Abundance and diversity of herbaceous weeds in sheep/beef pastures, South Island, New Zealand

Abstract

This study compared species diversity, abundance and size of broad-leaved herbaceous weeds on 28 South Island sheep/beef farms that employed either organic, integrated management (IM) or conventional management (CM) systems. Three or six paddocks per farm were surveyed using walked transects in November 2005, and the presence and number of individuals of each weed species encountered were recorded. 39 broad-leaved herbaceous weeds were recorded on all the farms in the study, but 76.7% of occurrences were of just three species (Californian thistle (Cirsium arvense L.), dandelion (Taraxacum officinale) Weber and daisy (Bellis perennis L.)). The ten next most abundant species made up 21% of records and the remaining 26 species just 1.9%. Very few significant differences were found in the geographic distribution, species richness or Shannon diversity index, abundance, cover or size of broad-leaved herbaceous weeds present on farms employing the different management techniques. However, there were significantly fewer Californian thistle per m² on CM than on organic or IM farms and cover of all herbaceous weeds averaged 5.0, 5.6 and 2.1 on organic, IM and CM farms, respectively. Weed infestation varied enormously by region and between individual farms, so the statistical power of the comparisons was relatively low. Until further research is reported, the authors caution against general and unquantified assertions that the sustainability of organic pastoral farming is, or is not, compromised by weed infestations.

Keywords:

• sustainable agriculture
• herbaceous weeds
• pasture
• organics
• integrated management
• conventional management
• sheep/beef farming systems

Introduction

The traditional forage species such as rye grass and clover make up the bulk of the productive biomass of improved sheep/beef pastures on the South Island, but these co-exist with a range of other plant species, some of which are considered to be weeds. Broad-leaved herbaceous weeds are generally acknowledged as a significant impediment to sheep and beef farming on the South Island, but there is a distinct lack of information available as to the true nature and extent of this problem. This paper defines weeds as non-forage broad-leaved herbaceous plant species present in the sward that are not grasses or species (such as clover) that are specifically planted as animal feed or as part of farm management actions. The definition includes species such as thistles and the ‘flat weeds’ from the family Asteraceae (dandelion, daisy, plantain), and it is these species that are most frequently of concern to farmers (Van Toor & Stuck 1993; Popay et al. 2002). Relatively unproductive grass weeds were not included in this study, as the authors considered that their potential impact by lowering animal production is overshadowed by the complete displacement (and zero productivity) of productive grasses by broad-leaved weed species.

The little information available about herbaceous weeds in New Zealand comes either from small paddock-scale studies (for examples see Wardle et al. 1995; Edwards et al. 2005; Seefeldt et al. 2005) or from larger low-resolution ad hoc distribution studies (Bourdôt et al. 2007). Efforts at full costing of weeds on New Zealand farming have been very informal or approximate (Hackwell & Bertram 1999). There have been estimates of cover taken by weeds (Bascand & Jowett 1982; Bourdôt & Kelly 1986) and surveys of stakeholders’ perceptions of weeds (Popay et al. 2002). Hackwell & Bertram (1999) estimated the economic losses attributable to weeds in New Zealand (including to all primary production systems and the natural estate) at $440 million annually. To date, there have been no replicated farm-scale studies of broad-leaved herbaceous weeds or the factors affecting their prevalence and impacts, and there are no reliable estimates of how weeds limit the economic and environmental performance of sheep and beef farms specifically.

Weeds can adversely affect farming operations and farm viability by: reducing the amount of pasture biomass available for stock, thus affecting productivity; imposing major time and financial costs on farmers attempting to control weeds to maintain productivity; and posing significant health risks to stock, such as the case of cattle poisoning by ragwort (Senecio jacobea L.) (Wardle 1987). Alternatively, some broad-leaved species may have beneficial nutrition or animal health properties and may be tolerated or actively encouraged by farmers (Wilman & Derrick 1994; Daly et al. 1996; Sanderson et al. 2003). These species may also lead to increased biodiversity on farms, with potential benefits for increased socio-ecological resilience for the farming operation (Matson et al. 1997; Fischer et al. 2006). Consequently, the prevalence and impacts of broad-leaved herbaceous species on pastoral farming operations are of interest to farmers, agricultural and environmental scientists and policy-makers alike.

Pastoral agriculture is evolving so that land-use systems achieve more appropriate and enduring accommodations with the New Zealand environment, as well as continuing to satisfy the demands of market and community stakeholders (Popay et al. 2002; Bourdôt et al. 2007). Managing weeds to minimise economic and environmental damage is critical to achieving sustainable agriculture because weeds can dictate land-use choices, limit productivity and limit profitability. Furthermore, management techniques can be energy intensive, labour intensive and/or increase pesticide load in the environment.

