Road age and its importance in earthworm invasion of northern boreal forests

Authors


*Correspondence author: E-mail: ecameron@ualberta.ca

Summary

  • 1Roads are an important conduit for the spread of invasive species. Road age is a key factor that could influence the susceptibility of roads to invasion as older roads are typically subject to higher cumulative levels of human disturbance and propagule pressure than younger roads. We investigated the effects of road age on the spread of non-native earthworms, which act as ecosystem engineers.
  • 2We sampled earthworms and habitat variables at 98 roads in the boreal forest of Alberta, Canada, to determine the influence of road age on non-native earthworm occurrence at the landscape level. The extent and rate of local spread were also assessed at seven sites adjacent to old and young roads. Generalized estimating equations and zero-inflated negative binomial regression were used to analyze landscape- and local-level results, respectively. We used our models to create maps that predict the current and potential future extent of earthworms in north-eastern Alberta.
  • 3Probability of earthworm occurrence and extent of spread increased as road age increased. Areas closer to agriculture and towards the south and west of our study area were also significantly more likely to be invaded by earthworms.
  • 4Our spread model indicated that approximately 9% of the boreal forest of north-eastern Alberta is likely invaded by earthworms currently. This is projected to increase to 49% of suitable forest habitat over the next 50 years as human development intensifies in this region.
  • 5Synthesis and applications. Although the effects of roads and linear features are commonly investigated in relation to native species, our results emphasize the importance of considering the impacts of linear feature creation on the spread of invasive species. We demonstrate that road age in particular can be an important factor affecting the spread of invasive species. In the boreal forest, reducing the number of roads being constructed, restricting traffic, and reclaiming temporary roads will be critical to reduce the future extent of earthworm invasions.

Introduction

Roads are a key pathway in the introduction and spread of non-native species, including pathogens, plants, invertebrates, and vertebrates (Jules et al. 2002; Gelbard & Belnap 2003; Gundale, Jolly & Deluca 2005; Brown et al. 2006). Non-native species invasions are more likely to occur near roads due to increases in: (i) level of human activity (Tyser & Worley 1992); (ii) permeability to invasions when linear features function as dispersal corridors (Tyser & Worley 1992); and (iii) availability of disturbed habitats caused by road construction and maintenance that allow establishment of invasive species (Parendes & Jones 2000; Gelbard & Belnap 2003). Vehicles may play a particularly significant role in promoting invasions along roads due to their ability to transport propagules over long distances (von der Lippe & Kowarik 2007).

The factors which influence the likelihood that one road corridor vs. another is colonized by non-native species are not well understood, however. It is believed that roads with greater human disturbance are more likely to be invaded than less-disturbed roads, due to greater propagule pressure (a measure of the number of individuals introduced and number of introduction events) and increased habitat alteration associated with human activities in roaded areas (Tyser & Worley 1992; Parendes & Jones 2000; Watkins et al. 2003; Lockwood, Cassey & Blackburn 2005). For example, the cover and richness of non-native plants has been found to increase as road improvement level increases from ungraded four-wheel-drive tracks to paved roads (Gelbard & Belnap 2003). More improved roads experience greater traffic levels, habitat alteration, and frequency of use by vehicles and road maintenance equipment, all of which contribute to increased disturbance (Gelbard & Belnap 2003). Similarly, roads which are subject to heavy traffic and regular maintenance experience increased rates of plant invasion as compared to abandoned roads (Parendes & Jones 2000). Another factor that could influence propagule pressure and amount of disturbance is the age of the road. If older roads are typically subject to higher cumulative levels of traffic and maintenance than young roads, this might result in an increase in non-native species occurrence near old roads simply due to the longer period of exposure.

