Experimental treatment of a selection of individuals with limited clonal extension was set up in four heavily infested dune sites (Figs 1, 2; Table 2). Individual B. aquifolium plants selected for treatment were located with a hand-held Garmin Foretrex 401 GPS, photographed and marked with a flag so as to easily relocate them in the field after treatment. Height (cm) and diameter (cm) were measured and the number of stems was counted to get an idea of the dimension of each individual or clone. Within the selected plants, clone diameter was 85 cm on average (minimum 10 cm, maximum 6 meter). In April/May 2013, plants were subjected to one of the following management treatments: (1) manual uprooting by digging with shovels (DIG), (2) leaf treatment using a spray bottle with a 5% Roundup Max (450g/l glyphosate) solution (LEAF), (3) stem treatment (cut and paint) with the same glyphosate solution (STUB) or (4) stem treatment with a saturated salt (NaCl) solution (SALT). Each ramet of a clone was treated similarly. We alternated treatments in sequence to different plants. When one plant received a given treatment, we moved to the next plant and applied the following different treatment, making sure plants were sufficiently far apart (minimally 10 meters) so as not to treat the same plant clone and making sure the most closely located plants got different treatments. The 5% glyphosate solution corresponds with the recommended concentration for cut stump treatment of P. serotina which is higher than the recommended concentration (1.5–2%) for leaf treatment (Agentschap voor Natuur en Bos 2013). The salt treatment is regarded as an environmentally friendly alternative because the used quantities of salt, considered on a m² basis, result in soil salt concentrations far below the natural values (Rozema et al. 1983). The direct effect of treatments was compared in terms of categories of stem regrowth (dead, limited regrowth, vigorous regrowth), after six months following treatment (November 2013) and after one year (May 2014). Glyphosate application was performed on rainless days with an outside air temperature below 25 °C. Since January 2015 the use of herbicides in Flanders has been banned in areas that belong to or are used for public services, areas located in drinking water protection zones, in a zone of six meters alongside surface water such as canals, waterways and ponds and on roadside verges (Decision of the Flemish Government laying down detailed rules for the reduction of pesticide use by public services of 19 December 2008). A derogation on this ban for invasive species removal is subjected to specific permits. For the purpose of this experiment we obtained permission from the competent Agency for Nature and Forest who was also the owner of the sites. In total, 127 clones were treated. Unfortunately, nine could not be retrieved, probably because flags were removed by site visitors. Nonetheless, the resulting 118 clones measured were more or less equally distributed over sites and treatments (Table 2, Fig. 2).

A minority of treated plants (8 out of 118) showed limited regrowth after one year. We therefore lumped limited and vigorous regrowth and considered those plants as vital after treatment. The product of plant diameter and plant height was used as a proxy for plant volume. We then investigated the relationship between kill rate (%), treatment (as a categorical variable with 4 values), dune area (as a categorical variable with 4 values), plant volume and number of stems using a generalized linear model with a binomial distribution and logit link (Quinn and Keough 2002) since we had a two-level response (dead or vital regrowth). We log transformed plant volume and number of stems to account for their skewed nature. We checked for correlations between factors in the model using Pearson’s product-moment correlations (number of stems * plant volume), Pearson’s Chi-squared test (dune area * treatment), Fisher’s Exact Test for Count Data (regrowth * dune area and regrowth * treatment) and analysis of variance for plant volume and number of stems with dune area and treatment. We performed a multiple comparisons Tukey test to compare treatments, with a simultaneous p-value at 0.05. All analyses were performed in R version 3.1.2 (R Development Core Team 2014).

The data underpinning the analysis reported in this paper are deposited in the Dryad Data Repository at https://doi.org/10.5061/dryad.zkh189361.

Ornamental exotic species are increasingly causing problems for native biodiversity in Belgian coastal dunes. As the dunes are highly fragmented by urban development, the gardens surrounding the sites are the primary source of these plant species (Verloove 2002). However, because large populations of invasives have built up within the dunes, these currently most probably act as the major secondary seed source causing further expansion (Fig. 2). Most of the problems caused by invasions are due to, at most, 15 shrub or tree species. Differences in growth form and ecology urge for a variety of removal techniques. Important characteristics for invasiveness are a high dispersal potential and the capacity for vegetative reproduction. Berberis aquifolium is among the worst in these respects, given its long-distance dispersal potential through frugivorous birds and its vegetative growth potential through root suckers. It has been widely planted for its ornamental value and functionality as an evergreen hedge plant. It shows vigorous vegetative growth and has a rooting system well adapted to the coastal dune environment. Therefore, the experience with B. aquifolium can be useful for a wider range of other invasive species such as B. julianae which has recently been observed in the Belgian dunes (pers. obs. WV).

