Unexpectedly high 20th century floristic losses in a rural landscape in northern France

H. Van Calster, R. Vandenberghe, M. Ruysen, K. Verheyen, M. Hermy, G. Decocq

First published: 12 August 2008

Summary

• 1Resurveys of regional floras allow assessment of long-term floristic change and to pinpoint driving forces behind these changes. Causes of floristic decline derived from such resurveys depend on the considered landscape, but are often associated with human activities especially in urbanized areas. Much could be learned from comparisons of contrasting landscapes.
• 2Vascular plant abundance categories from a detailed, late 19th century flora were matched with distribution data from a late 20th century plant database in a rural area (Thiérache; 1673 km²) in northern France. Floristic change was characterized both in absolute numbers of extinct and extant species and by a measure for relative change in range size. Knowledge of land use changes and ecological correlates of floristic change permitted the causes for species change to be inferred and to identify which traits are associated with species vulnerability. Extinction rates were compared with similar studies from contrasting landscapes, taking into account the negative relationship between standardized extinction rates and log-transformed size of the study area.
• 3Of the 959 species from the historical list, 186 (19.4%) may be considered regionally extinct. Most extinct species were already rare historically. Differences among habitats indicated strong declines for aquatic species and arable weeds and least change for forest species. Percentage species loss per year in Thiérache equalled 0.22, which was higher than two other similar sized predominantly rural landscapes.
• 4Across the main habitat groups, relative species decline was always associated with relatively higher stress tolerance and lower competitiveness and biased towards therophytes. The main causes of species decline were management intensification, eutrophication, secondary succession in semi-natural open habitats and land use change.
• 5Synthesis. Rural landscapes with no population density increase, no urbanization and a stable landscape configuration are believed to be less vulnerable; yet, on a relatively short time scale, without nature conservation measures, these landscapes may suffer strong species loss comparable to that in urbanized landscapes. Effective management of landscapes for biodiversity conservation whether they are urbanized or rural requires the development of a network of nature reserves.

Introduction

Human activities have an impact on the environment through changes in land use, addition or removal of species or populations, and alterations to biogeochemical cycles (Vitousek et al. 1997). Accelerated increases in human population size and resource demand during the past 200–300 years are held responsible for the present biodiversity crisis (May et al. 1995; Vitousek et al. 1997). During this period, landscape-scale loss of native plant species has been documented by a limited number of studies in temperate regions and has occurred at differing rates depending on the mechanisms threatening species persistence (Waller & Rooney 2004).

Habitat loss results in loss of populations (decline) or even species (extinction) belonging to that habitat. In particular, loss of habitat with a long temporal continuity is detrimental because dispersal of species to new habitat patches is often too slow in relation to current rates of land use change (Verheyen et al. 2004; Matlack 2005). The ongoing habitat loss and fragmentation of ancient forests (e.g. Peterken & Game 1984) and ancient grasslands (e.g. Ihse & Lindahl 2000) therefore forms a major threat for many plants. Another important cause of floristic change is related to alterations of biogeochemical cycles which often increase the availability of otherwise limiting plant nutrients. Increased nitrogen deposition especially may cause eutrophication and acidification in semi-natural habitats (e.g. Bobbink et al. 1998) and can result in random loss of rare species or directed loss of species which are functionally susceptible irrespective of initial abundance (Suding et al. 2005). In farmed landscapes, type and intensity of agricultural activities impinge upon the floristic composition of arable weed species and of semi-natural landscape features embedded in the agricultural matrix (e.g. Andreasen et al. 1996). Agricultural intensification reduces the diversity of weed species by a simplification of cropping systems, the application of pesticides and fertilizers, and drainage or irrigation (Stoate et al. 2001). Field sizes are often expanded which removes hedgerow habitats (Baudry et al. 2000). In other regions, however, abandonment of agriculturally degraded lands initiated secondary succession towards forest (e.g. Harrelson & Cantino 2006). In general, abandonment or intensification of traditional management, be it in agriculture (e.g. Ihse & Lindahl 2000) or in forestry (e.g. Decocq et al. 2005), deviates from the historical disturbance regime, causing possible floristic loss of species adapted to that regime (White & Jentsch 2001).