Earthworms are one group of non-native species which appear to be introduced and spread along road networks (Dymond, Scheu & Parkinson 1997; Gundale et al. 2005; Cameron, Bayne & Clapperton 2007; Holdsworth, Frelich & Reich 2007). Native earthworms were extirpated from most of Canada and the northern USA during the Pleistocene glaciations (Gates 1970; Gates 1982). During European settlement of North America, European earthworms were probably introduced with the dumping of soil used for ship ballast and with the transport of soil associated with plant material (Tiunov et al. 2006). Within the last several decades, these species have begun spreading into northern forests (Scheu & Parkinson 1994; McLean & Parkinson 1997). Introduction of non-native earthworms can result in a number of significant changes to forest ecosystems, such as loss of soil carbon (Alban & Berry 1994; Bohlen et al. 2004), reduction of soil organic content (McLean & Parkinson 1997; Gundale 2002), decreased microarthropod species richness (Eisenhauer et al. 2007), and decreased abundance and diversity of understorey plants (Hale, Frelich & Reich 2006). Human activities are continuing to play an important role in promoting introduction and spread of earthworms in forests, as earthworms have limited active dispersal abilities (Marinissen & van den Bosch 1992). A key uncertainty is whether the spread of earthworms along roads is more likely to occur via initial transport of earthworms or their cocoons in soil or gravel during road construction or via transport by vehicles (i.e. in tyre treads or wheel wells) after the road has been built (Dymond et al. 1997; Gundale et al. 2005).

In northern Alberta, Canada, six species of earthworms are invading boreal forest stands (Cameron et al. 2007). Linear features, including roads, pipelines, and seismic lines, currently average 1·8 km km−2 in this area and are projected to increase to 8 km km−2 over the next century (Schneider et al. 2003). Given this dramatic increase in linear features, it is important to understand the risk that such features pose as invasion conduits. Our primary objective was to assess the effects of road age on earthworm introduction and spread in Alberta's boreal forest. Specifically, we investigated earthworm introduction at the landscape level by determining the probability of earthworm occurrence in forest stands adjacent to roads of varying ages. Once a road corridor is invaded, earthworm populations may begin to spread. Therefore, to examine the rate of spread once introduced, and to assess local-level spread in relation to road age, we intensively sampled earthworms at 500 × 500 m grids. We hypothesized that earthworms would be more likely to occur and have spread farther at older roads as compared to newer roads. A positive relationship between road age and earthworm occurrence would suggest that earthworms are typically introduced by traffic. However, if earthworms are introduced during the construction process via initial importation of gravel and soil, earthworm occurrence should be a function of whether or not the road fill used in construction contained earthworms. This should be independent of road age. Using these data, we then created a spatial model to predict the current and future distributions of earthworms in relation to roads in north-eastern Alberta over the next 50 years.

Methods

study area

Our study was conducted in the boreal forest of northern Alberta, Canada, between 54·4°N and 58·8°N latitude and 110·1°W and 119·3°W longitude during May to early August in 2005–2006 (Fig. 1). This region is subject to multiple overlapping land use practices, which are leading to large cumulative impacts on forest structure and function (Schneider 2002; Schneider et al. 2003). Both the volume of forest harvested and rate of oil and gas drilling have increased rapidly in recent years (Schneider 2002). Expansion of the oil sands industry due to world demand for fuel has been the major driver increasing linear features in boreal Alberta. Additionally, this area experiences considerable agricultural activity and conversion of forest to agricultural use continues to occur.

Figure 1.

Locations of study sites within the boreal forest of northern Alberta. Black circles represent landscape-level sites and white squares represent local-level sites.

Northern Alberta is vegetated by deciduous, mixedwood, and lowland forests (Natural Regions Committee 2006). Grey luvisols and brunisols are the dominant soils in upland forests (Beckingham & Archibald 1996). Mean annual temperature is −0·2°C and mean annual precipitation is 469 mm (Natural Regions Committee 2006). At our study sites, trembling aspen Populus tremuloides Michx., balsam poplar Populus balsamifera L., and white spruce Picea glauca (Moenech) Voss were the most common tree species. Common shrubs included rose Rosa acicularis Lindl., low-bush cranberry Viburnum edule (Michx.) Raf., and saskatoon Amelanchier alnifolia Nutt.

site selection

We used Alberta Ground Cover Classification (AGCC) and Alberta Vegetation Inventory (AVI) data obtained from Alberta-Pacific Forest Industries and aerial photos from Air Photo Distribution in Edmonton, Alberta to select study sites. All sites were located in mature deciduous or mixedwood stands that were adjacent to roads. Based on historical aerial photography back to 1949, we found no evidence that any of these stands had ever been cleared for agriculture or harvested. Road ages were determined by reviewing aerial photos taken between 1949 and 2005. Photographs were black and white, and had scales between 1:15 000 and 1:60 000. In almost all cases, the date that a road was built could not be narrowed down to a single year (due to gaps in coverage ranging from 1 to 16 years in length), and construction was assumed to have occurred in the median year within the range of uncertainty.

