Effects of landscape composition on spread of an herbicide-resistant weed
Widespread adoption of genetically modified glyphosate-resistant (GR) crops in the US has dramatically changed the agricultural landscape to one that selects for establishment and spread of weedy species resistant to glyphosate, a commonly applied herbicide. Weed species that possess the means to readily spread across the landscape will be contained by weed management strategies that limit weed establishment and prevent seed set. An empirically-derived simulation model was developed to explore GR Conyza canadensis spread in relation to characteristics of the agricultural landscape. C. canadensis seeds are carried in the wind and move among fields and therefore, access high quality habitat (GR crops) at long distances. The baseline scenario was the current GR adoption levels in many US agricultural landscapes with corn and soybean rotated annually. Alternate scenarios examined the interacting effects of management uniformity (GR crop adoption) and increased landscape richness (three crops: corn, soybean, alfalfa, instead of two), over a 10 year simulation period. When landscape uniformity increased (increased GR corn adoption), 3× more fields would be infested with the resistant biotype and a specific field would have up to 24% greater likelihood of being infested compared to the current GR crop adoption levels. Increased landscape richness (adding alfalfa as a third crop) slightly decreased GR C. canadensis abundance. Reduced GR management uniformity by way of reducing GR soybeans to half their current adoption levels had the greatest impact on spread and prevented GR C. canadensis from reaching high abundance. Large-scale reliance on glyphosate for weed management has increased high-fitness habitat and will result in rapid spread of glyphosate-resistant weeds. Without significant reductions of glyphosate use and without spatial coordination of weed and crop management practices, GR weeds will continue to spread rapidly and impact agricultural practices in areas reliant on glyphosate.
KeywordsConyza canadensis Dispersal Erigeron canadensis Genetically-modified crops Horseweed Resistance management
In the last 10 years, the diversity of weed management practices in US agroecosystems has declined precipitously and resulted in a massive increase in habitat suitable for weedy plants adapted to the new management. Instead of within field dynamics, the wholesale adoption of practices raises an important question: will the change in agricultural practices at landscape-level scales affect the spread of the adapted organisms?
Spread at this level depends on landscape processes of composition, distance among suitable sites, and connectivity (Dunning et al. 1992; Taylor et al. 1993; Margosian et al. 2009). One example is how increased distances between suitable sites and concurrent habitat conversion can potentially interfere with spread of forb species among sites (Soons and Ozinga 2005). In contrast, increasing area of suitable habitat facilitates spread of invasive genotypes and species (Mortensen and Jongejans 2006; Hamilton et al. 2006). We sought to isolate the change in agricultural landscape composition by developing a spatially explicit model that reduces the effects of distance among sites and connectivity. Distance among suitable sites was reduced as a cause of spread through model construction (see “Simulating spread”).
Connectivity effects were equalized among our tested scenarios due to the biophysical properties of the plant species being studied. Conyza canadensis L. (common names: horseweed, marestail, Canada fleabane) is a winter annual weed that has become an increasing problem in no-tillage agriculture (Tremmel and Peterson 1983), but is also common on roadsides and waste areas (Holm et al. 1997). It outcrosses less than 5% of the time (Smisek et al. 1998) and a single plant has been estimated to produce 70,000–200,000 seeds each with an unbeaked pappus which aids in wind dispersal (Regehr and Bazzaz 1979; Bhowmik and Bekech 1993). The seed bank is short-lived with 76% of seeds germinating or dying within 10 months of release (Davis et al. 2007). Glyphosate-resistant populations of C. canadensis were first confirmed in 2001 in Delaware and resistant populations have been documented in 15 states in the US, Brazil, and China (VanGessel 2001; Heap 2008). As a wind-dispersed species, seeds can travel long distances and have been collected 140 m above the ground surface and to distances of 500 m from a source population (Shields et al. 2006; Dauer et al. 2007, 2009). We assume that seed that are elevated above the crop canopy are able to move freely among fields, and therefore, the landscape topography will not facilitate or impede seed movement. This assumption will need to be revisited as more data are collected about seed trajectories through heterogeneous environments (Levey et al. 2008). If seed are able to freely move above the landscape, they will be deposited according to their distance from the seed origin in a diffusion-like manner. The seed’s ability to germinate after deposition, and weed management strategies, will determine the spread of the resistant biotype.