Historical flora dating from the 18th to early 20th century have been used to assess long-term species changes in range size and extinction rates (e.g. McCollin et al. 2000; Van der Veken et al. 2004; Walker & Preston 2006; Stehlik et al. 2007), especially because they reflect reference conditions before or at the onset of the industrial revolution. Although few in number, these studies offer the opportunity to investigate mechanisms of species loss operating within landscapes that are only detectable by a cross-comparison of floristic loss between landscapes. Long-term floristic changes have mostly been studied in regions with strong urbanization (e.g. Drayton & Primack 1996, Duncan & Young 2000; Chocholousková & Pysek 2003; Van der Veken et al. 2004; Stehlik et al. 2007). For these studies, the most immediate threat came from habitat loss. In rural areas biodiversity loss was associated with agriculture and forest-related land use and management changes (Harrelson & Cantino 2006; Walker & Preston 2006). In any landscape, the amount of semi-natural habitat left and the strength of conservation actions may delay plant species responses to background environmental degradation (Bengtsson et al. 2003; Gaston 2005). In this respect, residual semi-natural habitat provided refuges for unimproved grassland species across a range of land use intensity, but within the limits set by the size of the residual habitat, the land use intensity and local conditions (Smart et al. 2006). Native species loss may also depend on resilience to species invasions. Urbanized and new world regions are more prone to invasion by exotic species (e.g. Drayton & Primack 1996; Chocholousková & Pysek 2003; Harrelson & Cantino 2006), although at large spatial scales a negative influence of invasive exotic species on native species is generally lacking (cf. Melbourne et al. 2007).

In the present study, a 19th century floristic list with species abundances for a rural area in Northern France (Thiérache) was compared with plant species distribution data from the second half of the 20th century. The region, which experienced no urbanization, is characterized by a mixture of grasslands and arable fields, together with stable remnants of ancient forests. Despite a fairly stable landscape matrix, the extinction rate was high.

As the Thiérache region encountered neither landscape changes (i.e. the area covered by forests, grasslands and arable lands were the same) nor population density increase since the late 19th century, we propose the three following hypotheses:

• 1Changes in the regional flora are primarily due to qualitative alterations of habitats, such as eutrophication, management intensification or discontinuation, and thus the extinction rate is low compared to landscapes that undergo urbanization and associated semi-natural habitat destruction.
• 2Species loss and population decline are primarily biased towards historically rare species.
• 3Distributional changes of species can be predicted from life-history traits, irrespective of their habitat type and historical rarity.

Methods

study area

The Thiérache region (1673 km²) is located in the NE of the Aisne department in northern France (centred at: 49°56′ N, 3°54′ E). Altitude increases from SW to NE (50–285 m), which causes variation in annual rainfall (627 mm in SW to 1010 mm NE) and average annual temperature (10.6 °C SW, but 9.0 °C NE). The climate is humid Atlantic. The entire area lies within the watershed of the Oise river, flanked by alluvial plains in the downstream section (W–SW) of Thiérache. The geological substrate consists of Cretaceous calcareous chalks, marls and clays, except in the NE corner where Paleozoic schists and sandstones and Jurassic limestones outcrop; these bedrocks are largely covered by Quarternary loess.

landscape dynamics

Currently, the study area has only 49 inhabitants per square kilometre (the largest city, Hirson, has only 10 340 inhabitants). Population declined from 117 573 in 1820 (Brayer 1824) to 94 435 in 1962, to 76 492 in 2002 (Seguin 2007). Forests, fields and grasslands occupy 86% of the land, yet forestry and agriculture currently employ only 2.8% of the people. A century ago most of the population worked in those sectors.

The Thiérache landscape by the late 18th century was characterized by open field agriculture for grain cultivation. Due to the combined effect of grain price collapse (–45% between 1860 and 1895) and climatic crisis (Little Ice Age 1550–1850), open fields were converted to grasslands with hedgerow networks (so-called bocage; Catrin 1870). Early land use inventories in nine administrative subdivisions (i.e. ‘cantons’) indeed showed a shift from arable land to grassland between 1820 (Brayer 1825) and 1999 (AGRESTE 2000) (Fig. 1). The former decreased from 67.3% to 42.9% (–24%) and the latter increased from 8.2% to 29.6% (+21%). Forest area (1999: 13.5%) changed little (–3%), resulting in a net change of +6% for other land uses (1999: 13.9%, mainly build-up areas and gardens; the road network had not become denser). The main period of arable field to grassland conversion occurred before the second half of the 19th century (Catrin 1870). However, in the last couple of decades, a conversion from grasslands to open fields is moving from the S to the N due to the crisis encountered by dairy farmers. As a result, grassland-dominated and arable field-dominated landscapes are located in the administrative subdivisions of the N and NE and of the S and SW, respectively (Fig. 1).