The two major potential pathways to more sustainable agriculture currently employed in New Zealand are certified organic production and integrated management (IM) supermarket-based accreditation schemes (Wharfe & Manhire 2003, Campbell et al. 2006). Weeds are often considered a significant barrier to sustainable organic production and there is the perception that weed numbers or sizes are harder to control using organic techniques (Mackay et al. 2002). Consequently, information is urgently required on the relative environmental, economic and social attraction of or limitations to these management systems. Specifically, reliable information on weeds is required at the whole-farm scale, as these are the key sites of management actions and outcomes for New Zealand farmers and agriculture (Darnhofer et al. 2010).

This work hypothesised that the management system of a farm will have a significant impact on broad-leaved weed communities. With this in mind, the specific aims of this study were to compare:

1.	percentage coverage and species richness of broad-leaved weeds on farms with organic, IM and conventional management (CM)
2.	the abundance of broad-leaved weeds on farms under the three management systems
3.	the size of individual plants of the most common weeds on organic, IM and CM farms.

This focus on broad-leaved herbaceous weed abundance and cover in pastures forms part of wider studies undertaken by the Agriculture Research Group on Sustainability (ARGOS 2010) that investigates the environmental, social and economic outcomes of different market accreditation systems as complementary pathways to more sustainable and resilient agriculture (Rosin & Campbell 2008, Campbell et al. 2009, Carey et al. 2009, Fairweather et al. 2009, Darnhofer et al. 2010). This study of herbaceous weeds is part of a wider investigation of weed management options and techniques, pathways for invasion of woody weeds into agricultural land and assessment of future weed management research needs.

Materials and methods

Ten of ARGOS's 12 clusters of sheep and beef farms from eastern South Island were sampled (Fig. 1). Each cluster consisted of three farms within 25 km of each another that were matched for altitude, rainfall and soil type. Each cluster contained a CM farm (no accreditation scheme), a certified organic farm and an IM farm.

Figure 1  Locations of the sheep/beef farms sampled in the current study. Cluster locations: 1, Marlborough; 2, Amberley; 3, Banks Peninsula; 5, Methven; 7, Fairlie; 8, Outram; 9, Owaka; 10, Gore; 11, Oamaru; 12, Waimate. With the exception of cluster 12, each cluster consists of an organic, IM and CM farm. Cluster 12 also includes a farm undergoing organic conversion at the time of the survey. ARGOS clusters 4 (Leeston) and 6 (Ashburton) were not include in the weed study.

Timed paddock walks

The ‘focal paddocks’ selected for this study were those used for soil sampling in July and August 2005 (see the works of Pearson et al. (2005) and Carey et al. (2010) for details). Three landforms (flat, slope and hill-crest) were recognised as being potentially important explanatory variables for soil quality and pasture ecology because of the way nutrients are expected to flow within farm landscapes. For clusters with only one landform, three paddocks were sampled per farm. For clusters with more than one landform, three paddocks were sampled per farm in each of the two most common landforms for the cluster as a whole (giving a total of six paddocks per farm). This design ensured a balanced design within the clusters to maximise the power to detect differences in sustainability indicators between the three ARGOS panels (i.e. organic, IM and CM farms).

Estimation of weed abundance

The relative abundance of broad-leaved herbaceous plants in each focal paddock was calculated from the number of individuals of each species counted along a walking route starting from a pre-selected random start point. This start point was found using a Garmin eTrex global positioning system (GPS) (Garmin Ltd, USA). The direction of travel from the start point was determined by twirling the direction setting on a Silva compass for 10 s and walking in the direction of the compass arrow. The direction of the walk changed when a fence or land feature prevented continuation of travel in that direction. In such cases, the angle of reflection away from the feature was the same as the angle of incidence. The finishing point was determined by a digital timer set for the appropriate paddock size and survey duration and its location recorded by GPS. A geographic information system (GIS) map of each farm was used to calculated the total length of transect walked in each paddock.

Individuals of all broad-leaved herbaceous plants that were encountered in a corridor 1 m either side of the walking route were counted and identified. Their common names, scientific names and authorities are listed in Table 1. Any plant that could not be named was collected for later identification. The duration of the walk varied with paddock size: paddocks up to and including 5 ha were surveyed for 10 min; paddocks between 5 and 10 ha were surveyed for 15 min; paddocks greater than 10 ha were surveyed for 20 min. Data were recorded using a dictaphone and later transcribed. To check for edge effects, individuals within 5 and 10 m of a fence were recorded. Preliminary analyses found no differences in weed abundance or species richness in edge habitats, so all data were pooled for subsequent analysis. Provided that weeds are relatively randomly distributed around paddocks, robust abundance estimates should have been obtained by this practical and efficient timed-transect sampling approach.