In the landscape-level survey, we sampled transects at 98 roads ranging in age from 6 to 56 years old. We ensured that roads of different ages were evenly distributed across the study area. In the local-level survey, we sampled at three newer roads built after 1980 and four older roads built between 1957 and 1962. These seven sites were in locations where earthworms were known to be present along the road and where there was an area of forest at least 550 × 550 m that did not contain any other anthropogenic features.

data collection

At each site in the landscape-level survey, earthworms were sampled along a 50-m transect located parallel to the road and 1 to 2 m from the forest edge. The average distance between the forest edge and road was 13 m. Six 0·0625 m2 (25 × 25 cm) quadrats spaced 10 m apart were sampled on each transect, with every second quadrat located 5 m farther into the forest interior. In the local-level survey, each site consisted of a 500 × 500 m square grid, with 50 m spacing between sampling quadrats. This resulted in a total of 121 0·0625-m2 quadrats sampled in each grid. Each grid started approximately 5 m into the forest from the road, with the rest of the grid extending into the forest. When earthworms were not present in two consecutive quadrats in a transect running from the road to the forest interior, a quadrat 25 m from the last quadrat where earthworms were found was searched in order to more accurately map the location of the invasion front.

At each sampling location, we removed the Oi, Oe, and Oa horizons from our quadrats and hand-sorted this material to determine the abundance of adult earthworms and juvenile earthworms. Quadrat size was selected based on a pilot study which indicated that earthworm abundance in 25 × 25 cm vs. 50 × 50 cm quadrats was strongly correlated (r2 = 0·88). Although no methods of earthworm sampling allow complete enumeration, hand-sorting provides an accurate estimation of the number of individuals in an area for most species (Callaham & Hendrix 1997; Lawrence & Bowers 2002). We did not sample the mineral soil layers in our quadrats, as previous research in northern Alberta indicated that sampling these layers did not significantly increase detection rates (Cameron et al. 2007). This is particularly likely to be the case when conditions are consistently wet as they were in 2005 and 2006. Earthworms were preserved in 70% ethanol and adults were later keyed to species (Reynolds 1977). Juveniles were grouped into three categories (Dendrobaena octaedra (Savigny)/Dendrodrilus rubidus (Savigny); Lumbricus sp.; and Aporrectodea sp.) based on ecological and taxonomic similarities because, in some cases, it is impossible to identify juveniles to the species-level (Hale, Frelich & Reich 2005).

We measured the depth of the organic horizon in each quadrat using a metal metre stick. The percentage cover of moss, leaves, grass, and forbs for each quadrat was estimated using five categories (0, none present; 1, 1–25% cover; 2, 26–50% cover; 3, 51–75% cover; 4, 76–100% cover). The distance to the nearest tree, species of the tree, and distance to the nearest shrub were also recorded. In addition, we used the Field Guide to Ecosites of Northern Alberta to categorize the ecosite, ecosite phase, and plant community at our sites (Beckingham & Archibald 1996). Ecosites are areas with similar moisture and nutrient characteristics, while ecosite phases are classified by dominant canopy species, and plant community types are based on understorey species composition and abundance. Distance to the nearest agricultural area was calculated for all transects in the landscape-level survey using ArcGIS 9·1 (ESRI Inc. Redlands, California) and Hawth's Analysis Tools (Beyer 2004).

statistical analysis

Landscape level

We assessed the effects of road age on earthworm occurrence by using a population-averaged generalized estimating equation (GEE) with a logit link and binomial error structure in stata 9·1 (StataCorp, College Station, Texas). GEEs are a modification of generalized linear models in which an estimate of the correlation structure within panels (i.e. transects in this case) is calculated to account for the hierarchical structure of the data (Hardin & Hilbe 2003). The primary sampling unit is the transect and each quadrat occurs within a transect. We used a semi-robust estimator of variance which generates standard errors that are robust to any lack of independence even if the correlations within transects were not as hypothesized by the specified correlation structure. There were 588 quadrats and 98 transects in the analysis. Easting, northing, distance to agriculture, and vegetation variables were included in the global model and removed using a stepwise backwards elimination procedure if found to be non-significant. Easting, northing, and distance to agriculture were converted from metres to kilometres. All variables with P ≤ 0·15 were retained within the model when testing for significance of other predictor variables (Hosmer & Lemeshow 2000).

