Long-term changes in the flora of the cereal ecosystem on the Sussex Downs, England, focusing on the years

Transcription

1 Journal of Applied Ecology 2010, 47, doi: /j x Long-term changes in the flora of the cereal ecosystem on the Sussex Downs, England, focusing on the years G. R. Potts*, J. A. Ewald and N. J. Aebischer Game & Wildlife Conservation Trust, Fordingbridge SP6 1EF, UK Summary 1. There has been a surge of interest in the effects of modern agriculture on biodiversity but studies of farmland flora have lacked continuity and historical context. Here we present the results of 38 years of annual monitoring of the weed flora of cereal crops on the Sussex Downs. 2. This study investigates the long-term changes in abundance of 214 weed species, two subspecies and one forma found in the cereal fields of a 62-km 2 area of the Sussex Downs. Species occurrence and weed abundance were recorded annually in June from 1970 to 2005 inclusive. Stubbles were surveyed in 1968, 1971, 2004 and Annual archaeophytes and perennial natives predominated and the community belonged to the Papaver rhoeas Silene noctiflora association (OV16) of the UK National Vegetation Classification. 4. Overall, 97% of fields were treated with herbicides prior to sampling, reducing dicotyledonous weed abundance by 64% and taxon occurrence by 52%. From 1970 to 2005 there was no trend in overall abundance of dicotyledons, although monocotyledons decreased by 13% relative to the early 1970s. 5. Of 66 taxa monitored from 1970 to 2005, 18 increased, 38 rose and fell (or vice versa) and 10 showed no trend. Annuals increased until the early 1980s, when many were not susceptible to herbicides, before levelling off or declining slightly as the efficacy of herbicides expanded. 6. Perennial dicotyledons increased steadily throughout the study. This latter change was due to the loss of traditional leys, not to changes in herbicide efficacy. 7. Ninety-two species of dicotyledons were found on stubbles, with no significant overall change in occurrence from to In both stubbles and crops, species uncommon at the start have tended to increase whereas common species have tended to decrease. 8. Combining this study with earlier records, we estimate that 16 weed species have been lost from the study area and 15 gained. Before 1970, the loss rate of archaeophytes and the gain rate of neophytes were both higher than for other species. Most species lost were historically uncommon whereas many of the species gained are now common. 9. Synthesis and applications. The soil seed bank remains sufficient to enable a rapid restoration of the pre-herbicide flora where needed for wildlife conservation purposes, without enhancement, i.e. seeding. The means to do this are available through the UK s agri-environment in-field measures, but these are very unpopular with farmers. Incentives need to be much improved to ensure the future conservation of the traditional arable flora. Key-words: agroecology, cereal ecosystem flora, conservation, crop biodiversity, farmland biodiversity, herbicides, weeds Introduction *Correspondence author. twyneham1@btinternet.com Cereal growers have worked for at least 13 millennia to eradicate weeds by seed-cleaning, hand-weeding, hoeing or rotational cropping (Hillman et al. 2001). Despite this a characteristic flora evolved symbiotically with cereals until the commercial use of hormone weed-killers. Most farmers hoped that dominant annuals such as Papaver rhoeas and Sinapis arvensis would be eradicated by these herbicides and the initial results were spectacular as herbicide usage increased through the 1950s (Salisbury 1961; Woodford 1964). Massive losses of floristic diversity were documented in Germany during Ó 2009 Game & Wildlife Conservation Trust. Journal compilation Ó 2009 British Ecological Society

2 216 G. R. Potts, J. A. Ewald & N. J. Aebischer (Bachthaler 1969) and (Rademacher et al. 1970), in England from 1951 (Fryer & Chancellor 1971) and in Hungary in 1953 (Ubrizy 1968). Predictions followed that crops may become free of weeds as herbicides became more effective (Sagar 1968). More recently, however, there have been reports that the diversity of the cereal flora increased from 1980 to 2000 in Finland (Hyvo nen, Ketoja & Salonen 2003b; Hyvo nen et al. 2003a), from 1987 to 2004 in Denmark (Andreasen & Stryhn 2008) and was at its highest during in the Czech Republic (Lososova et al. 2004). There were favourable trends for many arable weeds in the UK between 1987 and 2004 (Braithwaite, Ellis & Preston 2006) with some species reported from cereals for the first time, e.g. Lactuca serriola (Sutcliffe & Kay 2000; Last 2004). In summary, the long-term effects of changes in agriculture on the cereal field flora have become unclear. Established in 1968, our study on the Sussex Downs in southern England was designed to monitor, evaluate and explain long-term changes in the cereal ecosystem, measuring changes in the abundance of weeds, insects and birds on farmland. Herbicides were first used on the area in 1946, by % of fields were treated annually and by 1968, 90%. Interim broad-brush results about the flora have been reported previously (Potts 1970, 1986, 2003; Potts & Vickerman 1974; Aebischer 1991; Ewald & Aebischer 2000) but this paper is the first detailed review of arable flora in the Sussex study. Materials and methods The Sussex study area consists of 62 km 2 of farmland on the hills known as the Sussex Downs, situated between the rivers Adur and Arun along the south coast of the UK. The dominant soils are chalk rendzinas with abundant flint, isolated caps of clay on higher parts and post-glacial deposits along the lower parts of a series of dry valleys. We defined the cereal weed flora as any flowering plant within a cereal crop or stubble (up to November), but excluding the seedlings of shrubs or trees, the crop and the remnants of previous crops. Species were designated as annual, biennial or perennial and classified as archaeophyte, neophyte or native using Preston, Pearman & Dines (2002). Some species were difficult to identify unless flowering or fruiting and had to be dealt with at the genus level, e.g. Fumaria. Nomenclature follows Stace (1999). Some taxa remained unresolved until the later stages of the study, e.g. Polygonum aviculare agg. For the consistent analysis of long-term changes, such taxa were retained after being resolved. The cereal flora was studied in four ways. IN JUNE IN GROWING CROPS: INCLUSIVE Each year all cereal fields (average 106 per year for 36 years) were sampled on or as soon after 16 June as weather permitted; this required a minimum of 2 days. Sample