confidently examined. greater than the average over all sites, over • where vegetation is sparse, the spread The next step is to decide whether the 10 year period. increased by 0.391 km on average dur- the biological model for Scotch broom The estimated coefficient for the slope ing the period, needs to be modified by including the sig- measurement at the site had a value of • where slopes are steep, the spread is nificant environmental and management -0.953 and a P-value of 0.001. This coef- 0.953 km less than average, and factors that influence the density and the ficient implies that on steeper areas the • where there are floods, spread is 0.354 rate of spread. By doing this, the relation- spread was reduced by 0.953 of a kilometre km less than average. ship between the parameters used in the on average. Unfortunately, this measure Another way of interpreting the results is biological model of Scotch broom may be included both uphill and downhill slopes, to predict from the OLS model the ranges improved to fit the case of Barrington Tops and these could be usefully separated in of spreads, under different environmental National Park, and more reliable policy any future work. and management factors, as metres per implications may be able to be derived. The estimated coefficient for sparse year, with all other variables at their mean vegetation had the expected positive sign values. From such a prediction, the three References with a coefficient estimate of 0.391 and a variables contributing most to an expan- Carter, K. and Signor, A. (2000). Control- P-value of 0.107, just outside the 10% criti- sion of broom are sparse vegetation (62 m), ling broom (Cytisus scoparius) in na- cal level. The coefficient implies that the flat slopes (23 m) and State Forest land (23 tive forest ecosystems. Plant Protection spread was 0.391 of a kilometre greater m). The three variables contributing most Quarterly 15, 165-6. than the average on areas with sparse to a reduction in broom spread are NPWS Griffiths, W.E., Hill, R.C. and Judge, G.G. vegetation compared to more dense vege- general treatment (-14 m), NPWS specific (1993). ‘Learning and practising econo- tation cover. This agrees with the literature wetlands treatment (-40 m) and steepest metrics’. (John Wiley and Sons, New that broom spreads more in open areas. slope (-59 m). Many of these factors work York). Finally, the variable indicating that the together. For example, the old State For- Harlen, R.L. (1989). Seed Dispersal and site was prone to floods had an unexpect- est land has sparser vegetation and flatter Dormancy in Cytisus scoparius broom ed negative and significant effect, with a slopes as well as less management input. on Barrington Tops, New South Wales. coefficient estimate of -0.354 and a P-value The predicted ranges of spread from the Honours Thesis, University of New of 0.012. This implies that in flood-prone OLS model, in metres per year, of certain England, Armidale. areas the spread of broom was 0.354 of a management and policy variables, are as Heinrich, A. and Dowling, B. (2000). Rare kilometre less than the average over the follows: and threatened plant survey of Bar- ten-year period. Rather than floods being rington Tops National Park: Plateau In ex State Forest land Average = 23 m responsible for spreading broom down Area. Unpublished report prepared for In NPWS land Average = -4 m watercourses, a possible explanation for the NSW National Parks and Wildlife the opposite sign is that the NPWS concen- Service, Gloucester. Without treatment Maximum = 17 m trated more management effort on flood Hobbs, R.J. (1991). Disturbance a precursor With treatment Minimum = -14 m prone areas to reduce the spread and also to weed invasion in native ecosystems. to prevent a reduction in water yield to For management purposes, it would Plant Protection Quarterly 6, 99-104. downstream users. be useful to say how much of the spread Hosking, J.R., Smith, J.M.B. and Shepp- could be reduced by another $1 spent on ard, A.W. (1998). Cytisus scoparius (L.) Discussion and conclusion broom control, but this is difficult. All that Link ssp. scoparius. In ‘The biology of In summary, the OLS analysis reported can be reported from these results is to Australian weeds’, Volume 2, eds F.D. above demonstrates that NPWS manage- say whether the control was effective or Panetta, R.H. Groves and R.C.H. Shep- ment activities, the effects of floods and not. Treatment stops the spread by 0.198 pard, pp. 77-88. (R.G. and F.J. Richard- the influence of the environmental factors of a kilometre on those sites where it was son, Melbourne). such as slope, crown land and natural veg- undertaken2. This result suggests that the Hosking, J.R. and Smith, J.M.B. (1999). etation density have a significant impact containment policy adopted by the NPWS ‘Broom Best Practice Management on the spread of Scotch broom in Bar- within Barrington Tops National Park has Guide’. (Cooperative Research Centre rington Tops National Park. The spread been effective, while any control measures for Weed Management Systems, Ad- is greatest on crown land, where there applied in the old State Forests land have elaide). has been limited treatment, on less steep not been effective. In this respect, the use Maddala, G.S. (1992). ‘Introduction to slopes and where the natural vegetation of the aerial maps has proven to be effec- econometrics’. (Macmillan, New York). is sparse. Pig activity also contributed to tive in isolating the factors impacting on Moodie, J.C. (1985). An evaluation of con- a greater spread in