The fairly dynamic nature of open habitats contrasts with the stability of forested areas in the 19th and 20th century (Fig. 2), which were mostly of ancient origin (i.e. forest continuity since at least the 11th century). Forest land use history since the 11th century was characterized by a gradual deforestation. Small-scale afforestation was only practiced in the 20th century. The forested area decreased from at least 27% in the 11th century to 13% in the 20th century (Fig. 2). Most deforestation occurred during the 18th century (–11%), whereas the 11th–17th and 19th century deforestation periods only accounted for –2% and –1%, respectively.

floristic survey data

Nineteenth century floristic data were available from Riomet (1891) and Riomet & Bournérias (1952–1957). The former flora contained all floristic data collected by Riomet in Thiérache between 1880 and 1891. The second, which dealt with the entire Aisne department, contained all unpublished data from Riomet. We extracted only those data collected between 1891 and 1900 in Thiérache from the latter flora. For each species, both floras give the degree of abundance and habitat types. Riomet (1891) ranked abundance according to six categories (see Table 1), from very rare (RR) to very common (CC). The other flora extended this scale with two extreme classes, RRR and CCC, at both ends of the range. These extreme abundance classes were integrated into the RR and CC categories, respectively. Data for 808 species came from Riomet (1891) and 151 species were added from Riomet & Bournérias (1952–1957).Table 1. Conversion of recent range sizes to historical range sizes, and vice versa

Ranking	Historical class	Rarity index thresholds*	Number of grid cells			Proportion†
			Minimum	Maximum	Mean	Minimum	Maximum	Mean
1	RR	100–97.5	1	3	2.0	0.01	0.03	0.02
2	R	97.5–93	4	9	6.5	0.04	0.07	0.05
3	AR	93–84.5	10	19	14.5	0.08	0.15	0.12
4	AC	84.5–68.5	20	40	30.0	0.16	0.31	0.24
5	C	68.5–36.5	41	82	61.5	0.32	0.64	0.48
6	CC	36.5–0	83	129	106.0	0.64	1.00	0.82

• * Adapted from Toussaint et al. (2005). RR, very rare; R, rare; AR, fairly rare; AC, fairly common; C, common; CC, very common.
• † Proportion was calculated as (x + 0.5)/(n + 1), with x the number of occupied 4 km squares and n = 129.

Recent floristic data were obtained from the DIGITALE database (Centre Botanique National de Bailleul, url: http://www.cbnbl.org/centrederdigit.htm), containing plant distribution data for Northern France, in 4 by 4 km grid squares. Data came from systematic surveys, botanical literature sources, herbaria and individual botanists. Data were retrieved for each 4 km square needed to cover Thiérache (n = 129; squares near the perimeter, which contained territory inside and outside Thiérache, were included). For each record, the year when a species was last seen was available (range 1839–2005). The threshold year used to include records was set at 1958 to ensure that each 4 km square was sufficiently recorded (see Appendix S1 in Supplementary Material). A total of 24 226 records (i.e. species – 4 km square combinations) were retained. The drawback of using a long end period is that species might have arrived/gone extinct after 1958, giving a false impression of changes since 1900. Contrary to Stehlik et al. (2007), some supposedly extinct species might have been simply overlooked. But that bias was reduced by using a long period (48 years) of cumulative recordings. Moreover, all major forests in Thiérache were intensively surveyed by one of us (G. Decocq) from 1995 to 2000.