sites were consistently in the same locations within fields with the number of fields sampled varying according to the availability of wheat, barley or oat crops across the 12 farms of the area. The main purpose of the visit to each field was to sample invertebrate abundance, with a D-vac at five 4-m intervals within the outer 20 m of the field. The flora was recorded from the immediate vicinity of the five places where the D-vac had been placed to sample invertebrate abundance. The crops were tall at the time of sampling and quadrats were not practicable. Areas subject to shading by trees or tall hedges were avoided as were gateways. For the flora the objective was to record weed species occurrence and assess general weed abundance. General weed abundance The taxa were recorded in two groups (i) dicotyledons dicots and (ii) monocotyledons or grasses monocots. The abundance of the plants in each group was assessed and ranked in six categories: 0 = no weeds; 1 = only one weed specimen found; 2 = few weed specimens, not obvious; 3 = weed specimens numerous and obvious; 4 = weed specimens as abundant as crop; 5 = weeds dominating crop; crop not obvious. Species diversity was ignored in this ranking. Small seedlings were excluded. The senior author made all the decisions regarding identification and abundance, preserving consistency throughout. CEREALS NOT TREATED WITH HERBICIDES The usage of herbicides was recorded by farmers and the records made available to us (see Ewald & Aebischer 2000). Throughout the study 99 fields (2Æ6% of the total) were sampled in June that had not been treated with herbicides and the results from these samples were combined with those from remaining samples when examining longterm trends. The frequencies of dicots in the untreated fields were used to determine the phyto-association as per Rodwell (2000). We calculated the log ratio of the average occurrence of taxa in crops treated with herbicides to those in crops not treated with herbicides, as well as the ratio in occurrences between early untreated crops ( ) and later untreated crops ( ) to measure the long-term effects of herbicide treatment. STUBBLE SURVEYS In mid-october 1968, 1971, 2004 and 2005, surveys were carried out of the dicots on cereal stubbles. Ten randomly placed quadrats each of 0Æ83 m 2 (1 yd 2 ) were placed at 25 m intervals along random transects across each field, excluding the outer 20 m. The presence of dicots was recorded. Thirty fields were chosen each year across the five farm units (see below). From 1972 this work was suspended as more than 90% of non-undersown stubbles were burned after harvest. This practice was banned in Government grants to maintain over-winter stubbles were adopted in parts of the study area in By 2004 there was a representative spread of stubbles and the surveys were reinstated. The data for 1968 and 1971 are combined as are those for 2004 and 2005 and we calculated the ln-transformed ratio of the occurrences of taxa in the latter samples to those in the early ones to see if there have been any changes in the stubble flora between the two periods. These data were analysed separately from the June surveys. NON-SYSTEMATIC SURVEYS: SPECIES GAINS AND LOSSES Throughout the study the senior author recorded species not encountered during the systematic surveys. Also included were the records in Wolley-Dod (1937) that referred to specific sites within the study area. The Sussex Botanical Recording Society surveyed the area between 1966 and 1978 (Hall 1980) with many original records made available to us, and recording of rare species has

3 Long-term changes in arable flora 217 continued (Briggs, Harmes & Knapp 2001). We collated the number of species lost in three time periods: before the first widespread use of herbicide (1946), from the first widespread use of herbicide until the start of monitoring for the Sussex study ( ) or during monitoring (from 1968). A species was considered lost during our monitoring if it had not been recorded since We collated species gained similarly. CHANGESINCROPROTATIONS As before, we divided the study area core into five farm groupings (Aebischer 1991) with the type of farming varying from traditional ley rotations (with undersowing resulting in grazed pastures, and spring fallow), to mixed arable and grass (without undersowing and using break crops such as oilseed rape and linseed) to continuous winter wheat (Fig. 1). Each year we characterized the farming system of each of the farm units recording changes. HERBICIDE EFFICACY From 1970 to 2004, we collected data on herbicide use through on-site interviews with farmers and their agronomists. We collated information on active ingredients used, season of treatment (autumn or spring) and rates of usage. For each herbicide, the manufacturers have designated certain plant taxa as susceptible, mostly susceptible, mostly resistant or resistant. We accepted the manufacturer s designations. STATISTICAL ANALYSIS We used the key and diagnostic species frequencies of candidate floras in Rodwell (2000) to determine the phyto-sociological association. We used correlation analysis to compare candidate species frequencies given by Rodwell et al. from the National Vegetation Classification with those in our untreated samples and diagnosed the community from the highest Pearson correlation coefficient. For the yearly average of the abundance ranking, the average of the total number of taxa per sample, the number of dicot and monocot taxa per sample and the number of annuals and perennials, we used generalized additive models (GAM) with a normal error and identity link (weighted by number of samples per year) to examine the long-term trends in average values per year (Hastie & Tibshirani 1990). GAMs were used instead of polynomial models (e.g. quadratic or cubic functions of year) as they allow greater flexibility in describing the changes in the response variable (in this case abundance or number of taxa) through the course of time (Fewster et al. 2000). We used smoothing splines to construct GAMs with zero to 2 d.f. (corresponding to the maximum degree of the spline s polynomial parts) and used Akaike s Information Criterion (AIC, Akaike 1974) to select the best fitting model (Whittingham et al. 2006). We limited the GAMs to a maximum of 2 d.f. because our primary objective was to identify broad long-term trends, not short-term fluctuations. We particularly wanted to determine when each trend was at a maximum during the 36 years of continuous monitoring, and limited our GAMs to 2 d.f. to accomplish this. Trends in the percentage occurrence of weed taxa were examined