Quantifying the recent expansion of native invasive rush species in 1 a UK upland environment 2

12 Rushes, such as soft rush (Juncus effusus L.), hard rush (Juncus inflexus L.) and compact rush 13 (Juncus conglomeratus L.) have become problem species within upland grasslands across the 14 UK and the coastal grasslands of western Norway. Indeed, being largely unpalatable to 15 livestock and having a vigorous reproductive ecology means that they can rapidly come to 16 dominate swards. However, rush dominance results in a reduction in grassland biodiversity and 17 farm productivity. Anecdotal evidence from the UK suggests that rush cover within marginal 18 upland grasslands has increased considerably in recent decades. Yet, there is currently no 19 published evidence to support this observation. Here, we use recent and historical Google Earth 20 imagery to measure changes in rush frequency over a 13-year period within four survey years: 21 2005, 2009, 2015 and 2018. During each survey year, we quantified rush presence or absence 22 using a series of quadrats located within 300 upland grassland plots in the West Pennine Moors, 23 UK. Data were analysed in two stages, first, by calculating mean rush frequencies per sample 24 year using all the available plot-year combinations (the full dataset), and second by examining 25 differences in rush frequency using only the plots for which rush frequency data were available 26 in every sample year (the continuous dataset). The full dataset indicated that rush frequency 27 has increased by 82% between 2005 and 2018. Similarly, the continuous dataset suggested that 28 rush frequency has increased by 174% over the same period, with the increases in frequency 29 being statistically significant (P<0.05) between 2005-2018 and 2009-2018. We discuss the 30 potential drivers of rush expansion in the West Pennine Moors, the ecological and agronomic 31 implications of grassland rush infestations, and priorities for future research. 32 33

In contrast, we do know about the reproductive ecology of rushes. For example, they 48 can produce between 4500 and 8500 seeds per stem per year (McCarthy, 1971;  Derda et al., 2014), which, on rush infested ground, equates to approximately 4 to 6.7 million 50 seeds per square metre per season (Moore and Burr, 1948;Ervin and Wetzel, 2001). To produce 51 such large amounts of seed, a single rush plant only uses 0.27% of its annual net biomass 52 production (Ervin and Wetzel, 2001). Depending on species, seeds ripen between July and 53 September and are shed (mainly by the wind during dry conditions) up to the following spring 54 (Richards and Clapham, 1941a, b, c). After shedding, seeds can remain dormant at the soil 55 surface for up to 60 years (Moore and Burr, 1948), and, during this time, they may be dispersed 56 by wind or surface run-off and/or germinate in areas disturbed by cultivation or livestock 57 poaching (Agnew, 1961;McCarthy, 1971;Cairns, 2013). Once established, rushes persist for  The vigorous reproductive ecology of rushes may be a contributing factor behind their 62 recent invasion of upland grasslands across the UK and the coastal grasslands of western 63 Norway (Cherrill, 1995;Østrem et al., 2018). Indeed, there is anecdotal evidence from farmers 64 and ecologists in the UK of rush infestations within upland grasslands (Hamilton et al., 2018). 65 Such infestations are problematic because they significantly reduce the agricultural and

89
We chose to measure rush expansion within the WPM SSSI for two reasons. First, the 90 SSSI contains large areas of marginal grassland, i.e., semi-improved and enclosed permanent 91 pasture at or below the moorland line (above this line the land is generally unimproved and 92 unenclosed). These grasslands are vital to hill farmers because they tend to be the most 93 productive areas of their farm (Mansfield, 2008;Nielsen and Søegaard, 2000). Also, by 94 providing suitable nesting habitat, marginal grasslands can support large populations of wading 95 bird species (Baines, 1988 119 We decided to use aerial imagery instead of field surveys because there is a lack of 120 historical field data on rush expansion within the marginal grasslands of the WPM SSSI.