Germination is assumed to be uniform within a population, but landscape composition will vary with weed management strategies. In 1995, 16 different herbicide active ingredients were applied to more than 1 million hectares of soybean in the US (USDA-NASS, Environmental 2007). With the introduction of glyphosate-resistant soybeans in 1996 and their rapid adoption, glyphosate has become the dominant soybean herbicide and is currently applied to 89% of the soybean area (USDA-NASS, Environmental 2007). Furthermore, glyphosate-resistant varieties of corn, cotton, and alfalfa are now commercially available resulting in greater reliance on glyphosate as the sole or primary weed management tactic. As adoption of GR crops has increased, the number of weedy species resistant to glyphosate has also increased from the first reported case in 1996 to 15 species in 2008 (Heap 2008). Wide-spread adoption of glyphosate-resistant crops increases management uniformity and the suitable habitat for GR weeds when producers rely on glyphosate as the primary form of weed control (Wiens et al. 1993).
Monitoring spread on the landscape scale is difficult and simulation modeling offers an opportunity to explore the importance of spatial and/or temporal cropping system diversity on weed spread. Weed dispersal modeling, often focusing on non-wind-dispersed weed species, has explored spread among and within fields (Perry and Gonzalez-Andujar 1993; Gonzalez-Andujar and Perry 1995; Wang et al. 2003) and the effects of land manager decisions on weed spread (Gonzalez-Andujar et al. 2001; Wadsworth et al. 2000). Our model differs from previous studies because C. canadensis dispersal distances link multiple fields through aerial dispersal, eliminating connectivity as a major driving force of spread, and instead, emphasizing the change in landscape composition. We also chose to use irregular shaped polygons matching actual fields instead of a cellular automaton grid with equal sized cells to describe the landscape (Hanski and Gilpin 1997; Law and Dieckmann 2000). The benefit of representing the complex landscape at our study site as a jigsaw rather than a grid resides in the spatial scale of correlation between life cycle transitions within an agricultural field. Within this natural unit (fields), weed life cycle transitions are correlated by a uniformly applied mortality pressure (weed management). Once surviving plants disperse seeds, the individual seeds are decorrelated as they land in sites with a range of land uses and associated mortality pressures. We selected a study site corresponding to a real landscape with polygons outlining actual fields and inter-field distances calculated among these fields. Whereas cellular automaton models have accepted ways to calculate distances between cells and hence calculate dispersal probabilities (Silvertown et al. 1992; Wang et al. 2003; Bianchi and van der Werf 2004), distances among polygons vary according to size and shape. Instead, distances were calculated among randomly selected points within polygons which estimated mean distance between the polygons. This method accounted for variation in distances and shapes of polygons. A polygonal approach has an additional benefit that allows the landscape to be further subdivided (e.g. adding roadsides that may serve as refugia, fields that are divided, development and urbanization) as additional landscape information is obtained.
Our spatially explicit model was used to explore the following important questions regarding the spread of established herbicide resistant weeds (not including new cases of evolved resistant weeds) in the face of landscape-level management choices. What is the predicted spread of glyphosate-resistant Conyza canadensis if the current GR weed management system is maintained for 10 years? How quickly does the resistant biotype spread if management uniformity changes through increased GR crop adoption? Can short term spread be reduced through increasing crop richness in a landscape or reducing management uniformity by reducing GR crop adoption?
Equation 3 was computed using Monte-Carlo integration, randomly drawing points in source and target polygons (Beyer 2004), then computing the size of the dispersal kernel at the corresponding distance. We explored the number of samples required to converge and found a sample size of twenty points per polygon was accurate to within 4%. This level of accuracy was more than sufficient given the other approximations in our spatial population model.