using GAMs with a binomial error and a logistic link function, again using smoothing splines with 0 2 d.f. and AIC to determine the bestfitting model. We used general linear models (GLM) weighted by number of samples per year to examine the effect of farm management and herbicide efficacy on trends in the number of annual and perennial dicots and monocots recorded in each sample. We examined the effect of taxon-specific herbicide efficacy on weed taxa whose trends in occurrence rose, then fell (two-degree GAM). Yearly herbicide efficacies (measured as the percentage of fields where the taxon was susceptible to the herbicide regime used) were grouped into three 7-year periods corresponding to (i) the peak phase, centred on the year where the taxon had the highest smoothed value, with 3 years either side, (ii) the 7 years preceding this the increase phase, and (iii) the 7 years following the peak phase the decrease phase. We used analysis of variance (anova) to compare the susceptibility of the taxon to the herbicides used in the three phases. We restricted this Fig. 1. Changes in crop rotation on the five farm units in the Sussex study area, Farm units 1 (a) and 3 (c) have followed an arable and grass regime not necessarily rotating with no undersowing throughout (AG). Farm unit 2 (b) changed from traditional ley rotations (TL) to AG management in Farm unit 4 (d) went from TL management to AG management in 1983, followed and then switched to mainly continuous winter wheat (MW) in Farm unit 5 (e) has continued with TL management throughout the time period. (a) Year Year Cropped area (%) (c) Cropped area (%) (e) Cropped area (%) (b) Cropped area (%) (d) Year Year Cropped area (%) Year Winter wheat Winter barley/oats Spring cereal Set-aside Miscellaneous Grass (rotational) Grass (non-rotational)

4 218 G. R. Potts, J. A. Ewald & N. J. Aebischer analysis to taxa whose highest smoothed values were during , in order to have sufficient data for the increase and decrease phases. For each individual taxon we estimated the relative change from 1970 to 2005 by taking the GAM fitted to the percentage occurrence as above, and calculating the ratio of the fitted value from 2005 to that from In order to evaluate the change in occurrence of species between the beginning and end of the study period, the geometric mean of these ratios was compared to 1, denoting no change. This was tested statistically after transforming the ratios to logarithms and comparing their mean to log(1) = 0 using a t-test. If the ratio was significantly greater than 1 it indicated an increase in occurrence and vice versa. We divided the data set into taxa whose percentage occurrence in 1970 was equal or above the 1970 median value (referred to as common taxa) and those whose percentage occurrence was below it ( uncommon taxa). We compared the geometric mean relative change of taxa in the two groups using a t-test on the log-transformed ratios; if they were significantly different we compared the relative change of each group to 1 (i.e. no change) as above. A similar approach was used to compare the percentage occurrences of taxa in crops not treated with herbicides to those treated with herbicides, and to compare percentage occurrences in stubbles in to those in stubbles in We compared the relative number of archaeophyte, neophyte and native species lost and gained from the study area using chi-square analysis. We used Genstat 10.2 (VSN International Ltd, Hemel Hempstead, UK) for the GAM and GLM analyses and Systat 12 (Systat Software Inc., Chicago, IL, USA) for the t-tests, anova and chi-square analyses. Results THE FLORA A total of 214 species, two subspecies and one forma were identified in the cereal weed flora of the Sussex study area (Appendix S1, Supporting Information). Of this, 116 were annuals (53%), including 63 (54%) archaeophytes, 40 (34%) native species and 13 (11%) neophytes. The combined biennials and perennials numbered 101 including eight (8%) archaeophytes, 79 (79%) native species and 14 (14%) neophytes. The flora in the cereals not treated with herbicides was determined to be the P. rhoeas Silene noctiflora association (OV16) of light well-drained calcareous soils (Rodwell 2000). The fit of the 41 species frequencies of this association to those in the untreated crops during was significant (r 40 =0Æ81, P <0Æ001). JUNE SURVEYS: LONG-TERM CHANGE IN OVERALL ABUNDANCE OF WEEDS We identified 134 species in the mid-june samples, combined (where necessary) into 66 taxa. The mean dicot abundance for the herbicide-treated fields was 1Æ2 ± 0Æ01, 64% less than the 3Æ3 ±0Æ12 for the untreated fields (t 3634 = )20Æ06, P <0Æ001). There was no difference in the mean monocot abundance between the treated (mean 1Æ2 ±0Æ02) and untreated fields (mean = 1Æ2 ±0Æ15, t 3361 = )0Æ31, P = 0Æ759). The long-term overall abundance of dicots showed no detectable trend from 1970 to 2005 (Fig. 2a; although 2005 was relatively high, its removal did not affect this overall conclusion). There was stability or a slight increase in monocot abundance through the 1970s with a downward trend subsequently, best described by a two-degree GAM (Fig. 2b). JUNE SURVEYS: LONG-TERM CHANGES IN AVERAGE NUMBER OF TAXA PER SAMPLE For both dicot and monocot taxa the long-term trend in the average number of taxa per sample increased from the beginning of the study to the mid-1980s, with dicots peaking in 1984 (this conclusion is not affected by the peak in 2005) and monocot taxa in 1988 (Fig. 3). From the mid-1980s, the average number of taxa in both groups levelled off, though dicots show a sharp, short-term increase in For annual dicots the average number of taxa per sample increased through the 1970s, reaching a maximum in 1984 and then levelling off, though with the sharp increase in 2005 (Fig. 4a). For perennial dicots, there was a steady increase from the beginning of the study (Fig. 4b). Annual monocots also increased throughout the period, though this increase levelled off after 1990 (Fig. 4c), whilst perennial monocots increased through the 1970s, peaking in 1980, with a decline thereafter (Fig. 4d). JUNE SURVEYS: CHANGES IN NUMBER OF TAXA IN RELATION TO HERBICIDE EFFICACY AND FARM MANAGEMENT Information on herbicide effectiveness was available for 80 taxa (36 annual dicots, 24 perennial dicots, 11 annual and nine perennial monocots). We examined trends in the efficacy of the Fig. 2. Overall annual abundance of (a) dicotyledonous arable weeds, and (b) monocotyledonous ones in the cereal fields of the Sussex study, For dicots, there was no trend through time. Monocot abundance best fitted a GAM with 2 d.f., as measured by AIC.