Only native (n = 952) and naturalized (n = 43) species for both floristic lists were considered; thus excluding casuals, sub-spontaneous and cultivated species (n = 46; sensu Toussaint et al. 2005). Species nomenclature for the historical flora was updated using the synonyms index for the French flora (Bock et al. 2007) and nomenclature of final species lists followed Lambinon et al. (1998).

species trait data

• 1Species habitat preferences were obtained from Riomet (1891) and Riomet & Bournérias (1952–1957). We allocated their habitat descriptions to 14 secondary habitat types belonging to six primary habitat types. A branched key (adapted from Grime et al. 2007: p. 21) was used to allocate habitat descriptions to primary and secondary habitat types (See Appendix S2). Species could be affiliated with more than one habitat.
• 2Ellenberg’s indicator values (Ellenberg et al. 1992) for light, temperature, continentality, soil moisture, soil pH, and soil nitrogen were used as environmental species trait indicators. These ordinal values are on a scale of 1–9 (soil moisture: 1–12) with higher values indicating higher resource requirements.
• 3Two traits represented the regenerative phase. Dispersal potential was coded on a three-point scale: no adaptation for long-distance dispersal, adaptations for one vector of long-distance dispersal and adaptations for at least two vectors for long-distance dispersal (defined as dispersal by water, wind, birds and mammals) (Tamis et al. 2004). Reproduction mode (Klotz et al. 2002) was coded on a five-point ordinal scale ranging from strictly sexual to strictly vegetative reproduction.
• 4Four traits represented the established phase. The standard coordinates (values between 0 and 1) for competitiveness, stress tolerance and ruderal character of a species’ CSR-strategy were taken from Hunt et al. (2004) and Grime et al. (2007). Species life forms, that is, according to Raunkiaer’s (1934) classification, were retrieved from the BiolFlor database (Klotz et al. 2002).

data analysis

To determine change in the historical and recent area occupied by a species, historical abundance had to be compared with recent abundance (i.e. number of occupied 4 km squares). Therefore, a rarity index, for which plausible ranges corresponding to an RR–CC scale were known (cf. Toussaint et al. 2005), was calculated as the percentage of unoccupied 4 km squares (Table 1). This permitted conversion of numbers of occupied 4 km squares to an RR–CC scale, or, to convert RR–CC classes to mean numbers of occupied 4 km squares (Table 1). The former conversion was used as a coarse way of inferring which species declined, increased or remained stable (i.e. negative, positive or no shift along the RR–CC six-point scale, recoded as 1–6). The numbers of declined, increased, stable and extinct species were calculated for all species, for each primary habitat group and within each historical abundance class (for a list of species presumably extinct in Thiérache see Table S1).

For a more detailed way of inferring species change, relative change in range size (i.e. area of occupancy) was calculated following McCollin et al. (2000) and Telfer et al. (2002). They used linear regression to predict recent from historical abundance and used the standardized residuals as a measure of relative change in range size. Whereas McCollin et al. (2000) directly regressed numbers of occupied grid cells against ordinal ranked historical abundance classes, Telfer’s method was for situations in which species abundance data from both surveys were counts of occupied cells in the same region. McCollin’s method fails to reflect the intuitive notion that, going from RR to CC, the increase in area of occupancy reflected by a change from RR to R is much smaller than a change from C to CC. The conversion in Table 1 reflects this and was used to convert historical abundance to mean numbers of occupied 4 km squares, after which we proceeded with Telfer’s method. First, recent and historical numbers of occupied 4 km squares, x, were converted to proportions, P, as (x + 0.5)/(n + 1), with n the total number of grid cells. Next, logit-transformed proportions, ln [P/(1 – P)], were used in a weighted linear regression procedure to obtain standardized residuals. This transformation is suited to normalize range sizes, attain an unbounded distribution and make the historical-recent range size relationship approximately linear (Telfer et al. 2002). Weighting reduced increased variation near 0 and 1 proportions.

Standardized residuals from a regression using all species were related to secondary habitat type using one-way anova. Standardized residuals obtained by Telfer’s method for arable land species, grassland species and woodland species separately, were related to environmental species indicator and life-history traits (regenerative and established phase). Arable land, grassland and woodland were chosen because they comprised ±90% of the land. Spearman’s rank correlations were used for traits on an ordinal scale with more than three levels; for the other traits one-way anova was used. If a factor was significant overall (P < 0.05), post hoc homogeneous subsets were determined by Tukey’s honestly significant difference. Because the standardized residuals had a hump-shaped relationship with soil pH and moisture, these species indicators were recoded to make the relationship linear for correlation analysis. Both variables were reduced to a five-point scale going from neutral to more extreme conditions (acid or basic, wet or dry) (soil pH: 5: 1, 4 and 6: 2, 3 and 7: 3, 2 and 8: 4, 1 and 9: 5; soil moisture: 6: 1, 5 and 7: 2, 4 and 8: 3, 3 and 9: 4, 2 and 10: 5).