Foliar application of a 5% glyphosate solution seemed by far the most effective way to remove isolated B. aquifolium clones, suggesting good uptake of the active compound through the stomata despite the species’ glossy, leathery leaves. Manual uprooting of individuals is labor intensive and regrowth from thin root or stolon fragments is nearly inevitable. Cutting, even with glyphosate stem treatment, gives poor results and therefore seems inadequate. However, our experiments only included a single cut and we do not know the results of several years of mowing. Also, the results of the described experiment only give an impression of the aboveground regrowth after one year. Excavation of a number of individuals revealed that some roots did reshoot, even if the aboveground parts of the plant looked completely dead. Kill rates were therefore probably overestimated and retreating sites will probably be necessary in order to obtain complete removal or at least more accurate figures on the success of each treatment.

The results described here are in contrast with Stahl and Schwab (2014) who measured no effect of chemical treatment in invasive B. aquifolium populations of orchid-rich forest in central Germany. In this study, the herbicide used was Clinic (Nufarm) in a 33% glyphosate concentration which was applied in early July on the leaves with a paintbrush (Stahl and Schwab 2014). This higher concentration may have hindered effective uptake of the product by the leathery leaves or differences in environmental circumstances (e.g. shade, soil conditions) might explain the contrasting results. In our study, we did not consider control plants as the emphasis was on establishing causal relationships between a set of treatments and an outcome and had hardly perceived any natural fall-out during field surveys prior to the experiment. Due to the lack of control plants, we can, however, not entirely rule out that kill rates of our glyphosate treatments would have been overestimated, although leaf-treated plants consistently turned brown within days after treatment. Furthermore, it should also be noted that the exact identity of invasive Berberis hybrids/cultivars in Belgian coastal dunes is unknown for the moment (see introduction). Ross et al. (2009) found no evidence for local adaptation of different populations in central Germany, but for the time being it cannot be ruled out that Belgian populations are of a different genetic constitution and therefore could react differently to management treatments.

Non-target effects of a proposed management method on the environment, economy or society are important to consider when deciding on management options for invasive species (Booy et al. 2017). This is especially true in nature reserves where the use of herbicides might impact other species of concern or may face public opposition. Since leaf application requires more glyphosate than the other control techniques, this method has more potential for non-target effects. However, visual inspections showed very little collateral damage around treated plants. This is logical as application on the leaf was performed with a hand sprayer on days with little wind and therefore was very precise. Moreover, non-target effects of the other chemical as well as mechanical treatments on non-target plant species, invertebrates and soil biodiversity cannot be ruled out and remain undocumented.

Optimizing the use of herbicides currently seems the most appropriate way to tackle B. aquifolium. The technique of leaf spraying is particularly useful in dunes which are inaccessible for heavy machinery, or in situations where mechanical removal using machines is inappropriate because of great conservation value or sensitivity of the local habitat. In our experiments, we followed the general advice for glyphosate application in P. serotina management. Herbicide treatments were performed on days without rain to prevent solution run-off from the leaves and on days with an outside air temperature below 25 °C to maximize the efficiency of the active compound glyphosate. Further experiments should be carried out, however, testing different types and concentrations of herbicide and optimal treatment timing and conditions. As B. aquifolium is an evergreen shrub and its flowers are conspicuous, it can easily be detected throughout the year. However, it often occurs under or in between native Hippophae rhamnoides where plants can be a lot harder to detect or (re)treat and where non-target effects of control are more difficult to prevent. Also, some of the dune areas are grazed by introduced cattle as a management technique. The effect of grazing on B. aquifolium is unknown although grazers can defoliate older plants. The choice of method should reflect on the characteristics of each site. For example, the experiments described here were performed in nature reserves and were therefore not allowed during the breeding season. Although potentially more cost effective and broadly applicable, chemical control might not be the preferred option everywhere. We acknowledge many questions might need to be addressed before chemical control can become a viable option. For example, the use of herbicides is heavily restricted near areas used for drinking water extraction, several of which are located in the coastal dunes. Also, horizon scanning of new potentially effective compounds (e.g. triclopyr) and products (e.g. aquamaster, agridex) is often hindered by legal constraints.