5 Long-term changes in arable flora 219 Fig. 3. Annual mean number of (a) dicotyledonous taxa, and (b) monocotyledonous taxa per sample in the cereal fields of the Sussex study, In both cases the trend was best described by a GAM with 2 d.f., as measured by AIC. Fig. 4. Long-term ( ) trends in the annual mean number of arable weed taxa per sample in the cereal fields of the Sussex study for (a) annual dicots, (b) perennial dicots, (c) annual monocots, and (d) perennial monocots. In all cases, a GAM with 2 d.f. described the trend through time best, as measured by AIC. herbicide cocktails measured as the number of annual and perennial dicots and monocots susceptible to the herbicides used. The number of susceptible annual dicots increased (Fig. 5a). For perennial dicots, there was a shallow decline in the number of susceptible taxa (Fig. 5b). For annual monocots, the number of susceptible taxa increased until the early 1990s, and then levelled off (Fig. 5c), as did the number of susceptible perennial monocots (Fig. 5d). Across weed types, only in the case of perennial dicots did the trends in the number of taxa differ significantly according to farm management (Table 1). Farms with arable and grass crops but no undersowing (no traditional leys) showed a significant increase in the average number of perennial dicot weeds throughtimewhereastherewerenosignificanttrendsinperennial dicots on farms with either traditional ley or continuous winter wheat. There was no relationship between the number of perennial dicots and herbicide efficacy, and herbicide efficacy could not explain the different trends according to management. Using the same type of analysis, the average number of annual dicots per field increased through time, even after discounting significant decreases associated with an increase in herbicide efficacy, though this was driven by farms with arable and grass crops but no undersowing (Table 1). In the case of the annual monocots per field per farm there was no relationship with herbicide efficacy and over all types of management the number of taxa increased through time. For the perennial monocots as a group there was no significant trend through time or relationship with herbicide efficacy. JUNE SURVEYS: LONG-TERM CHANGES IN THE OCCURRENCE OF INDIVIDUAL TAXA No trend was detected for 10 (15%) of the 66 continuously monitored taxa (Table 2). The 18 taxa best modelled by a GAM with 1 d.f. showed increases throughout the study in the percentage of fields where each taxon was present. The pattern in the remaining 38 taxa (58%) was more complex (Figs S1 8). There were 13 taxa where the maximum value in the trend was in the first half of the study (1987 or before) and 25 taxa reaching their maximum value in the latter half of the study (1988 and later); this did not differ from random expectations (V 1 =3Æ79, P =0Æ052). For all taxa whose trend in occurrence rose, then fell, the herbicide efficacy in the decreasing phase was around three times higher, on average, than it was in the increasing phase (Table 3). Herbicide efficacy during the peak phase was intermediate. This indicates that at least some of the humped trends in occurrence

6 220 G. R. Potts, J. A. Ewald & N. J. Aebischer Fig. 5. Long-term trends ( ) in the annual efficacy of the herbicide cocktail applied to cereal fields in the Sussex study. Trends are shown for the number of (a) annual dicots, (b) perennial dicots, (c) annual monocots, and (d) perennial monocots susceptible to the average herbicide cocktail applied to cereal fields in Sussex. In all cases, a GAM with 2 d.f. described the trend through time best, as measured by AIC. Table 1. Farm management and herbicide efficacy effects on trends in the number of arable weed taxa per sample in cereal fields on the Sussex study area, Taxa are grouped into annual dicots, perennial dicots, annual monocots and perennial monocots Slope of linear trend through time Slope of relationship with efficacy of herbicide cocktail Taxa Type of farm management Type of farm management AG (n) TL MW Test for equal trends AG TL MW Test for equal trends Annual dicots Within farm management +0Æ038 (±0Æ014)* )0Æ017 (±0Æ019) )0Æ018 (±0Æ052) F 2,123 =1Æ53 P =0Æ221 )0Æ037 (±0Æ018) +0Æ005 (±0Æ025) )0Æ023 (±0Æ021) Overall +0Æ028 (±0Æ009)** )0Æ032 (±0Æ010)** Perennial dicots Within farm +0Æ009 +0Æ002 )0Æ007 F 2,123 =10Æ21 +0Æ002 )0Æ009 )0Æ003 management (±0Æ001)*** (±0Æ001) (±0Æ006) P <0Æ001 (±0Æ007) (±0Æ006) (±0Æ015) Overall )0Æ004 (±0Æ005) Annual monocots Within farm management +0Æ011 (±0Æ004) +0Æ007 (±0Æ004) )0Æ001 (±0Æ013) F 2,120 =0Æ69 P =0Æ504 +0Æ026 (±0Æ0Æ028) )0Æ019 (±0Æ041) Overall +0Æ010 (±0Æ003)*** +0Æ025 (±0Æ018) Perennial monocots Within farm )0Æ003 )0Æ001 )0Æ007 F 2,120 =0Æ31 +0Æ014 )0Æ013 management (±0Æ002) (±0Æ003) (±0Æ006) P =0Æ732 (±0Æ010) (±0Æ017) Overall )0Æ003 (±0Æ001) +0Æ010 (±0Æ006) )0Æ010 (±0Æ060) +0Æ004 (±0Æ012) F 2,123 =0Æ27, P =0Æ764 F 2,123 =0Æ29, P =0Æ749 F 2,120 =0Æ38, P =0Æ683 F 2,120 =0Æ70, P =0Æ498 AG, arable and grass not necessarily rotating, no undersowing; TL, traditional ley rotation with undersowing and fallow; MW, continuous winter wheat. *P <0Æ05, **P <0Æ01, ***P <0Æ001. can be explained by changes in herbicide regimes, with taxa at a competitive advantage until new products or combinations of products succeeded in reducing their occurrence. For each of the 66 taxa, we compared the percentage of fields where the taxon was present at the beginning of the Sussex study (1970) with that at the end (2005, Fig. 6) using smoothed values from the GAM model. For the taxa without a significant GAM, we used a ratio of 1 (i.e. no change). The geometric mean relative change from 1970 to 2005 was 5Æ9 (95% confidence interval 3Æ5 10Æ1), significantly higher than 1 (t 65 =6Æ61, P <0Æ001). For 33 taxa that were uncommon in 1970 (i.e. the percentage of fields where they occurred was less than the 1970 median occurrence of 0Æ49%) 27 (82%) became more common, whereas for the 33 remaining common ones, 18 (55%) increased in occurrence. The average relative change for the two groups was 22Æ5 (95% confidence interval 11Æ1 45Æ4) and 1Æ6 (95% confidence interval 0Æ9 2Æ6) respectively, significantly different from each another (t 64 =6Æ24,