121
Furthermore, while field surveys are likely to be more accurate, rush expansion can be 122 measured more efficiently using aerial imagery, which means that larger areas of grassland can 123 be surveyed. Furthermore, the use of aerial imagery is much more convenient for sampling 124 more remote or inaccessible areas and you do not require prior permission from landowners. fast-growing grasses such as Lolium spp., and also white clover (Trifolium repens), on fertile, 134 neutral soils. Improved Grasslands are typically either managed as pasture or mown regularly 135 for silage production" (NERC, 2017).

136
In total, 340 improved grassland parcels lay within or intersected the WPM SSSI 137 boundary. However, 40 grassland parcels were excluded from our survey because Google Earth 138 imagery revealed that non-grassland habitats constituted ≥ 25% of their extent. We used the 139 remaining 300 grassland parcels as discrete sampling units in which we measured temporal

155
A total of 205 high-resolution Google Earth images were downloaded ( Table 2). All 156 images were selected from an eye altitude of 1 km while all Google Earth layers were switched 157 off. Also, before a Google Earth image was captured, the compass and tilt were reset, and the 158 'Atmosphere', 'Sun' and 'Water surface' options from the 'View' menu were also deselected.

159
After an image was downloaded, it was imported into ArcGIS and then georeferenced. Google 160 Earth images are orthorectified, but the original images are captured using different camera 161 angles (Google Inc). Therefore, to enhance subsequent alignment, the images were  Root Mean Square (RMS) error is a measure of the difference between known locations and 168 locations that have been georeferenced, i.e., it is a measure of georeferencing accuracy. 169 Therefore, care was taken to ensure that the RMS error of each georeferenced image was <1 170 (Table 2). Additional information about the aerial images used in this study is contained within  (rather than exact) area within each grassland parcel between sample years. This is because 211 Google Earth imagery is orthorectified, but the source images are captured using different 212 camera angles, which means perfect alignment between survey years is impossible.

213
Nevertheless, the RMS error of georeferenced images was extremely low during each survey 214 year (Table 2). Furthermore, during the georeferencing process, care was taken to ensure that 215 the field boundaries of the sample grassland parcels were aligned between survey years. 216 Finally, it is also worth noting that other types of tall vegetation (e.g. thistles or nettles) may 217 look similar to rushes on aerial imagery. However, such vegetation was rare within validation 218 plots. In short, while our approach is not perfect, we believe that we have minimised error 219 sufficiently to be confident that our approach is an accurate and valid technique for measuring 220 rush frequency within marginal grasslands.   For the 91 plots for which we had continuous data, we recorded an increase in rush frequency 264 during each consecutive study year (Fig. 6a). Overall, mean rush frequency increased by  of 51.9 ± 17.2% and 53.8 ± 15.7% recorded during these periods respectively (Fig. 6b). Overall,     rush expansion) within marginal grasslands (Wathern et al., 1985;Fuller and Gough, 1999;334 Sutherland, 2002). For example, sheep grazing can increase soil bulk density and reduce soil 335 infiltration capacity within upland grasslands (Marshall et al., 2014). Overstocking of sheep 336 may also lead to poaching, especially on undrained fields with wet soils (Bilotta et al., 2007). 337 The creation of bare ground via poaching would facilitate the spread of rushes by providing 338 the germination niches required by overwintering seeds lying dormant at the soil surface 339 (Agnew, 1961;McCarthy, 1971;Cairns, 2013). Poaching induced rush germination may even 340 occur at low stocking densities in rush dominated grasslands because, due to the low

373
To truly understand if and what field-level factors are contributing to rush expansion, 374 we need to combine our satellite imagery approach with historical management data.

375
Unfortunately, accurate historical data was not available for the grassland parcels used in this 376 study, but such data is likely to be available in other areas across the UK. The expansion of rushes within upland grasslands has several negative consequences. First and 402 foremost, as rushes increase, palatable and productive grasses tend to be outcompeted.

415
Rush dominated fields, particularly bordering heather moorland, could also be a 416 significant, but currently unidentified, wildfire risk, especially given that we know rushes are 417 combustible (e.g. as highlighted by the historical practice of swaling, but also see Ghantous 418 and Sandker, 2015). Furthermore, fields in which rush cover exceeds 50% will have a 419 significant amount of biomass that is likely to become very dry (and thereby more combustible)