As our experiment focused on individual shrubs, it offers prospects to more effectively control scattered clones which currently still represent a widespread type of infestation in Belgian dunes. However, in high density areas, almost entirely covered by B. aquifolium, a different management approach might be required for various reasons. The removal of large surface areas of high density B. aquifolium requires landscape scale measures that often involve mechanical removal with heavy machinery rather than manual removal. As a demonstration project, a heavily infested area with 100% B. aquifolium cover in the Noordduinen, was mechanically removed over a surface area of 350 m² in November 2013 using a 42 tons excavator equipped with a barred shovel aiming at sifting sand from plant material (Suppl. material 1). This was accompanied by intensive manual raking (approximately 9 man hours) which enabled the removal of most of the smaller remaining stolon fragments. Thickets and plant remains were removed using a tractor and trailer fitted with low pressure tires and using a fixed route in order to limit track formation and damage to adjacent areas. The site was revisited in May 2014 and the outcome measured in terms of B. aquifolium regrowth from different depths. The rooting system appeared to be relatively shallow (30–40 cm). The limited regrowth from superficially buried stolon fragments could easily be pulled out. This shows large patches of dense B. aquifolium can be removed mechanically. Several hundreds of square meters per day could be harvested, depending on the terrain conditions (relief) and soil moisture content. Even though the use of herbicides is strictly regulated in Flanders, and the potential for non-target effects was considered high with large-scale application, mechanical removal was preferred in this situation. Evidently, this method should fit the nature management goals of the area as it removes all vegetation and completely disturbs the soil profile. As very few native plant species can survive under the dense and evergreen B. aquifolium cover, botanical losses are generally limited. In our experiment, only Rosa spinosissima L.was of conservation concern and was also removed. Soil disturbance can also be seen as an opportunity for landscape-scale dune restoration. All over northwestern Europe, fixation and landscape senescence is seen as a threat to the specific biodiversity of coastal dunes (Arens and Geelen 2006, Provoost et al. 2011). As such, the removal of invasive plant species, and notably scrub, can be a lever for landscape rejuvenation and various management options are available (Day et al. 2003). Working in dry conditions is essential when performing this type of removal, as these facilitate the separation of soil fraction and plant material. Care should be taken to correctly dispose of the plant material which in this case involved a tractor with low tire pressure (Suppl. material 1). Manual aftercare on site and revisiting the sites during the next growing season is essential, as some regrowth of B. aquifolium from stolons cannot be ruled out. However, any remaining or new shoots mostly originated from superficially buried fragments and could easily be pulled out by hand. It should be noted that only very limited regrowth was observed at the treated site six months after removal. This was still the case during field visits in summer 2015 (personal observation WVG) and is in sharp contrast with similar mechanical removals of other invasive plant species in Belgian dunes such as Rosa rugosa and Ribes aureum Lindl. which seem to have a higher potential to reshoot. The success of the mechanical scrub removal also contrasts with the manual digging treatments performed on individual clones, where large regrowth was observed. It is possible that the crane can dig up the entire root system while manual digging does not remove all of the roots.

The potential of B. aquifolium to become invasive in Belgian dunes was already predicted by Verloove (2002) who also advised quick removal at the time. Also, permanent monitoring of biodiversity in the Belgian coastal dunes revealed an urgent need to tackle invasive shrub invasions (Provoost et al. 2010, 2015). Berberis aquifolium is currently still confined to the western coastal dune areas (Fig. 2). In order to prevent its further spread to the eastern parts of the Belgian coastal dunes, removal of the current populations acting as sources of secondary spread is urgent. Berberis aquifolium is subject to risk communication and recommendations towards the general public as well as horticulture professionals within the framework of the Belgian Life+ project AlterIAS, in order to limit its use near habitats of high conservation value (Halford et al. 2014). This incentive should further be put into practice near coastal dunes in order to raise awareness with local horticultural stakeholders (garden centers, horticulturists, park managers etc.), public bodies and private owners to prevent the species from being planted and used in gardens and public greenery near coastal dune reserves. In parallel, the potential of promoting native alternatives such as B. vulgaris can be explored.

On 1 January 2015 Regulation 1143/2014 on the prevention of the introduction and spread of invasive alien species (IAS) entered into force which prohibits trade and possession of invasive species on a Union List and enforces surveillance, rapid eradication, prevention and management actions on them. However, with the exception of B. halimifolia L. and A. altissima Swingle, none of the problematic shrub and tree species mentioned in this study, nor detrimental dune invasives such as Carpobrotus edulis (L.) L. Bolus or Acaena novae-zelandiae Kirk, are on the current list of regulated species. The drafting of a list of IAS of regional concern for Atlantic dunes could be a good alternative to prevent establishment of invasive species detrimental to this unique ecosystem and to prioritize action on already established invasives. The drafting of such regional lists should be based on sound risk assessment methodologies (Gallardo et al. 2015; Roy et al. 2018) but the prioritization should also properly consider risk management options and their feasibility (Booy et al. 2017; Vanderhoeven et al. 2017).