7 Long-term changes in arable flora 221 Table 2. Long-term trends in the occurrence of individual weed taxa in Sussex from 1970 to The year with the highest smoothed value is given for fitted GAMs with more than 1 d.f. (see Figs S1 8) No trend GAM with 1 d.f. (all increases) GAM with 2 d.f. (humped trends) Annual dicots Spergula arvensis; Fallopia convolvulus; Sinapis arvensis; Euphorbia exigua; Galeopsis tetrahit; Linaria vulgaris Papaver spp. 2 ; Fumaria spp. 3 ; Euphorbia helioscopa; Geranium spp. 4 ; Anchusa arvensis; Chaenorhinum minus; Kickxia spp. 5 ; Sherardia arvensis; Centaurea cyanus Papaver rhoeas (Max in 1996); Atriplex & Chenopodium 10 (Max in 2005); Stellaria media (Max in 1970); Persicaria spp. 11 (Max in 1970); Polygonum spp. 12 (Max in 1978); Viola arvensis (Max in 1989); Sisymbrium officinale (Max in 2005); Capsella bursa pastoris (Max in 2005); Coronopus squamatus (Max in 1997); Anagallis arvensis (Max in 2005); Aphanes arvensis agg. (Max in 1984); Scandix pectin-veneris (Max in 2001); Aethusa cynapium (Max in 2005); Umbellifers 13 (Max in 2005); Lithospermum arvense (Max in 1991); Myosotis arvensis (Max in 1973); Lamium spp. 14 (Max in 1985); Veronica arvensis (Max in 1979); Veronica persica 15 (Max in 1985); Odontites vernus (Max in 1998); Legousia hybrida (Max in 1987); Galium aparine (Max in 1995); Valerianella dentata (Max in 2005); Matricaria, Tripleurospermum 16 (Max in 1984); Senecio vulgaris (Max in 2005) Perennial dicots Ranunculus repens; Reseda lutea; Potentilla anserina; Epilobium spp. 1 Silene spp. 6 ; Convolvulus arvensis; Plantago spp. 7 ; Lapsana communis; Taraxacum officinale; Artemisia vulgaris; Tussilago farfara Cerastium fontanum (Max in 1998); Rumex spp. 17 (Max in 2005); Malva sylvestris (Max in 2005); Arctium minus sens. lat. (Max in 2005); Cirsium spp. 18 (Max in 2005) Dicot taxa with mix of annuals and perennials Annual monocots Urtica spp. 8 ; Vicia spp. 9 Sonchus spp. 19 (Max in 2005) Poa annua (Max in 1972); Avena fatua (Max in 1997); Alopecurus myosuroides (Max in 1998); Bromus hordeaceus (Max in 1999); Anisantha sterilis (Max in 1997) Perennial monocots Poa trivialis (Max in 1984); Elytrigia repens (Max in 1970) 1 Epilobium parviflorum, E. lanceolatum, E. tetragonum, E. ciliatum. 2 Papaver argemone, P. hybridum, P. dubium ssp. Lecoqii. 3 Fumaria officinalis, F. parviflora, F. densiflora. 4 Geranium dissectum, G. molle, G. columbinum. 5 Kickxia elatine, K. spuria. 6 Mainly Silene latifolia, S. vulgaris. 7 Plantago major, P. lanceolata, P. media. 8 Urtica dioica, U. urens. 9 Vicia cracca, V. hirsuta, V. sativa. 10 Atriplex prostrate, A. patula, Chenopodium album, C. ficifolium, C. polyspermum, C. rubrum. 11 Persicaria lapathifolia, P. maculosa. 12 Polygonum aviculare, P. arenastrum, P. rurivagum. 13 Anthriscus sylvestris, Pastinaca sativa, Heracleum sphondylium, Torilis arvensis, T. nodosa. 14 Lamium amplexicaule, L. purpureum. 15 Includes some Veronica polita, V. hederifolia. 16 Matricaria recutita, M. discoidea, Tripleurospermum inodorum. 17 Rumex crispus, R. obtusifolius ± equally abundant, R. acetosa. 18 Cirsium arvense, C. vulgare. 19 Sonchus arvensis, S. asper, S. oleraceus. P <0Æ001). The average relative change was significantly greater than 1 in the first case (t 32 =9Æ00, P <0Æ001) but not in the second (t 32 =1Æ77, P =0Æ086). The occurrence of dicot taxa in cereals not treated with herbicides was compared to that in herbicide-treated cereals (Fig. 7). On average, treatment with herbicides reduced the

8 222 G. R. Potts, J. A. Ewald & N. J. Aebischer Table 3. Comparisons of herbicide efficacies in the increase, peak and decrease phases of annual dicotyledons taxa showing trends in occurrence that were humped ; from Table 2 where sufficient data were available (see Materials and methods) Mean susceptibility (±SE) Taxa Peak Increase phase Peak phase Decline phase F-statistic Aphanes arvensis agg Æ2±6Æ6 a1 55Æ3±7Æ0 b 62Æ5±10Æ5 b F 2,18 =5Æ50, P =0Æ014 Matricaria, Tripleurospermum Æ8±4Æ4 a 84Æ6±5Æ6 a,b 95Æ1±2Æ4 b F 2,18 =3Æ63, P =0Æ047 Lamium spp Æ3±7Æ5 a 31Æ0±6Æ8 a 69Æ5±10Æ5 b F 2,18 =10Æ11, P =0Æ001 Veronica persica Æ2±5Æ7 a 36Æ9±4Æ9 a 81Æ3±3Æ3 b F 2,18 =34Æ25, P <0Æ001 Legousia hybrida Æ6±2Æ8 a 16Æ0±8Æ2 a 53Æ7±10Æ0 b F 2,18 =10Æ00, P =0Æ001 Viola arvensis Æ4±7Æ7 a 37Æ7±10Æ5 a 78Æ4±3Æ8 b F 2,18 =11Æ79, P =0Æ001 Poa trivialis Æ1±7Æ4 a 55Æ7±7Æ6 b 75Æ8±2Æ7 c F 2,18 =14Æ40, P <0Æ001 1 Means with the same letters are not significantly different at P <0Æ05. Fig. 6. Percentage of fields (smoothed by GAM) where presence was recorded in 1970 compared to that in 2005 for 66 taxa where trends were significant. The broken line represents the case where percentage in 1970 equals that in 2005 (slope of 1). The geometric mean observed relative change was 5Æ9 (95% confidence interval 3Æ5 10Æ1) significantly greater than 1 (t 66 =6Æ61, P <0Æ001). Fig. 7. Percentage of fields treated with herbicides where presence was recorded compared with percentage presence in untreated fields (logarithmic scale) for 43 arable weed taxa. The broken line represents the case where the occurence for sprayed fields equals that for unsprayed fields (1 : 1 ratio). The geometric mean ratio across taxa of the percentage presence in herbicide-treated fields to that in untreated fields was 0Æ48 (95% confidence interval 0Æ38 0Æ61), significantly different from 1 (t 42 = )6Æ23, P <0Æ001). probability of recording a dicot in a field by 52% (geometric mean across dicot taxa of the ratio of percentage taxon occurrence in herbicide-treated fields to percentage taxon occurrence in untreated fields was 0Æ48 with 95% confidence interval 0Æ38 0Æ61, significantly different from 1: t 42 = )6Æ23, P <0Æ001). This herbicide effect was similar across 21 taxa that were uncommon (for which the percentage of unsprayed fields where each taxon occurred was below the median of 4Æ5%) and 22 common ones (the rest) with the average