Table 1  The common and scientific names of all herbaceous species recorded on organic, IM and CM farms in the study. Also shown are the total numbers of plants of each species recorded on all farms in the paddock walk survey.

Species		Scientific name	Number of plants
1	Californian thistle	Cirsium arvense L.	66960
2	Dandelion	Taraxacum officinale Weber	35341
3	Daisy	Bellis perennis L.	22720
4	Nodding thistle	Carduus nutans L.	6377
5	Winged thistle	Carduus tenuiflorus Curtis	5903
6	Plantain	Plantago lanceolata L.	5010
7	Scotch thistle	Cirsium vulgare (Savi) Ten.	4346
8	Buttercup	Ranunculus repens L.	3398
9	Dock	Rumex obtusifolius L.	3235
10	Storksbill	Erodium cicutarium L.	2840
11	Rushes	Juncus spp.	1679
12	Shepherds purse	Capsella bursa-pastoris L.	1601
13	Hedge mustard	Sisymbrium officinale L.	474
14	Patterson's curse	Echium vulgare L.	445
15	Stinging nettle	Urtica urens L.	339
16	Willow weed	Polygonum persicaria L.	338
17	Orange hawkweed	Hieracium aurantiacum L.	308
18	Chickweed	Stellaria media L.	295
19	Unknown		271
20	Dwarf mallow	Malva neglecta.Wallr	189
21	Foxglove	Digitalis purpurea L.	167
22	Fathen	Chenopodium album L.	154
23	Doves foot	Geranium molle L.	125
24	Piripiri	Acaena novae-zealandiae Kirk	104
25	Cutty grass	Carex lessoniana Steud.	82
26	Fumitory	Fumaria muralis Sond.	75
27	Native flax	Phormium tenax J.R Forst. & G. Forst.	65
28	Staggerweed	Stachys arvensis L.	52
29	Wireweed	Polygonum aviculare L.	31
30	Redroot	Amaranthus retroflexus L.	26
31	Speedwell	Veronica persica Poir.	14
32	Horehound	Marrubium vulgare L.	12
33	Variegated thistle	Silybum marianum L.	12
34	Yarrow	Achillea millefolium L.	5
35	Ragwort	Senecio jacobaea L.	3
36	Burdock	Arctium minus Bernh.	3
37	Mushrooms	Basidomycota	2
38	Plumeless thistle	Carduus acanthoides L.	1
39	Stinking mayweed	Anthemis cotula L.	1

Species diversity on transects

Three measures of species diversity were calculated for each farm. First, the number of different weed species encountered (species richness) was calculated for each paddock and then averaged for each farm. Second, the total number of different species encountered on all transects on each farm was scaled against the total area of paddock searched. Third, the Shannon Diversity Index (SDI), which combines information on both number of species and distribution of individuals among those species, was calculated for each paddock and then farm, using the formula (Begon et al. 2006):

where P _i is the proportion of total individuals in each sample in species i.

Measurement of plant size

Individual plant size was estimated as the area of a circle of which the length of the longest leaf in a rosette formed the radius. The longest leaf was measured on the first 20 plants of each species encountered on each transect (following Bourdôt & Kelly (1986)).

Pasture composition

A circular 37 cm diameter quadrat was dropped over the observer's shoulder at the start point and repeatedly at 5 min intervals while moving along transects. Pasture that was flattened was teased from the internal edge of the quadrat. Pasture cover was visually assessed into the following categories: grasses; broad-leaved weeds; other herbaceous plants; clovers. Assessments were made by eye on a percentage of cover basis. Average percentage cover estimates for the most common species were estimated from the paddock walk data and were calculated as:

where A _p is the average individual plant area, N is the number of individuals on a transect and A is the area surveyed. The individual plant area was given by Π×(longest leaf length)² and area sampled (m²) was given by (transect length)×2.

Altogether, 41, 43 and 16% of the quadrats fell on flat, slope and hill-crests landforms, respectively, so most of the weed information gathered is for flat and sloping areas.

Statistical analysis

Due to the variable sampling efforts in differently sized paddocks, the weed abundance data were standardised prior to analysis to allow comparison. The data could be standardised using either the time or the length of the walk. As the exact speed of the observer was affected by the terrain of the paddock and the abundance of weeds (more species and/or individuals to record resulting in a slower speed), parameters for abundance of weeds were standardised by the length of the transect and are expressed as weeds/m².