Mortality sources, natural and human-mediated, affecting the survivorship of Conyza canadensis from seeds in year, t to plants in year, t + 1
Fall to early winter
Regehr and Bazzaz (1979)
Regehr and Bazzaz (1979)
Regehr and Bazzaz (1979)
Late spring to seed set
Seedling mortality due to chemical or mechanical control further reduced the seed-to-seed survivorship percentage. In non-GR crops, herbicide efficacy (mortality) was assumed to be 99.5%. Mortality was lower in GR crops (i.e. survivorship was higher), but still eliminated the majority of C. canadensis plants, which has been observed in resistant populations (M. VanGessel, personal communication). Mortality in GR soybeans was assumed to be 75% while the use of other herbicides in conjunction with glyphosate in corn increased mortality to 90%. Mortality in alfalfa was assumed to be 100% because of repeated intraseasonal clipping. Average C. canadensis fecundity in corn was 74,700 seeds (Regehr and Bazzaz 1979) while average fecundity in soybeans was higher at 133,000 seeds (Dauer et al. 2006). Multiplying herbicide-induced mortality by average yearly mortality and crop-specific mortality resulted in an estimate of the number of plants expected per seed produced. For example, seeds that land in a polygon that will transition to GR corn in year (t + 1), will have a 0.11% chance of surviving to produce a plant in year t, multiplied by 10% chance of surviving the herbicide treatment. Therefore, 9,000 seeds would need to land in this polygon for the successful establishment of a single plant.
To assess the effect of increased GR adoption (increased management uniformity), the subset of corn polygons using GR technology was doubled from 21 to 42% (HIGR), with percent of GR soybeans unchanged from the CUR scenario. A fourth scenario simulated the possible antagonism between increased crop diversity and decreased herbicide diversity by reducing the polygons in the corn/soybean rotation and adding alfalfa to the HIGR scenario (creating HIGR + ALF).
Two scenarios were included to simulate a lower percent of GR corn and soybean compared to the CUR scenario. The PRE96 scenario represented a corn-soybean rotation landscape, prior to introduction of GR crops where none of the polygons were assigned the GR technology. With no safe sites, GR C. canadensis was expected to diffuse into the environment and virtually disappear from the landscape because of the high C. canadensis mortality in all non-GR crops. This reduction would also be true for susceptible C. canadensis which is generally controlled by many common herbicides. There is little evidence of a fitness cost associate with glyphosate-resistance in C. canadensis (M. J. VanGessel, personal communication), but if a cost existed, it would likely hasten the loss of GR C. canadensis in the absence of glyphosate use (Matthews 1994). The LOWGR scenario represented a reduction in GR soybeans that could result from mandated reductions in planting GR soybeans (as is practiced for Bt corn, see Cerda and Wright 2004), planting the same percent of GR soybeans as CUR, but using additional herbicides to reduce the area only receiving glyphosate, or area-wide cooperation to reduce the spread of a GR weed. The LOWGR scenario reduced the percent of GR soybean by half from 89 to 45%, while GR corn was held at 21% (the CUR level).
The Monte-Carlo simulation of spatial dynamics occurred in time steps of 1 year and simulated spread was continued to the end of the tenth year and repeated 100 times per scenario. With the anticipated rapid spread, producers are unlikely to maintain the same rotation beyond 10 years if confronted with an impending infestation. Prior to each run, a single resistant plant was assigned to a central soybean field (within 1 km of the center of the landscape photo), the crop arrangement was shuffled, and the simulation began with the fall dispersal of seeds from the source population. The random location of crops and weed management strategies is important because any field is equally likely to be adjacent to a suitable and unsuitable habitat and therefore guarantees that distance among suitable habitats does not vary by scenario or by year. The number of plants per polygon was stored at each time step. There were insufficient data on C. canadensis effects on crop yield to parameterize a yield loss curve (Bruce and Kells 1990). Nonetheless, polygons were subjectively classified into zero plants, easily overlooked (<5 plants ha−1), noticeable (5–25 plants ha−1), potentially problematic (25–50 plants ha−1), and potentially yield reducing (50+ plants ha−1). Hereafter, densities greater than 25 plants ha−1 will be referred to as infested while fields with lower densities will be considered uninfested, but containing resistant plants.