An extension of the Hosmer–Lemeshow test for goodness of fit, designed specifically for GEEs, was used to assess the fit of the model (Hosmer & Lemeshow 1980; Horton et al. 1999). This statistic is calculated by dividing the predicted values obtained from the model into groups defined by deciles, and performing a chi-squared test of the observed vs. predicted values.

Local-level grids

We created maps of earthworm abundance at each grid using ordinary kriging in ArcGIS 9·1. Ordinary kriging is an interpolation technique that uses known values obtained at sampling points to predict unknown values at other points within the sampling area (Isaaks & Srivastava 1989). Zero-inflated negative binomial regression was also used to test the effects of distance from the road, age class of the road, and the interaction between age class and distance on earthworm abundance and occurrence within stata 9·1. Zero-inflated regressions use a two-part modelling approach, consisting of a binary outcome model which models the probability of obtaining a zero count and a truncated count model which models the non-zero counts (Cameron & Trivedi 1998). This type of regression is suitable for data that displays overdispersion and a high incidence of zero counts (Cameron & Trivedi 1998). To account for the probable lack of independence of quadrats within sites, a robust clustering approach was employed. This technique uses a variance estimator to adjust standard errors, and thereby accounts for within-site correlations (Rogers 1993). Non-significant vegetation variables were removed using a stepwise backwards elimination procedure, with all variables with P ≤ 0·15 retained within the model (Hosmer & Lemeshow 2000). We assessed goodness of fit using a chi-squared analysis of predicted vs. observed values, and compared the fit with that obtained from Poisson, negative binomial, and zero-inflated Poisson regressions. Finally, we calculated the approximate average rate of earthworm spread from the roads by dividing the distance of the invasion front from each road by the age of the road.

Spatial modelling of earthworm distribution

We used ArcGIS 9·1 to generate a model of current and future earthworm distribution in the Alberta Pacific Forestry Industries Forest Management Area (Al-Pac FMA), a 59 054-km2 area in north-eastern Alberta. There is little data available on the age structure of the road network in northern Alberta, although we were able to obtain some ages from the Mistakiis Institute. Ages were assigned to the remaining roads based on the average age of the nearest wells and cutblocks, because much of the road network was built to access these features. To create a future road network with a realistic road density, we used ALCES® (A Landscape Cumulative Effects Simulator) to obtain projected road densities over the next 50 years (see Schneider et al. 2003 for details of the model). Road networks from the 15% of townships with the highest road densities were then randomly added to the existing network in ArcGIS until the overall road density was equal to the projected ALCES estimate for each decade.

The Convert Paths to Points tool in Hawth's Analysis Tools for ArcGIS (Beyer 2004) was then used to generate a point every 10 m along the road network. These points were randomly invaded at a rate of 1·03% per year, excluding the year in which the road was built. This invasion rate was obtained from the slope of a linear regression of our earthworm occurrence data in relation to road age. Buffers were created around each invaded point in order to determine the areal extent of earthworms in 50 years, assuming a spread rate of 10 m per year. To determine the total area of forest likely to be invaded, we intersected this buffered layer with a layer containing forest habitat suitable for invasion. Suitable habitat included all forest types except those dominated by black spruce Picea mariana or tamarack Larix laricina, as such forests have highly acidic soils and are thus less likely to be colonized by earthworms (Bouché 1977; Edwards & Bohlen 1996).

Results

landscape level

Of the 588 quadrats sampled, 204 (35%) had earthworms present. Similarly, earthworms occurred along 44 out of 98 transects (45%). Densities within quadrats ranged from 0 to 1040 earthworms m−2, with an average of 42 earthworms m−2. The most common species at our sites was D. octaedra, an epigeic (litter-dwelling) species (99·8% of adults). D. rubidus and Aporrectodea tuberculata (Eisen) were also present. Most juveniles were D. octaedra or D. rubidus, although there were also four Aporrectodea sp. individuals.