Differences in weed abundance and weed plant size were analysed using residual maximum likelihood (REML) routines in GenStat 9th edition (Payne et al. 2006) to accommodate the unbalanced nature of the data. Geographical location is likely to have a strong influence on weed species presence and abundance. Consequently a randomised blocked design was used, with individual farms nested within each cluster and with each management system—organic, IM and CM (termed ‘panels’). Of the 12 clusters in the ARGOS study, data could not be collected from two of them (Leeston and Ashburton) and from one individual farm in two other clusters (Fairlie and Waimate). This left a total of 28 properties with weed data, and eight clusters for which the full nested analysis could be conducted. All tests for panel effects on weed abundance and size were conducted only on the 24 farms in complete clusters, but the estimates of overall prevalence of different weeds and species richness include information from all 28 farms.

An attempt was made to build REML models that included landform as an explanatory variable. However, they were unstable in nearly all cases because of unbalanced and small sample sizes resulting from the patchy distribution of most weed species between clusters. However, the selection of the same number of focal paddocks in equivalent landforms at all farms within each cluster means that comparison of weed measures between panels will not be confounded. We also cross-checked for the potential importance of landform as an explanatory variable by calculating a single average percentage cover of all weeds in circular quadrats for each landform in each property (data were aggregated at farm level to avoid pseudo-replication). A Kruskal–Wallis non-parametric test found no evidence of landform affecting overall weed prevalence (H=0.68, df = 2, p=0.71). Landform was this ignored as a potential explanatory variable. However, readers are cautioned not to interpret the averages presented for each farm as representative of weed prevalence and abundance for that entire farm (estimates of the proportion of the farm area in each landform would have been required to take that next step).

The data for the abundance of individual weed species were highly positively skewed and too sparse for most species to analyse. There were, however, enough records to test statistical models for eight species, named here ‘common’ species (buttercup, Californian thistle, daisy, dandelion, dock, nodding thistle, plantain and Scotch thistle). In most cases data were log₁₀(x) or log₁₀(x+1) transformed for analysis, but other transformations were sometimes employed where the logarithmic transformations were inadequate (detailed in each case where results are presented). The remaining species were not recorded frequently enough to analyse parametrically, so for these species the non-parametric Friedman's randomised block analysis of variance (Anova) or Kruskal–Wallis tests were used (Payne et al. 2006).

Biogeography theory predicts that the number of different species recorded (species richness) will increase rapidly at the start of each walk and then curve towards an asymptote as successively greater areas have been searched on each farm (Begon et al. 2006: p. 613). REML models were thus built to predict the total number of different weed species found on a farm from the total area searched there. The effect of panel on the species–area relationship after the effects of cluster had been excluded was then tested.

Results

Occurrence of weed species

A total of 39 different broad-leaved herbaceous species were recorded during the surveys. By far the most abundant species on the farms were Californian thistle, dandelion and daisy, with over 66, 35 and 22 thousand individual plants of each species respectively recorded (Table 1). Collectively, these three species constitute 76.7% of the records, with the ten next most abundant species comprising 21.4% of the records and the remaining 26 species making up just 1.9% of the records. Californian thistle occurred on 27 of the 28 farms with data (96.4%), dandelion occurred on all 28 farms, while daisy occurred on 17 farms (60.7%). Other frequently occurring species were Scotch thistle (92.9% of farms), dock (82.1%), plantain (67.9%), buttercup and rushes (64.3%), chickweed (60.7%), nodding thistle (46.4%), shepherd's purse (39.2%) and storksbill (39.2%).

There were 24 species that occurred on fewer than five of the study farms. Of the 15 species that were found only on one farm, seven were found on organic farms (burdock, fumitory, hieracium, mushroom species, piripiri, redroot and willow weed), four on IM farms (cutty grass, geranium, Paterson's curse and staggerweed) and four on CM farms (chamomile, hawkweed, horehound and plumeless thistle). Traditional forage species such as ryegrass and clover dominated the pastures (Fig. 2), but weeds were widespread within farms (i.e. the median percentage occurrence of other herbaceous plants was 20% for all the circular quadrats (25% for organic farms, 14.3% for IM farms and 15 for CM farms). A Kruskal–Wallis test accepts a null hypothesis that these medians are the same between panels (H=2.55, df = 2, p=0.28).

Figure 2  Average percentage cover of grass, clover and herbaceous species in the sward of organic, IM and CM farms in the study. Error bars are binomial 95% confidence intervals.

Percentage cover of weeds

Using data from the timed walk survey, Californian and nodding thistles had the greatest percentage cover, although the total amounts were still small (less than 6% of the sward (see Table 2)). On average, all thistles combined constituted 4.3% of the sward on organic farms, 4.7% on IM farms and 2% on CM farms. These differences between panels border on being formally statistically different (p=0.067; Table 2). Coverage of buttercup was low on all farms, while dock made up 1% of the organic sward, compared with 1.6% on IM and 0.4% on CM farms. When all species were combined, the common weeds constituted a mean percent coverage of 5.0, 5.6 and 2.2 on organic, IM and CM farms, respectively (Table 2). These differences are very nearly statistically significant (p=0.054) despite the very high coefficients of variation recorded between farms (Table 2).