We speculated that the spread rate of the species would depend on the degree to which the dispersing seed had access to suitable patches. From the perspective of a producer, the relative fitness of resistant plants may be much larger than susceptible plants within the GR cropping system. Therefore, this model only projects existing resistant plant dynamics. Metrics of spread vary widely (Hilborn and Mangel 1997) but we chose to use the likelihood that a field separated from the initial infestation contained resistant plants within 10 years. In year 10, fields with densities greater than 25 plants ha−1 were assigned a value of 1 and fields with a lower density were assigned a value of 0. A logistic equation (y = exp (a + b × x)/1 + exp (a + b × x)) determined the infestation probability (y) as a function of distance from the source (x) and two shape parameters a and b (MacDonald and Rushton 2003) and fit to simulation output data. This likelihood was a conservative estimate because the defined dispersal capability does not include the rare long-distance events that are known to occur with this species (Shields et al. 2006), but does account for changes in source strength (Rieger et al. 2002; Dauer et al. 2007). Moreover, if one included random development of resistance, the likelihood of a field receiving the resistant biotype could be much higher. All simulations and analyses were carried out in R version 2.8 (R Core Development Team 2008).
Comparison of scenario effects on mean polygons infested by GR Conyza canadensis after 6 and 10 years following 100 iterations for each scenario
Density (plants ha−1)
As expected, when glyphosate-resistant crops were not planted (PRE96) scenario, the landscape provided no safe sites for the glyphosate-resistant plants and rarely were any plants present after 6 years. The LOWGR scenario reduced management uniformity and reduced spread but did not eliminate the resistant biotype. LOWGR represented the slowest rate of spread and an average of 12 polygons (out of 360) contained plants after 10 years (Table 3, Fig. 2).
If the simulations are extended to 15 years, the likelihood of infestation increases dramatically, at all distances from the source (Fig. 3). CUR and CUR + ALF remain close together, but a field at 4 km, now has a 35–40% likelihood of infestation. HIGR and HIGR + ALF also increased to a 65–70% likelihood of infestation.
Increased infestation rates when seedling survivorship is increased by 10% and when dispersal is increased by 10%
Density (plants ha-1)
10% greater survivorship
Greater Kurtosis in 2Dt
The main objective of this model was to test the sensitivity to variation in management across a landscape. Heterogeneous environments may limit dispersal in some landscapes (Taylor et al. 1993; Levey et al. 2008), but wind-dispersed species move with relatively little hindrance. With the importance of connectivity reduced in our model, the primary difference between scenarios was landscape composition manifested in changes to management uniformity. To some, the result of this model is obvious: increase widespread reliance on a single management strategy and an organism will increase if they are resistant or tolerant to the strategy. However, producers are either unaware of this fact or continue to increase adoption of GR crops despite this fact. The former is unlikely, so we will focus on the later and how this simple model may suggest proactive management may be critical in slowing GR weed spread through agricultural settings.
In the face of an impending invasion by resistant weeds, a producer is likely to consider options ranging from changing management on their own field to coordinating management among regional producers. The most effective solution is to increase tillage because C. canadensis is controlled by this technique (Brown and Whitwell 1998; Buhler 1992). The most accepted practice is applying different herbicides which have increased herbicide efficacy on C. canadensis, in combination with or in replacement to glyphosate. Glyphosate replacement is not occurring as evidenced by the steady increase in glyphosate use while the acreage receiving other herbicides has steadily declined since 1996 (Coble 2008; USDA-NASS, Environmental 2007). Instead, it is likely that producers are relying on combining different herbicides with glyphosate to increase efficacy. If half of soybean producers implemented this strategy, the landscape would closely resemble the LOWGR scenario used in our model, which reduced infestation of GR C. canadensis more than any other scenario. This ‘solution’ masks a number of potential problems that are currently not considered in the model. In particular, reapplying additional herbicides changes the selection pressure, but C. canadensis populations have developed multiple-resistance to a number of commonly used herbicide active ingredients (Gressel et al. 1982; Pölös et al. 1988; Smisek et al. 1998; Trainer et al. 2005). Adding additional herbicides to manage for GR C. canadensis also ignores the impending development of glyphosate-resistance in other species (Heap 2008) that have different responses to herbicide mixtures. Producers will soon be asking, “which GR weed is the most harmful to me?” In this regard, the model could be enhanced to incorporate newly arising resistant biotypes and the evolution of multiple-resistance, either of which would accelerate the spread rate of resistant biotypes (Maxwell et al. 1990; Diggle et al. 2003).