Road age had a significant effect on earthworm occurrence at the landscape level, with older roads being more likely to have earthworms present than younger roads (odds ratio = 1·07, P = 0·001; Fig. 2). The correlation of quadrats within transects was 0·67, indicating that strong spatial autocorrelation existed within transects. Distance to the nearest shrub (odds ratio = 0·994, P = 0·036), distance to agriculture (odds ratio = 1·02, P = 0·021), easting (odds ratio = 0·993, P < 0·001), and northing (odds ratio = 0·988, P < 0·001) also had significant effects at P ≤ 0·05 on earthworm occurrence. Distance to the nearest tree was retained in the model as well, but did not significantly influence earthworm presence. The goodness-of-fit test was not significant (χ2 = 13·2, P = 0·11), which suggests that the fit of the model was adequate. We also used standard logistic regression to conduct a similar analysis of earthworm occurrence at the transect level, rather than within quadrats, which produced similar results.

Figure 2.

Regression plot of predicted probability of earthworm occurrence (solid line) in relation to road age in years. Dotted lines represent 95% confidence intervals. Other model covariates are held at their mean values. Black dots represent actual earthworm presences and absences.

local-level grids

Three hundred forty-three out of 847 quadrats (40·5%) on the local-level grids were occupied by earthworms. Earthworm densities in the quadrats ranged from 0  to 1200 m−2, with an average density of 81 earthworms m−2. D. octaedra was the most common species (96·9% of adults), followed by D. rubidus (3·0% of adults), and then by L. rubellus (0·0005% of adults). All juveniles were D. octaedra or D. rubidus.

The kriged maps for the older roads showed that earthworms had spread to the back of the grid at each site (Fig. 3). For the newer roads, the maps showed that earthworms had spread less than 100 m from the road in each case (Fig. 4). Earthworm abundance decreased significantly as distance from the road increased (incidence rate ratio = 0·9989, P = 0·001). However, age class and the interaction between distance and age class did not significantly affect earthworm abundance. Tree distance was the only habitat variable which significantly influenced earthworm abundance (incidence rate ratio = 1·0015, P = 0·01). The probability of earthworms being absent also increased significantly as distance from the road increased (odds ratio = 1·0491, P = 0·0001). In addition, the distance and age class interaction had a significant effect on earthworm absence. Newer roads were more likely to have earthworms absent as distance increased from the edge than older roads (odds ratio = 0·9596, P = 0·001). Age class did not significantly affect earthworm absence. No vegetation variables significantly affected earthworm absence, although nearest shrub distance and forb percentage cover were retained in the model. The goodness-of-fit test showed that zero-inflated negative binomial regression produced predictions that agreed more closely with the observed data than those produced by Poisson, negative binomial, or zero-inflated Poisson regressions. The average rate of earthworm spread at the younger roads was 1·8 m year−1. Because earthworms were present at the farthest points sampled at older roads, invasion fronts could not be identified, and we were unable to calculate spread rates from those sites.

Figure 3.

Kriged map of earthworm occurrence at a typical older road site. The legend shows the predicted number of earthworms 0·0625 m−2 quadrat.

Figure 4.

Kriged map of earthworm occurrence at a typical newer road site. The legend shows the predicted number of earthworms 0·0625 m−2 quadrat.

modelling

Total road length in the Al-Pac FMA is currently 22 068 km. The total area of suitable forest habitat for earthworms within the Al-Pac FMA was 24 447·6 km2. Using current road densities and estimated road ages, our model predicts that 2221·9 km2 (9·09%) of this area is currently invaded by earthworms. If rates of road development occur as predicted by projections from ALCES (see Supplementary Material Fig. S1), we expect there will be 54 477 km of roads built over the next 50 years. According to our model, this will result in 12 044·6 km2 (49%) of the Al-Pac FMA being invaded by earthworms within 50 years (Fig. 5).

Figure 5.

Predicted future extent of earthworm occurrence within the Al-Pac FMA in 50 years. Areas invaded by earthworms are indicated in black and the Al-Pac FMA is shown in white.

Discussion

landscape-level occurrence

In northern Alberta, road age appears to strongly affect the occurrence of non-native earthworms, with older road corridors being significantly more likely to have earthworms present than younger roads. Older roads probably have experienced a greater cumulative amount of vehicular traffic and other disturbances such as road maintenance than roads constructed more recently. This could result in greater propagule pressure and an increased probability of earthworm occurrence along older roads. Our finding is comparable to previous studies which found that invasive plant species richness increased with time since human settlement in American states (McKinney 2001) and protected areas (McKinney 2002).