Table 2  Transect estimates of percentage ground cover, standard error and% CV of common herbaceous weeds of management concern on organic, IM and CM sheep/beef farms in the study. Geographic location was included as a blocking term in the model for Californian thistle, Scotch thistle, all thistles combined and all weeds listed combined, but could not be included in the analyses for winged thistle, buttercup or dock due to a scarcity of data.

Species	Mean	Standard error	CV: %	F	P
Californian thistle
Organic	3.85	1.33	91.68	F 2,9=2.47	0.140
IM	4.04	1.42	86.14
CM	1.81	0.91	123.02
Nodding thistle
Organic	0.37	0.16	74.11	F 2,5=1.05	0.415
IM	5.20	5.03	136.76
CM	1.93	1.64	146.99
Scotch thistle
Organic	0.37	0.07	39.45	F 2,8=0.14	0.868
IM	0.73	0.78	138.57
CM	0.41	0.17	114.92
Winged thistle
Organic	0.01	0.01	103.85	F 2,4=4.62	0.091
IM	0.77	0.40	90.12
CM	0.07	0.05	107.97
All thistles
Organic	4.28	1.42	87.83	F 2,14=3.29	0.067
IM	4.68	2.12	136.16
CM	2.01	0.86	134.87
Buttercup
Organic	0.020	0.0003	2.03	F 2,4=0.32	0.742
IM	0.008	0.0030	46.19
CM	0.060	0.0600	149.33
Dock
Organic	1.00	0.64	142.27	F 2,11=0.33	0.726
IM	1.64	1.53	207.60
CM	0.36	0.18	99.34
All common species
Organic	5.00	1.55	81.99	F 2,14=3.61	0.054
IM	5.59	2.14	114.74
CM	2.17	0.95	138.74

There were some significant differences in weed coverage between the geographic clusters. Californian thistle constituted significantly more of the sward in the southern clusters than those further north, with the highest coverage (mean±standard error) in Catlins (7.63±1.16%) and the lowest on Banks Peninsula (0.13±0.06%). Nodding thistle covered significantly more ground in Oamaru (5.37±2.75%) than in either Fairlie (0.18±0.02%) or Waimate (0.33±0.05%), while Scotch thistle percentage cover was highest in the Catlins (1.56±0.56%) and Banks Peninsula (1.07±0.74%) and lowest in Methven (0.13±0.00%) and Gore (0.17±0.06%).

Weed species diversity

The average number (± standard error) of weed species per paddock across all farms was 6.14 (±0.19), and ranged from 2 to 12 species per paddock. REML model fits to predict species richness were best for log₁₀(number of species) versus log₁₀(area searched). Conclusions about area and panel effects were the same whether or not two outliers were excluded. As expected, more species were found on farms where the area searched was greater (p=0.06), but there was no evidence that any of the panels had higher weed species richness for a given area searched (p=0.19).

Species diversity, as expressed by the SDI, differed significantly between the clusters (F _7,73=2.45, P=0.03). It was significantly higher on Banks Peninsula than Outram, the Catlins, Gore or Oamaru, and it was also significantly higher in Fairlie than Gore and Oamaru. The average SDI (±standard error) was 0.77±0.07 on organic farms, 0.84±0.07 on IM farms and 0.93±0.07 on CM farms. Once the differences between clusters were controlled for, there was no evidence of difference in the average species diversity of all species on organic, IM or CM farms (F _2,73=2.24, P≥0.05).

Abundance of weeds

The average number of individual plants/m² of all species differed significantly between the clusters (log₁₀ transformation: F _9,16=4.86, P=0.002), with the highest abundance recorded in the Catlins and the lowest in Fairlie (Fig. 3). The abundance in the Catlins was significantly higher than in Amberley, Banks Peninsula, Fairlie and Waimate. The average abundance of plants (± standard error) on organic farms in the study was 1.11 (±0.32)/m², compared with 1.08 (±0.25)/m² on IM farms and 0.97 (±0.22)/m² on CM farms. Once the differences between clusters were controlled for, there was no evidence of a difference in the average abundance of all species on organic, IM or CM farms (Fig. 4).

Figure 3  Average abundance of weeds of all species per square metre of paddock (± standard error) for each sheep/beef property. The locations of the clusters are shown in Figure 1. The management system on each farm is indicated by the letter following the cluster code: A, organic; B, integrated management; C, conventional management.

Figure 4  Average number of plants (±standard error) of all herbaceous species recorded per metre of the transect on organic, IM and CM farms in the study. The data were significantly positively skewed so a fourth- root transformation was performed prior to the analysis. There was no significant difference between panels (F _2,16=0.26, P>0.05).