Often producers utilize fields across a wide swath of the landscape, renting or owning discontinuous fields. Movement of equipment among these fields may play an important role in spread of weed species. The fields are discontinuous, but the connections are frequent, assuring certain fields have a greater likelihood of correspondence. The increased connectivity between certain fields could be intertwined into the current framework by parameterizing the Φ matrix (Eq. 1). Instead of the dispersal kernel dictating probability of aerial transport among fields, probabilities could assess seed transport via equipment. A producer carrying resistant seeds to their various fields may create stepping stones or satellite populations that have been shown to affect spread of invasive species (With 2004). Recent studies have also documented the significant role human mediated dispersal plays via boats (Leung et al. 2006), automobiles (Von der Lippe and Kowarik 2007), and contaminated crop seed and hay (Magda and Gonnet 2001). Each of these methods will increase connectivity and may facilitate spread much more quickly than even wind-dispersed seed. Alternatively, if connected fields are utilized strategically, producers might be able to limit spread by reducing suitable habitat quantity and location. This is the basis for proactive landscape-scale weed management strategies.
Reducing suitable habitat is challenging when the two crops are using similar strategies. Increased crop diversity (creating a 3 crop landscape instead of 2) reduced spread, although the number of infested fields did not differ from the two crop rotations. Including alfalfa in the rotation increased the number of ‘zero’ fields by 5–8%. Alfalfa is a common third crop throughout the US, but alternative crops, for example wheat, can have a similar impact if the weed management strategy is different from GR and has high weed management efficacy. The key is to increase the number of weed management techniques. For example, spread of GR C. canadensis in Eastern Pennsylvania has been limited by the wide diversity of crops compared to spread in Delaware where corn and soybean dominate (D. A. Mortensen and J. T. Dauer, unpublished data).
The next step is coordinated crop arrangement which can increase distance between suitable habitats, a concept readily accepted by entomologists. Carriere et al. (2006) found the arrangement of alfalfa fields [sources of the western tarnished plant bug (Lygus hesperus)] to cotton fields directly affected the abundance of insects in the cotton fields. The size, structure, and arrangement of refugia have been a major component of maintaining the viability of Bacillus thuringiensis (Bt) crops by providing an area where potentially resistant insects would breed with susceptible insects. Randomly situated refugia increased persistence of Bt-resistance alleles when compared to permanent blocks (Cerda and Wright 2004; Peck et al. 1999). The greatest coordination would be barrier zones which have been employed successfully to reduce the spread of gypsy moth (Lymantria dispar) in Eastern US forests (Sharov and Liebhold 1998).
Clearly, the continued adoption of GR crops will further accelerate the spread of GR weeds, leading to a landscape with high management uniformity and many fields infested with herbicide resistant weeds. Our model indicates that factors which increase the diversity of mortality sources and reduce the selection pressure for resistant genes to persist, will most profoundly slow the spread of herbicide resistance. The crop seed and input industry is rapidly developing crops with multiple insect and herbicide resistance traits (Gurian-Sherman 2007) the deployment of which should be preceded by ecological risk assessments that include the likelihood of pest resistance. Models like that reported herein and Margosian et al. (2009) provide badly needed syntheses of data and observation to guide the design and certification of future GR crop releases and cropping practices. Such information is needed now for current and future GR crop regulatory decisions and cropping systems recommendations.
This work is a natural extension of a significant body of empirical work conducted over the past 4 years. The construct of the model also benefited from numerous discussions with ecologists, weed scientists and agriculturalists at our study sites and at regional and national meetings. In particular, this work was improved through contributions from B. Maxwell and W. van der Werf, and two anonymous reviewers. Additional insights were provided by O. Bjornstad, B. Curran, and S. Isard, and the Borer-Seabloom lab group. This work was funded through grant from the USDA-NRI Weedy and Invasive Plants (#2004-02158) Program.