Earthworm presence was also significantly influenced by site location. Earthworms were more likely to occur towards the south and west boundaries of our study area, where anthropogenic activities have historically been more intense. Therefore, there has probably been more opportunity for earthworm introduction in these areas over time than in the north-east. Agriculture, which is also concentrated in the south and west, significantly affected earthworm occurrence as well. This is consistent with a study which found that agricultural fields appear to act as major sources of non-native earthworm introduction in New York (Suarez et al. 2006). Earthworms and cocoons are probably transported to farms on the tyres of trucks or agricultural machinery, and in potted plants (Marinissen & van den Bosch 1992; Suarez et al. 2006). Earthworms can attain high densities in agricultural clearings and thereby act as important source populations for invasion of nearby areas (Suarez et al. 2006).

Other characteristics of roads appear to be important determinants of invasive plant occurrence, and thus may also affect earthworm establishment. First, higher levels of road improvement are associated with increased cover of non-native plant species (Gelbard & Belnap 2003). Because most of the accessible roads in our study area are gravel, we were unable to test whether road improvement affects earthworm occurrence. Secondly, high levels of traffic can result in higher rates of non-native plant invasions (Parendes & Jones 2000). We estimated traffic levels (none, moderate, or high) at 91 of our sites but found no relationship with earthworm occurrence. However, traffic levels obtained at a single point in time are likely to be a poor measure of intensity of use.

local-level spread

Because the processes governing dispersal of invasive species may differ between scales, we examined spread at the local level in addition to the landscape level (Pauchard & Shea 2006). Road age also appears to strongly influence local spread of earthworms as spread had occurred over significantly greater distances from older roads as compared to younger roads in our study sites. The opportunity for earthworm introduction via vehicular transport would probably have occurred much earlier in areas with older roads than in those with roads built more recently. This means that there should have been more time for earthworms to spread, resulting in populations that extend over a larger area at such locations.

Earthworm occurrence and abundance in our local-level survey were also affected by distance from roads, with fewer earthworms occurring as distance increased. This result is consistent with previous studies which examined earthworm spread at shorter distances from roads in Alberta (Dymond et al. 1997; Cameron et al. 2007). In contrast to this, in hardwood forests in the north-eastern USA, road presence appears to be more appropriate as a coarse-scale rather than fine-scale predictor of earthworm invasion (Suarez et al. 2006; Holdsworth et al. 2007). This may be related to the fact that earthworm invasions in the north-eastern USA probably began earlier than invasions in Alberta (100 years vs. 50–60 years), and consequently, earthworms would be expected to have dispersed farther from initial introduction sites in the hardwood forests (Holdsworth et al. 2007). Initial introductions thus appear to have occurred along the roads at our study sites, with subsequent spread occurring towards forest interiors. Cocoons of D. octaedra and D. rubidus, the most common species in our samples, are particularly likely to be transported by vehicles because they are produced in large quantities via parthenogenesis, and are found in the upper leaf litter layers (Gates 1974; Jaenike, Ausubel & Grimaldi 1982; Dymond et al. 1997; Terhivuo & Saura 1997).

Information on the spread rate and areal extent of an invasive species is crucial for the development of appropriate management strategies (Abbott 2006). The average spread rate at our younger road sites (1·8 m year−1) is likely to be an underestimation of the actual rate of earthworm spread because initial introduction of earthworms may often occur several years or more after road construction. No spread rates could be calculated for the older roads, which had an average age of 46 years, because earthworms had spread to the farthest points sampled. This suggests that the average spread rate at those roads was greater than 10 m year−1, particularly if initial introduction did not occur soon after construction. Although we were only able to calculate approximate rates of earthworm spread from our data, our results suggest that the rate of earthworm spread in the boreal forest is similar to or higher than rates reported in other areas (~5–10 m year−1) (Marinissen & van den Bosch 1992; Hale 2004).

modelling

Our model of earthworm spread in the Al-Pac FMA of north-eastern Alberta suggested that approximately 49% of suitable forest habitat will be occupied by earthworms in the next 50 years. Due to the ability of earthworms to act as ecosystem engineers, significant changes in the structure and functioning of this substantial area of boreal forest habitat may be expected in the near future. Furthermore, as development intensifies in the boreal forest across Canada, invasions are likely to occur on an even broader scale. Because there are no known methods of controlling earthworms once they have invaded an area, it is critical that measures (e.g. public awareness campaigns) be taken to reduce further human-mediated introductions in such systems (Callaham et al. 2006).