The abundance of each common species was compared between clusters and management systems (Table 3). Overall abundance was significantly different between clusters for Californian thistle (F _9,16=12.29, P<0.001) and was higher on the more southern clusters (Outram, Gore, Catlins and Oamaru) and lower in the northern clusters (Marlborough, Amberley, Banks Peninsula and Fairlie). Management system had a significant effect on abundance only for Californian thistle, with the number of plants/m² significantly lower on CM farms than on either organic or IM farms (Table 3).

Table 3  The mean abundance (individual plants/m²), standard error and coefficient of variation for common weeds on organic, IM and CM sheep/beef farms. Test statistics are from Friedman's non-parametric randomised block analysis of variance (F _s) in all cases, except for Californian thistle and total weeds of management concern, where an unbalanced analysis of variance was used. Data were log10(x+1) transformed for Californian thistle and total weeds of management concern to meet assumptions of normality for the unbalanced analysis of variance.

Species	Mean	Standard error	CV: %	Test statistic	P
Buttercup
Organic	0.02	0.01	169.52	F s=1.69	0.407
IM	0.01	0.01	192.81
CM	0.02	0.02	224.31
Californian thistle
Organic	0.53	0.21	120.81	F 2,16=3.68	0.048
IM	0.46	0.18	116.30
CM	0.23	0.14	193.65
Daisy
Organic	0.07	0.04	186.49	F s=0.341	0.341
IM	0.17	0.09	154.18
CM	0.17	0.09	168.05
Dandelion
Organic	0.21	0.07	104.04	F s=1.75	0.417
IM	0.10	0.04	124.59
CM	0.37	0.21	180.83
Dock
Organic	0.02	0.01	114.24	F s=1.75	0.417
IM	0.06	0.06	276.72
CM	0.01	0.01	172.01
Nodding thistle
Organic	0.01	0.01	173.00	F s=1.19	0.282
IM	0.06	0.06	293.33
CM	0.04	0.03	257.14
Plantain
Organic	0.03	0.03	280.45	F s=4.0	0.069
IM	0.03	0.01	235.59
CM	0.05	0.02	168.79
Scotch thistle
Organic	0.02	0.01	115.25	F s=1.75	0.417
IM	0.03	0.02	160.17
CM	0.02	0.01	67.33
All common species
Organic	0.91	0.31	102.41	F 2,16=0.16	0.851
IM	0.90	0.27	88.84
CM	0.91	0.23	79.99

Size of common weeds

Eight species were recorded frequently enough to compare weed size on organic, IM and CM farms (Table 4). Sizes of individual plants of most species were highly variable within and between management systems, and average size differed significantly only for dandelions, which were significantly smaller on IM farms than on either organic or CM farms.

Table 4  Summary statistics for weed size on organic, IM and CM farms in the study. Shown is the mean length of the largest leaf, standard error of the mean and percentage coefficient of variation. F-statistics are shown from an unbalanced analysis of variance, where mean length was the response variable and management system was the treatment variable. Californian thistle, dandelion and Scotch thistle were the only species with enough records to include cluster in the model as a blocking variable; for all other species only management system was included in the model. Californian thistle, daisy, dock and plantain were analysed using untransformed data, while dandelion was analysed using a fourth-root transformation ((x+1)^0.25), Scotch thistle was analysed using a square-root transformation ((x+1)^0.5), and nodding thistle and winged thistle were analysed using a log₁₀(x+1) transformation.

Species	Mean	Standard error	CV: %	F	P
Buttercup
Organic	3.52	0.44	17.80	F 2,4=2.79	0.175
IM	2.14	0.14	9.43
CM	2.94	0.40	23.41
Californian thistle
Organic	10.98	0.90	21.77	F 2,9=0.26	0.774
IM	10.85	0.61	13.84
CM	11.21	0.40	8.69
Daisy
Organic	1.98	0.48	34.01	F 2,4=2.72	0.180
IM	2.00	0.00	0.00
CM	1.25	0.25	28.28
Dandelion
Organic	9.00^a	0.85	26.70	F 2,12=7.24	0.009
IM	5.76^b	0.69	31.71
CM	8.58^a	0.78	27.23
Dock
Organic	17.72	2.92	36.79	F 2,11=1.44	0.278
IM	14.38	1.59	24.79
CM	11.40	3.18	55.74
Nodding thistle
Organic	20.15	1.27	10.92	F 2,5=0.10	0.910
IM	21.47	0.65	4.30
CM	24.45	7.41	52.50
Plantain
Organic	9.00	0.00	0.00	F 2,4=0.20	0.827
IM	8.78	1.78	28.61
CM	6.91	2.21	64.10
Scotch thistle
Organic	20.05	2.25	25.07	F 2,8=1.02	0.403
IM	17.01	1.57	24.45
CM	19.70	2.00	28.75
Winged thistle
Organic	10.94	1.06	13.74	F 2,4=0.05	0.950
IM	10.56	3.47	56.97
CM	10.71	4.29	56.72