relative change being 0Æ6 (95% confidence interval 0Æ4 0Æ8) and 0Æ4 (95% confidence interval 0Æ3 0Æ6) respectively (t 41 =1Æ11, P =0Æ273). The percentage occurrence of dicot taxa in cereal fields not treated with herbicides from 1969 to 1986 was compared to that in untreated fields from 1987 to 2005 (Fig. 8). For most taxa the percentage of fields where they occurred was higher in the second period. The geometric mean relative change was 2Æ3 (95% confidence interval 1Æ5 3Æ5), significantly different from 1 (t 59 =3Æ96, P <0Æ001). For 16 taxa that were uncommon in untreated fields in (the percentage of untreated fields where they occurred in was less than the median of 1Æ1%) all of them became more common, whereas for the remaining 44 common ones, 26 (59%) increased in occurrence. The average relative change for the two groups was 16Æ3 (95% confidence interval 13Æ0 20Æ6) and 1Æ12 (95% confidence interval 0Æ8 1Æ7) respectively, significantly different from one another (t 58 =8Æ25, P <0Æ001). This was significantly greater than 1 in the case of uncommon taxa (t 16 =25Æ88, P <0Æ001) but not in the case of common taxa (t 43 = )0Æ61, P =0Æ546). STUBBLE SURVEYS: CHANGES OVER TIME We identified 92 species of dicots in stubbles. We compared the percentage of stubble fields where taxa were found in to the percentage in stubbles in The geometric mean relative change was 1Æ4 (95% confidence interval 1Æ0 1Æ9), not significantly different from 1 (t 92 =1Æ76, P =0Æ082, Fig. 9) indicating there had been no change over the 35 years. For 49 taxa that were uncommon in (the percentage of fields where they occurred was less than the median of 1Æ18%) 35 (71%) became more common, whereas for the 44 remaining common ones, 19 (43%), increased in occurrence. The average relative change for the two groups was 2Æ9 (95% confidence interval 1Æ9 4Æ5) and 0Æ6 (95% confidence interval 0Æ4 0Æ9) respectively, significantly different from one another (t 91 =5Æ38, P <0Æ001). The average relative change was greater than 1 for the uncommon taxa (t 48 =5Æ12,

9 Long-term changes in arable flora 223 Fig. 8. Percentage of fields from 1969 to 1986 not treated with herbicides where presence was recorded compared with percentage presence in fields not treated with herbicides from 1987 to 2005 for 60 arable weed taxa. The broken line represents the case where the percentage for equals that for The geometric mean ratio across taxa of percentage presence in relative to was 2Æ3 (95% confidence interval 1Æ5 3Æ5), significantly higher than 1 (t 59 =3Æ96, P <0Æ001). Fig. 9. The percentage of stubble fields in 1968 and 1971 where presence was recorded compared to that in 2004 and 2005 (logarithmic plot) for 92 arable taxa. The broken line represents the case where the percentage equals that in (1 : 1 ratio); the geometric mean ratio across taxa of the two percentages was 1Æ4 (95% confidence interval 1Æ0 1Æ9), not significantly different from 1; no change (t 92 =1Æ76, P =0Æ082). P <0Æ001) but significantly less than 1 for the common taxa (t 43 = )2Æ54, P =0Æ015). LONG-TERM CHANGES IN NUMBER OF SPECIES PRESENT Eleven species were recorded before 1968 (Wolley-Dod 1937; Hall 1980) that were not recorded by us, and 35 additional species were recorded in non-systematic surveys during Fifteen species and one subspecies (7% of a total of 217) have been lost from the flora (Table 4a). Eleven were lost before the study (seven before widespread herbicide use and four following it) and five during the course of the study. There were 15 gains, six before widespread herbicide use and none during (Table 4b). Nine species were gained during the course of our study, i.e. from 1968 (Table 4b). Nine archaeophytes (13% of 71 total) were lost compared to only four (3%) of the 119 native species and three (11%) of the 27 neophytes (Table 4a). Two archaeophytes were lost after 1970 and three natives. Overall the archaeophytes were lost at a higher rate than were native or neophyte species (V 2 2 =6Æ69, P=0Æ035), but this effect vanished after 1970 (V 2 2 =0Æ62, P =0Æ733). Nine of the fifteen species gained since c were neophytes (33% of the 27 neophyte species identified in the Sussex study area), with three (11% of neophytes) gained since Three archaeophytes (4%) and three native species (3%) were gained, all since Overall, relatively more neophytes were gained than archaeophytes or native species (V 2 2 =33Æ65, P <0Æ001), although since 1970, the rate of gain has been similar between all groups (V 2 2 =4Æ09, P =0Æ130). Discussion INITIAL IMPACT OF HERBICIDES A marked reduction in arable weed abundance followed the initial use of herbicides in cereal crops; our study suggests a 64% decline in dicot weed abundance and a 52% fall in species frequency. The phyto-sociological community of our untreated crops, the P. rhoeas S. noctiflora association was first described from studies during , before the use of herbicides (Wasscher 1941). Silene noctiflorum was the only species not identified in our June samples but it was found in the non-systematic surveys. Our ability to identify this community (OV16) indicates that the soil seed bank can still regenerate the distinctive components of the pre-herbicide community. It is likely that this position will be similar in principle for the other arable plant communities given that herbicide usage in the study area was similar to that in England as a whole (Ewald & Aebischer 2000). TRENDS DURING OUR MONITORING Discounting the losses due to herbicide use, above, the overwhelming impression throughout the continuous monitoring of our study was of persisting stability of weed abundance with notable increases in species diversity. During our study there wasnosignificantoveralltrendinweedabundance,although monocot abundance was slightly greater in the early 1980s than at other times. We were able to identify the short-term effects of herbicides on particular taxa, with changes in herbicide efficacy driving the trends in all seven annual dicot taxa where trends were humped and where efficacy data were available. There was a sustained increase in the occurrence of perennial dicot species, tall species of considerable ecological importance (Heracleum, Artemisia, Arctium, etc.). Our analysis indicates that changes in perennial dicot occurrence are most probably due to the loss of traditional leys that interrupted perennial life cycles, with most grass leys down for 3 years of grazing and cutting, and not to changes in herbicide specificity. Set-aside, introduced in the study area in 1993, did not affect the pattern of increase (Fig. 4b). In addition, whether in crops