the initial sub-period. broom spread and could be usefully ap- trol techniques for broom (Cytisus sco- The spread is least with treatment, on steep plied to other weeds. parius) on Barrington Tops. B.Nat.Res. slopes, and where there have been floods. Given the low level of variation ex- Project Report, University of New In terms of their statistical levels, the most plained by all the models and the dif- England, Armidale. important factors are slope, floods, treat- ficulty of obtaining meaningful solutions McDonald, P. (1999). Scotch broom survey: ment in the wetlands, crown land, general from the Probit and Multinomial Logit Barrington Tops field work. Unpub- treatment and sparse vegetation. models, there are obviously other factors lished report prepared for the NSW In comparison to the average values influencing the spread of broom which National Parks and Wildlife Service, across all sites, the results reported in Ta- are not included in this analysis. Another Gloucester. ble 3 can be summarized as follows: approach may be to increase the size of McDonald, R.C., Isbell, R.F., Speight, J.G., • general treatment activities by the the sample to cover the whole area of the Walker, J. and Hopkins, M.S. (1990). NPWS reduced spread by 0.198 km broom infestation and then to redo the ‘Australian soil and land survey field over the 10 year period, although in analysis. It would be expected that the re- handbook’, 2nd edition. (Austral- the specific wetland areas the treatment sults would show some improvement in ian Collaborative Land Evaluation there reduced spread by 0.496 km, statistical properties and in the inclusion Program, Canberra). 10 Plant Protection Quarterly Vol.18(1) 2003 Odom, D.I., Cacho, O., Sinden, J.A. and Footnotes Griffith, G.R. (2001). Strategies for con- 1 An attempt was made to explore the trolling weeds in natural ecosystems: environmental and management factors a dynamic optimisation approach. that have influenced changes in both Australian Agricultural and Resource the spread and density of Scotch broom. Economics Society Conference, Ad- However meaningful results were not ob- elaide. tained for the density analysis due to the Odom, D.I., Griffith, G.R., Schroder, M. lack of changes in density across most of and Sinden, J.A. (2002). Using aerial the measurement points (only 14 of the 55 mapping to analyse factors affecting sites). The focus on the analysis reported the spread of Scotch broom. Working here is therefore on the spread of Scotch Paper, School of Economics, University broom. of New England. Pindyck, R.S. and Rubinfeld, D.L. (1998). 2 From the Probit model results (see Odom ‘Econometric models and economic et al. 2002), statements can also be made forecasts’, 3rd edition. (McGraw-Hill, about the probability of broom spreading New York). or contracting. For example, from the re- Rees, M. and Paynter, Q. (1997). Biologi- sults for the 1989-1993 period, with NPWS cal control of Scotch broom: modelling treatment the probability of broom spread- the determinants of abundance and the ing is reduced by about 40%. On the other potential impact of introduced insect hand, the probability of broom spreading herbivores. Journal of Applied Ecology is estimated to increase by about 4% on the 34, 1203-21. old State Forests land. Robertson, D.C., Morgan, J.W. and White, M. (1999). Use of fire to enhance control of English broom (Cytisus scoparius) in- A revised and shortened version of a pa- vading into a subalpine snowgum per presented to the 45th Annual Confer- (Eucalyptus pauciflora) woodland near ence of the Australian Agricultural and the Bogong High Plains, Victoria. Plant Resource Economics Society, Adelaide, Protection Quarterly 14, 51-6. January 2001. A longer version is availa- Schroder, M. and Howard, C. (2000). ble as Odom et al. (2002). We acknowledge Controlling broom (Cytisus scoparius) the help of other NSW National Parks and in natural ecosystems in Barrington Wildlife Service staff in providing us with Tops National Park. Plant Protection information and data for the analysis, in Quarterly 15, 169-72. particular Chris Howard for helpful dis- Smith, J.M.B. and Harlen, R.L. (1991). cussions during the initial stages of this Preliminary observation on the seed project and Cathy Ball for assistance with dynamics of broom (Cytisus scoparius) the map preparation. Two anonymous at Barrington Tops, New South Wales. referees provided very useful comments Plant Protection Quarterly 6, 73-8. on an earlier draft. Smith, J.M.B. (1994). The changing ecolog- ical impact of broom (Cytisus scoparius) This project was funded by the CRC for at Barrington Tops, New South Wales. Weed Management Systems. Plant Protection Quarterly 9, 6-11. Smith, J.M.B. (2000). An introduction to the biogeography and ecology of broom (Cytisus scoparius) in Australia. Plant Protection Quarterly 15, 140-4. Trudgeon, J. and Williams, J. (1989). Draft plan of management of Barrington Tops National Park, NSW. NSW National Parks and Wildlife Service, Raymond Terrace. Wapshere, T. and Corey, S. (1999). ‘Bio- logical control of Scotch broom Cyti- sus scoparius’. (Division of Entomology, CSIRO, Canberra). Waterhouse, B.M. (1986). Cytisus scoparius (broom) on Barrington Tops. Unpub- lished Honours thesis, University of New England, Armidale. Waterhouse, B.M. (1988). Broom (Cyti- sus scoparius) at Barrington Tops, New South Wales. Australian Geographical Studies 26, 239-48. Williams, P.A. (1981). Aspects of the ecol- ogy of broom (Cytisus scoparius) in Canterbury, New Zealand. New Zealand Journal of Botany 19, 31- 43.