Size variability within each farm's weed population was evaluated by comparing variance in weed sizes between management systems for each species. The variance in weed size was very nearly statistically different between panels for both buttercup (average variance±standard error: organic, 2.21±0.84; IM, 0.35±0.35; CM, 0.39±0.13; F _2,4=6.02, P=0.06) and dandelion (organic, 19.48±8.07; IM, 3.34±1.09; CM, 7.69±2.11; F _2,4=3.34, P=0.07). Variance in size did not differ significantly for any other species. Nor was there evidence of bimodal size distribution for the two biennial species (nodding thistle, Scotch thistle).

Discussion

Non-forage broad-leaved herbaceous plants, many of which are considered to be weeds, were recorded on all farms in the study. However, the term ‘weed’ has been used here as conventional shorthand throughout this paper while placing no specific positive or negative value on them. Certainly the majority of the non-forage herbaceous species focused on in this study are sources of concern and targets of active management by most farmers. However, this study emphasises the fact that weeds are also a significant source of floristic diversity on sheep/beef farms of South Island (Table 1), and they may in turn support considerable invertebrate diversity and ecosystem services on farms.

As well as their tremendous variety and widespread distributions, the other overarching conclusion from this study is that weed prevalence is highly variable between individual sheep/beef farms and geographic locations within South Island. The coefficient of variation (CV) of weed measures was often greater than 100% (Table 2 and 3).

It is interesting that mouse-ear hawkweed or hieracium (Hieracium pilosella) was not listed as a weed of concern on any of the farms, even though it is a serious weed of pastoral agriculture in some parts of the South Island (Hunter et al. 1992; Duncan et al. 1997). This reinforces the fact that the results of this survey cannot easily be generalised to other farming systems in the South Island, let alone North Island sheep/beef farms.

A general lack of differences between farming systems

Overall, very few significant differences were found in the distribution, abundance, diversity and size of herbaceous plants present on farms employing organic, IM or CM techniques. The only common species that differed in abundance between panels was Californian thistle (Table 3). The reason for this difference is unclear, but may relate to active targeting of thistles by CM farmers or ineffective management options for organic and IM farmers. Control options for Californian thistle include mechanical topping or mowing in areas where machinery can gain access (Knowler 2008) and either broad-scale or targeted herbicide application. There are currently no successful biocontrol agents for Californian thistle in New Zealand. The finding that the lowest densities of this species were recorded on CM farms may reflect the herbicide control options open to these farmers, which are absent for organic farmers and reduced for IM farmers.

Although the test for panel differences in abundance of all thistles combined was not quite formally significant (Table 2), it can nevertheless be concluded that a real difference does occur because Californian thistle (the most abundant thistle species by an order of magnitude) was significantly lower on CM farms (Table 3). These differences suggest that weed prevalence is marginally lower on CM farms than in either organic or IM farms, and that there is no evidence for weed prevalence being different between organic and IM farms.

Significant differences in the average size of individual weed plants (Table 4) were not detected. Weed size is an indicator of their vigour and/or the ability of farmers to control them. The only significant difference detected was for dandelions, which were smaller on IM farms than on organic or CM farms. However, there was higher variation in the size of buttercup and dandelions on organic farms than on their IM and CM counterparts, suggesting that the different weed control methods used by organic growers in some way trigger a change in these weed populations’ age or size structure.

There was no evidence for a higher proportion of circular quadrats on organic farms to have at least one weed species present. This suggests that weed distribution within farms is similar between panels.

Unfortunately, the high variability in weed prevalence between farms and clusters greatly reduces the power of this study to detect any significant differences between panels. The cluster design and matching of landforms and other ecological conditions between focal paddocks within each cluster will have minimised unexplained variation in order to better test for panel effects. However, it cannot definitely be concluded that weed prevalence, cover or size do not differ on average for organic, IM or CM farm management practices. This work can only conclude that, so far, there is no evidence for many panel effects, despite considerable sampling efforts.