10 224 G. R. Potts, J. A. Ewald & N. J. Aebischer Table 4. Species lost and gained from cereal crops in the Sussex Downs Study area, in relation to first use of herbicides in 1946 and the beginning of this study s monitoring. Approximate dates of losses and gains are given. Each species is classified as Archaeophyte (*), neophyte (**) or native (no symbol) Time period Species and sub species (year) Total (a) Species lost Pre-1946 Adonis annua* (1917), Agrostemma githago* (1930), Alyssum alyssoides** (1930), 7 Anagallis a. foemina* (1930), Caucalis platycarpus* (1887), Ajuga chamaepitys (1666), Melampyrum arvense** (1892) Ranunculus arvensis* (1952), Reseda phyteuma** (1965), Galium tricornutum* (1967), 4 Anthemis arvensis* (1967) Spergula arvensis* (1986), Torilis nodosa (1990), Torilis arvensis* (1974), 5 This study Clinopodium acinos (1971), Anthemis cotula* (1972) Total 16 (b) Species gained Pre-1946 Veronica persica** (1829), Lepidium draba** (1887), Senecio squalidus** (1898), 6 Matricaria discoidea** (1901), Galinsoga parviflora** (1920), Rapistrum rugosum** (1936) None June samples This study Non-systematic This study Chenopodium ficifolium* (1988), Nigella damascene** (2005), Artemisia vulgaris* (1976); 7 Epilobium ciliatum** (1978), Heracleum sphondylium (1983); Arctium minus sens. lat. (1992) Eupatorium cannabinum (2005) Lactuca serriola* (2004), Linaria purpurea** (2004) 2 Total 15 1 Not seen since or in stubbles, the taxa that were uncommon at the start of our monitoring have increased, whereas those that were common have shown consistency or have become less common, contrary to expectation. COMPARISON WITH OTHER STUDIES Overall the composition of the archaeophyte flora in the Sussex study area appeared to have stabilized relative to the native flora by 1970, whereas the neophyte flora appears to be still expanding. This supports the conclusions of Braithwaite et al. (2006). Many of the changes that we identified in the occurrence of individual taxa in our monitoring match those found where surveys were undertaken intermittently during the same period (Wilson et al. 1999; Sutcliffe & Kay 2000; Hyvo nen et al. 2003a,b; Weber & Gut 2005; Andreason & Stryhn 2008). The taxa that increased in Sussex, and in a majority of the other studies, included Fumaria spp., Euphorbia helioscopa, Geranium spp., Sherardia arvensis, Cirsium and Artemisia, while those taxa that showed little change in a majority of the other studies and in Sussex included Reseda lutea, Potentilla anserina and Epilobium spp. Taking into consideration the difficulties in comparing weak or variable trends, our results for Sisymbrium officinale, Lithospermum arvense, Myosotis arvensis, Veronica arvensis, Odontites vernus, Alopecurus myosuroides and Bromus hordeaceus are similar to those found in other surveys. An increase in perennials in Denmark (Andreason & Stryhn 2008), similar to Sussex, was attributed to minimum tillage but this was not practised in our study area during the time period reported here. We compared our results for the 46 single-species taxa for which we had trends (increase, decrease or no change) to those from national surveys over similar time frames (Braithwaite et al. 2006; who measured change from to in 811 tetrads and Preston et al. 2002; who also examined trends recorded at the tetrad level in the 1962 Atlas of British Flora (Perring & Walters 1962), the BSBI Monitoring Scheme (Rich & Woodruff 1990) and the New Atlas of British and Irish Flora (Preston et al. 2002)). There was agreement for 21 (46%) of the 46 species with the Braithwaite et al. (2006) data covering to The major discrepancy was with 12 species that had decreased in Sussex from 1987 but had not done so nationally, most notably Stellaria media, Myosotis arvensis and Elytrigia repens. This may reflect the greater use of herbicides in cereals than in other surveyed habitats. For the longer period of time covered by Preston et al. (2002), there was agreement with our results from Sussex for only 12 (26%) of the 46 species with again, S. media, M. arvensis and E. repens declining in our survey but not nationally, together with Veronica arvensis and Poa annua.forthelongertimeperiod(preston et al. 2002) there were nine species that increased in Sussex but declined in distribution nationally and 17 species that increased in Sussex whose distribution nationally was considered to be stable. Of the 17, two species, Arctium minus sen. lat. and Senecio vulgaris, also increased in Sussex but showed no change in distribution nationally over the shorter time span in Braithwaite et al. (2006). Only four species showed similar results in our study and in the two national surveys: Centaurea cyanus increased while Ranunculus repens, Sinapis arvensis and Reseda lutea all showed no change in distribution. EFFECTS ON FARMLAND BIODIVERSITY At the start of the monitoring, it was anticipated that there would be a gradual decline in the floral diversity and abun-