Our model was based upon a number of assumptions that may have affected our estimate of the future extent of earthworm invasion. First, we assumed that there were no barriers present that could prevent earthworm spread within the buffered areas. Some suitable forest habitats within the buffers may have been isolated from previously invaded stands by, for instance, black spruce or tamarack stands such that earthworms would be unable to spread into them. Secondly, introduction rates may vary depending on road improvement or traffic levels. In addition, introduction may occur along waterways, or near other anthropogenic features such as pipelines or seismic lines (Schwert & Dance 1979; Cameron et al. 2007). Thirdly, we set 10 m year−1 as the spread rate in our model but our local-level data suggest that spread may occur more quickly than 10 m year−1. Finally, the nature of the future road network in Alberta is difficult to ascertain as energy resources are often clustered.

conclusions and implications

The boreal forest of western Canada is viewed by many as one of the last great wildernesses in the world. However, the nature of this ecosystem is changing fundamentally as world-wide demand for timber and energy products has resulted in record levels of industrial development in the last 15 years. Recent acknowledgement of the oil sands as a proven oil reserve by the US Department of Energy has resulted in the size of Canada's reserves advancing from the 21st to the second largest in the world (Radler 2002). This recognition has led to changes in policy and, subsequently, unprecedented rates of road construction in Alberta. The environmental impacts of roads and other linear features in this region have typically been viewed in terms of how they impact native biodiversity. However, our results demonstrate that the effects of roads on non-native species may be equally important.

While an increased rate of earthworm invasion seems to be a consequence of roads, the effects on the boreal forest remain unclear. In our study area, there were no correlations between earthworms and vegetation variables that we thought might change with earthworm invasion (i.e. leaf litter depth). However, invasions of Alberta's boreal forest by earthworms are relatively recent and changes in vegetation structure may yet occur given more time. In addition, major changes to ecosystem function and structure often occur when endogeic and anecic earthworms invade. In northern Alberta, these species appear to be currently localized near boat launches (Cameron et al. 2007). However, Hale, Frelich & Reich (2005) hypothesized that changes to forest systems caused by litter-dwelling epigeic species facilitate the invasion of endogeic or anecic species. This suggests that the appropriate conditions for an invasional meltdown may exist, whereby the first invasive species introduced into an ecosystem can alter conditions in such a way as to facilitate invasion by other species (Simberloff 2006).

Reducing the number of new roads being constructed in Alberta is likely to be a key way to decrease earthworm introduction and spread. However, our study suggests earthworms are more often introduced by vehicle traffic than during initial importation of gravel and soil during construction. Therefore, earthworm introductions could also be reduced by restricting the amount of traffic on roads or by reclaiming temporary roads whenever possible. Restrictions on road use have been considered in northern Canada to protect sensitive wildlife species. However, enforcement of these policies has been lax due to political pressure from special interest groups. Our results add to the growing number of threats that we are failing to address by ignoring the science that shows roads have detrimental impacts. The very large area of northern Alberta that earthworms are likely to invade in the near future, if current rates of development persist, make it important to try to slow the introduction rate of this ecosystem engineer.

Acknowledgements

We thank N. Melnycky, J. Reimer, T. Williams, E. Buss, A. de Leon, Y. Ma, and B. Cameron for data collection assistance. M. C. Arienti and D. Park provided valuable advice on ALCES modelling and aerial photograph interpretation, respectively. This research was supported by the Integrated Landscape Management Group at the University of Alberta, Canadian Circumpolar Institute, Northern Scientific Training Program, Alberta Sport, Recreation, Parks and Wildlife Foundation, a Natural Sciences and Engineering Research Council Canada Operating Grant to EMB and Canada Graduate Scholarship M to E.K.C., and an Alberta Ingenuity Scholarship to E.K.C.

Ancillary