Care is also needed in assuming that any observed differences (or similarity) in weed prevalence between panels are caused by farm management itself. ARGOS's design is ‘quasi-experimental’ in that farming families choose whether or not to convert their land to organic or IM production (rather than researchers randomly assigning land and families to each panel in a truly experimental way). If the nature of the land (in this case prior weed prevalence) or the philosophy of the farmer (in this case their attitude to ‘weeds’) affects their decision about whether or not to convert, then current weed prevalence and management may have little to do with the act of accrediting their farm practices or subsequent rules for weed management. It may just reflect the way farmers and their land were before conversion. It may even be that organic growers deliberately chose land that had low weed prevalence or was known to be less vulnerable to weed infestation. Also, weed prevalence may have been gradually rising since conversion to now be about equivalent to that in integrated or conventionally managed land nearby. However, we believe that these potential alternative interpretations are unlikely in this study because herbaceous weed dynamics are labile and rapid. We would expect rapid change in weed prevalence, abundance and size between panels soon after conversion if the weed management methods authorised by each farming system differed greatly in effectiveness or cost. It is more likely that the lack of evidence for differences in weed prevalence, abundance or size genuinely indicates that weed management success is relatively unaffected by farming system practice.

Variation in weeds between clusters and farms within panels

Compared with the lack of statistically significant differences between farm management systems, this study found strong differences in weed abundance and cover at different farm locations. Further study is needed to confirm what causes such differences, but higher rainfall and greater variability in topographic relief, soil types or soil quality may all be involved. Some cluster differences may also have been driven by different proportions of landforms in the different clusters. These results also highlight the value of the ARGOS experimental design, where local physio-chemical and climatic factors can be explicitly considered and controlled for, thus leading to greater power to detect differences between management systems where they exist.

Future research may be more valuable if it focuses on what enables one farm or farming family to effectively manage weeds while another nearby struggles, rather than searching for average differences between farming systems. It may also help to understand attitudes of farmers to weeds as part of a broader orientation towards ‘tidy farms’ and their perceptions of what makes a ‘good farmer’ (Hunt 2010) because this will affect their investment in weed management and their resistance or willingness to change weed management in ways that promote resilience and sustainability.

National weed management priorities

According to ARGOS (unpublished data), farmers recorded their active management of 22 ‘pastoral weeds’ on the study farms in the past 5 years whereas the current survey identified 39 species of herbaceous weeds in their pastures. However, just a few species dominated the records of control events—the woody weeds (especially gorse and broom), the perennial Californian thistle and annual/biennial nodding and Scotch thistles. Van Toor and Stuck (1993) and Popay et al. (2002) also found that farmers were preoccupied with these species; Bascand and Jowett (1982) and Cockayne (1917) note that these species were dominant weeds for South Island pastoral farmers nearly a century ago. This raises three points. First, despite consistent weed management effort over many decades, there has not been significant long-term widespread change in the pest status of these major weeds. Second, over this period, no other species have increased in importance to supplant these weeds in the consciousness of sheep/beef farmers in the South Island. Third, while Californian thistle specifically, and thistles in general, were the herbaceous species most frequently mentioned by farmers (ARGOS, unpublished data), dandelion and daisy were more abundant than nodding thistle and plantain was more abundant than Scotch thistle in the current survey. These three non-thistle species are all less immediately visible than the thistles and, in the cases of dandelion and plantain, may be viewed as beneficial species by some farmers. Four of the 12 organic farmers surveyed felt that herbaceous species could prove beneficial to their farming operation, compared with only one of the 20 IM and CM farmers in the study. These included several organic farmers who placed value on thistles for providing fodder in dry periods and on Californian thistle in particular for conditioning the soil (ARGOS, unpublished data). Clearly, a more nuanced understanding of the prevalence, abundance and impacts of herbaceous species on South Island sheep/beef farms is required, and the authors advise caution against grouping all herbaceous species simply as ‘weeds’.

Lack of evidence for increased weed prevalence and vigour on organic sheep/beef farms contradicts a popular assertion that weed infestations are a major barrier to sustainable organic production (Mackay et al. 2002, Campbell et al. 2009). This study and other ARGOS data suggest that organic farmers spend more time but less money on controlling weeds than their IM and CM counterparts. The current study suggests that weed prevalence and its potential impact on food and fibre production of the farm are very similar between farming panels.

Obviously more and larger studies are needed in a variety of farming sectors to increase the statistical power and generalisability of weed comparisons between farming systems. In the meantime, unquantified generalisations that have asserted that organic agricultural sustainability is compromised by the force of weed infestations (Mackay et al. 2002) are to be taken with caution. Equally, so far, there is only tentative evidence from this study that weeds are not a more significant challenge to sustainable sheep/beef farming on organic farms.

Acknowledgements

The authors thank all the participating ARGOS farmers who granted access to their farms for this study. This work was funded by a Ministry of Agriculture and Forestry Sustainable Farming Fund grant and the Foundation for Research, Science and Technology (contract number AGRB0301) with additional assistance from Canterbury Meat Packers. We are grateful for the helpful comments from two initial referees and especially to a very perceptive biometrican whom the journal asked to provide further comments.