11 Long-term changes in arable flora 225 dance to a point of entirely weed-free crops but this has not yet happened. Part of the explanation seems to be the buffering effect of seed longevity in the soil seed bank. With a mean half-life amongst 21 species of annual arable dicotyledons estimated at 26 years (Wilson & Aebischer 1995) we could expect a maximum loss of 60% through our study; about the same as the loss of seeds from soils in arable landscapes between 1945 and (Robinson et al. 2004). However, with up to 7000 seeds per m )2 for P. rhoeas and 2845 m )2 for Sinapsis arvensis (Thompson, Bakker & Bekker 1997), with a pre-herbicide abundance of all dicotyledons species in the soil seed bank approaching m )2 (Brenchley & Warington 1930) and with annual replenishment (albeit at a lower rate than pre-herbicides), the cereal flora has considerable resilience to sustained herbicide treatment. It is a feature of arable weeds that most of them have long-lived seeds, but amongst a small group of arable weeds with relatively shortlived seeds are four species lost from our study area since herbicides were first used: Ranunculus arvensis, Galium tricornutum, Spergula arvensis and Torilis arvensis (Thompson et al. 1997). A trend towards a flora with longer seed longevity might therefore be expected but, apart from the above losses, this was not evident from our data. Because there are short-term changes in cereal weed flora it is difficult to estimate the status of species from snapshot studies. It is equally difficult to determine the dynamics of the soil seed bank from above-ground studies. Our results suggest that a fuller understanding of the status of arable weeds requires direct monitoring of the soil seed bank and separate consideration of the non-seed-dependent perennials that have increased through our study. INVERTEBRATES AND BIRDS Of the 11 weed taxa of most importance to invertebrates (Marshall et al. 2003), the trend in the occurrence of five (Chenopodium album, Cirsium arvense, Rumex obtusifolius, Senecio vulgaris and Sonchus oleraceus) has been upward throughout our study, reaching a maximum in Galium aparine increased until 1995, with a slight decrease since then. Of the remaining five taxa, three (Stellaria media, Polygonum aviculare and Poa annua) reached a maximum in the 1970s with a subsequent decrease, while Matricaria spp. and Tripleurospermum inodorum reached a peak in 1984, and Sinapis arvensis showed no change. A similarly mixed picture emerged from consideration of those weed taxa most important for seed-eating birds (Marshall et al. 2003): three showed a maximum occurrence in the 1970s, strongly declining since then (Stellaria media, Persicaria maculosa and P. lapathifolia); Polygonum aviculare and Fallopia convolvulus showed little or no change; while Chenopodium album increased throughout the study. The declines of seed-eating birds that began around 1977 (Fewster et al. 2000) do not parallel changes in the flora that we studied, perhaps strengthening the view that the loss of stubbles was responsible (Gregory, Noble & Custance 2004). In contrast, the grey partridge decline began during , before national monitoring of UK bird populations, earlier than the species mentioned above but in line with the introduction of herbicides and its initial effects (Potts 1986). Whether the population densities of any other bird species were similarly affected has yet to be demonstrated. PRACTICAL CONSIDERATIONS This study emphasizes the opportunities that exist to restore the cereal ecosystem flora in at least some landscapes by restricting the use of herbicides in the headlands of cereal crops. Based on the diverse flora that we have identified and the higher occurrences associated with untreated fields in the latterhalfofthestudyperiod,wepredict that limiting the use of herbicides on headlands and stubbles would result in a rapid increase in floral diversity as has occurred elsewhere (Walker et al. 2007). It follows that an assessment of the potential to regenerate the weed flora should be made before resorting to floral enhancement with additional seeding as available in the agri-environment measures. As well as being unnecessary, such floral enhancement may include species not hitherto present, confounding studies such as ours. Meanwhile the take-up of in-field options that would restore traditional cereal ecosystem floras has been very poor, accounting for only 0Æ35% of Entry Level Stewardship points agreed with farmers (Hodge & Reader 2009). Improved incentives are urgently needed. Acknowledgements We are very grateful for the help of Frances Abraham and Frank Penfold (who sadly died whilst this paper was in preparation). We thank Jonquil Ash, Kevin Walker and Alan Knapp for help with difficult species, helpful referees, the farmers in the study area and many colleagues who have helped through the course of the long study. The senior author received an Emeritus Fellowship from The Leverhulme Trust to support the preparation of this paper. References Aebischer, N.J. (1991) Twenty years of monitoring invertebrates and weeds in cereal fields in Sussex. The Ecology of Temperate Cereal Fields (eds L.G. Firbank, N. Carter, J.F. Darbyshire & G.R. Potts), pp Blackwell Scientific Publications, Oxford, UK. Akaike, H. (1974) A new look at the statistical model identification. The Institute of Electrical and Electronics Engineers Transactions on Automatic Control, 19, Andreasen, C. & Stryhn, H. (2008) Increasing weed flora in Danish arable fields and its importance for biodiversity. Weed Research, 48, 1 9. Bachthaler, G. (1969) Entwicklung der Unkrautflora in Deutschland in Abha ngigkeit von den veränderten Kulturmethoden. Angewandte Botanik, 43, Braithwaite, M.E., Ellis, R.W. & Preston, C.D. (2006) Change in the British Flora A report on the BSBI Local Change survey. Botanical Society of the British Isles, London. Brenchley, W.E. & Warington, K. (1930) The weed seed populations of arable soil. I. Numerical estimation of viable seeds and observations on natural dormancy. Journal of Ecology, 18, Briggs, M., Harmes, P. & Knapp, A. (2001) The Sussex Rare Plant Register of Scarce & Threatened Vascular Plants, Charophytes, Bryophytes and Lichens. Sussex Wildlife Trust, Henfield. Ewald, J.A. & Aebischer, N.J. (2000) Trends in pesticide use and efficacy during 26 years of changing agriculture in Southern England. Environmental Monitoring and Assessment, 64, Fewster, R.M., Buckland, S.T., Siriwardena, G.M., Baillie, S.R. & Wilson, J.D. (2000) Analysis of population trends for farmland birds using generalized additive models. Ecology, 81,