- Beyer HL (2004) Hawth’s analysis tools for ArcGIS. http://www.spatialecology.com/htools. Accessed 22 November 2006
- Bhowmik PC, Bekech MM (1993) Horseweed (Conyza canadensis) seed production, emergence, and distribution in no-tillage and conventional-tillage corn (Zea mays). Agronomy (trends in agricultural science) 1:67–71Google Scholar
- Brown CM, Whitwell T (1998) Influence of tillage on horseweed, Conyza canadensis. Weed Technol 2:269–277Google Scholar
- Bruce JA, Kells JJ (1990) Horseweed (Conyza canadensis) control in no-tillage soybeans (Glycine max) with preplant and preemergence herbicides. Weed Technol 4:642–647Google Scholar
- Buhler DD (1992) Population dynamics and control of annual weeds in corn (Zea mays) as influenced by tillage systems. Weed Sci 40:241–248Google Scholar
- Carriere Y, Ellsworth PC, Dutilleul P, Ellers-Kirk C, Barkley V, Antilla L (2006) A GIS-based approach for area wide pest management: the scales of Lygus hesperus movements to cotton from alfalfa, weeds, and cotton. Entomol Exp Appl 118:203–210. doi:10.1111/j.1570-7458.2006.00384.x CrossRefGoogle Scholar
- Coble HD (2008) The role, value, and importance of complementary herbicides for weed management in glyphosate tolerant crops. Proceedings weed science society of AmericaGoogle Scholar
- Dauer JT (2007) From emergence to impact: Conyza canadensis adaptations that facilitate seed dispersal. PhD thesis, The Pennsylvania State UniversityGoogle Scholar
- Gressel J, Ammon HU, Fogelfors H, Gasquez J, Kay QON, Kees H (1982) Herbicide resistance in plants. Wiley, New YorkGoogle Scholar
- Hanski I, Gilpin ME (1997) Metapopulation biology: ecology, genetics, and evolution. Academic Press, San DiegoGoogle Scholar
- Heap I (2008) The international survey of herbicide resistant weeds. http://www.weedscience.com. Accessed 06 May 2008
- Hilborn R, Mangel M (1997) The ecological detective: confronting models with data. Princeton University Press, Princeton, NJGoogle Scholar
- Holm L, Doll J, Holm E, Pancho J, Herberger J (1997) Conyza canadensis (L.) Cronq. (syn. Erigeron canadensis L.). World weeds: natural histories and distributions. Wiley, Toronto, pp 226–235Google Scholar
- Law R, Dieckmann U (2000) Moment approximations of individual-based models. In: Dieckmann U, Law R, Metz JAJ (eds) The geometry of ecological interactions: simplifying spatial complexity. Cambridge University Press, Cambridge, pp 252–269Google Scholar
- Matthews JM (1994) Management of herbicide resistant weed populations. In: Powles SB, Holtum JAM (eds) Herbicide resistance in plants: biology, biochemistry. Lewis Publishers, Boca RatonGoogle Scholar
- Maxwell BD, Roush ML, Radosevich SR (1990) Predicting the evolution and dynamics of herbicide resistance in weed populations. Weed Technol 4:2–13Google Scholar
- Mortensen DA, Jongejans E (2006) The influence of habitat type, extent and connectivity on the early stages of a plant invasion. Centre for Environmental Management of Degraded Ecosystems. International symposium on biology, ecology and management of the world’s worst plant invasive species, University of Delhi, Delhi, IndiaGoogle Scholar
- PASDA (2007) Pennsylvania spatial data access. http://www.pasda.psu.edu. Accessed 13 Sept 2007
- Peck SL, Gould F, Ellner SP (1999) Spread of resistance in spatially extended regions of transgenic cotton: implications for management of Heliothis virescens (Lepidoptera: Noctuidae). J Econ Entomol 92:1–16Google Scholar
- R Development Core Team (2008) R: a language and environment for statistical computing. R foundation for statistical computing, Vienna, Austria. ISBN 3-900051-07-0, URL http://www.R-project.org
- Smisek A, Doucet C, Jones M, Weaver SE (1998) Paraquat resistance in horseweed (Conyza canadensis) and Virginia pepperweed (Lepidium virginicum) from Essex County, Ontario. Weed Sci 46:200–204Google Scholar
- USDA-NASS Environmental (2007) Ag chemical usage reports. US Department of Agriculture. http://www.nass.usda.gov/Statistics_by_Subject/Environmental/. Accessed 13 Sept 2007
- USDA-NASS Publications (2007) Acreage US Department of Agriculture. http://www.nass.usda.gov/Publications/. Accessed 13 Sept 2007