Results

extinct and extant species in thiérache

Of 959 species recorded by Riomet, 773 were extant whereas 186 (19%) were not found again (Table 2, Table S1). The number of newly recorded species represented only 4.8% (n = 48) of the total species list. Twenty of them had a naturalized status. The remaining native species might have spontaneously colonized the region, since Riomet quoted 16 species as occurring to the south of Thiérache, or were overlooked. Extinction percentages differed between habitats, ranging between 13% (woodland species) and 24% (species on siliceous or calcareous substrates). Considering all species, 41% decreased, 25% remained stable and 14% increased. Across primary habitats, the percentage of decreasing species ranged from 31% (woodland species) to 53% (arable species). Woodland species were the only ones with more stable/increased species (56%) than extinct/decreased species.Table 2. Numbers of extinct, extant and newly recorded species, extinction percentages and total species numbers for each primary habitat type and for all species combined

Species group	Extinct	Extant			New	Total	Percentage extinct (excluding new)
		Decreased	Stable	Increased
Other habitats†	59	103	53	27	4	246	24.4
Arable land	55	137	50	15	8	265	21.4
Wetland	39	92	46	15	17	209	20.3
Grassland	33	110	72	20	13	248	14.0
Wasteland	39	132	95	32	11	309	13.1
Woodland	50	124	127	96	16	413	12.6
All	186	399	244	130	48	1007	19.4

• † Includes plants from skeletal, spoil, calcareous and siliceous/sandy substrates.

Species that were rare by the end of the 19th century were most prone to extinction (Fig. 3; a χ²-test for the association between historical abundance and the four classes of changed abundance was significant: χ² = 336, d.f. = 15, P < 0.001). A small number of historically very common species also went extinct: Agrostemma githago, Chrysanthemum segetum, Potentilla anglica and Trifolium filiforme (Table S1).

differences in relative change among habitat types

The intercept (–1.315) and slope (0.897) obtained by weighted linear regression using all species (Fig. S1) was a way to remove the effects of a change in recording probability (intercept) and the extent to which recording probability depended on the historical recording probability (slope). These parameters mainly reflected differences in methods between surveys rather than biological change (cf. Telfer et al. 2002).

Significant differences among secondary habitat types (Fig. 4) (n = 1900, d.f. = 13, F = 14.472, P < 0.001) indicated most decline for those belonging to wetland and arable primary habitats compared with those belonging to woodland. Others took intermediate positions, but among grassland secondary habitats, species of very dry grasslands declined most.

relative change in arable land, grassland and woodland species

Relative changes in range size were always positively correlated with soil nitrogen, and negatively with the species trait indicators temperature and soil moisture (the latter on a neutral to increasingly wet/dry scale; Table 3). Grassland and woodland species had significant negative correlations with light indicator values, and woodland species were negatively correlated with soil pH indicators (on a neutral to increasingly acid/basic scale). Relative change in range size of arable- and grassland species was positively correlated with an increasing reliance on vegetative reproduction. For the established phase (Table 3), arable land species correlated positively with competitiveness and negatively with ruderal character and stress tolerance. For grassland and woodland species, only the trade-off along the stress axis remained, that is, species adapted to low stress/low disturbance (competitors) gained importance at the cost of species adapted to high stress/low disturbance. Life forms differed significantly, with therophytes most negatively affected, followed by hemicryptophytes, geophytes, chamaephytes and phanerophytes (the final two only represented in woodland).Table 3. Relationships between environmental indicators or life-history traits with relative change in range size for arable land, grassland and woodland species separately

	Arable land species			Grassland species			Woodland species
	Test	Significance	N	Test	Significance	N	Test	Significance	N
Environmental indicators
Light	0.060	NS	230	–0.337	***	206	–0.287	***	360
Temperature	–0.240	**	200	–0.187	*	137	–0.318	***	264
Continentality	0.052	NS	200	0.038	NS	188	–0.092	NS	344
Soil moisture recoded†	–0.270	***	220	–0.351	***	193	–0.269	***	333
Soil pH recoded†	–0.001	NS	184	–0.035	NS	154	–0.148	*	299
Soil nitrogen	0.262	***	217	0.357	***	191	0.370	***	337
Regenerative phase
Reproduction mode	0.361	***	237	0.218	**	214	0.076	NS	367
Dispersal potential	F = 1.690	NS	211	F = 1.864	NS	194	F = 0.070	NS	340
Established phase
Competitiveness	0.452	***	148	0.306	***	161	0.242	***	282
Stress tolerance	–0.279	**	148	–0.260	**	161	–0.143	*	282
Ruderal	–0.208	*	148	–0.039	NS	161	–0.074	NS	282
Life form	F = 5.702	**	227	F = 9.554	***	205	F = 3.121	**	357
Therophyte‡	–0.219	a	68	–0.573	a	15	–0.544	a	11
Hemicryptophyte‡	0.045	ab	141	0.061	b	152	–0.009	b	200
Geophyte‡	0.193	b	18	0.140	b	38	0.034	b	68
Chamaephyte‡							0.122	b	16
Macrophanerophyte‡							0.200	b	33
Nanophanerophyte‡							0.202	b	29

• Test gives always Spearman’s rank correlations, except for dispersal potential and life form for which the F-value indicates the result of one-way anova.†5-point scale from neutral to more extreme soil conditions.‡rows indicate mean values of relative change in range size for each life form class and post hoc homogeneous subsets (a and b, Tukey’s honestly significant difference). NS, not significant; *P < 0.05, **P < 0.01, ***P < 0.001.

Discussion

land use and ecological change in open habitats

Arable field species had the highest extinction rate. Although arable fields were the only habitat to have encountered a significant net loss since 1820, several components of management intensification degraded the habitat and seem more likely to have caused this change. The use of agricultural machinery adapted to large scale processing, seed cleaning techniques and the use of pesticides and fertilizers have become increasingly important, especially after 1950 (Andreasen et al. 1996; Stoate et al. 2001). Because of herbicide selectivity, usually only species phylogenetically close to the crop species escape from herbicide effects. Furthermore, because the spectrum of cultivated species has greatly diminished, the associated weeds (e.g. Agrostemma githago, Bromus secalinus, Centaurea cyanus) have disappeared or been decimated (see also Decocq 2004; Hoste et al. 2006). Avena sativa, Fagopyrum esculentum, several varieties of Triticum spp., oleaginous plants and plants for textile use or colouration are no longer cultivated whereas cultivation of Zea mays has greatly expanded.

Eutrophication seemed an overall threat regardless of habitat type as indicated by the soil nitrogen species indicators. Excess of soil nutrients caused the decline of species adapted to environments where soil resources are limiting (stress-tolerators). The decline of therophytes might be indirectly related to competitive exclusion by plants with high rates of resource capture (competitives) which were stimulated by eutrophication. Studies of floristic change on landscape (e.g. Van der Veken et al. 2004; Walker & Preston 2006) and national scales (e.g. Smart et al. 2005) found similar responses to eutrophication. Eutrophication sources include direct fertilization of arable fields, forest plantations and improved grasslands, which causes N-leaching and run-off into wetland habitats (Preston et al. 2003). Atmospheric N-deposition loads in semi-natural open habitats (Bobbink et al. 1998) are probably low in Thiérache, but even low N-loads sustained for a long period of time have been shown to reduce plant diversity (Clark & Tilman 2008). For open habitats, eutrophication can result in stochastic loss of rare species, but functionally predictable declines are also possible depending on the habitat studied (Suding et al. 2005, for temperate grasslands see Fréville et al. 2007).

Survival of many species of open habitats implicates active management to prevent secondary succession towards forest. The relative decline of thermophilous and heliophilous species in grassland (primary habitat) is best understood in light of the strong decline of species thriving in very dry grasslands (secondary habitat, i.e. mostly calcareous grasslands). Although the area of very dry, calcareous grasslands in Thiérache is very small (a few hectares), they are species rich. The remaining calcareous grasslands are endangered by shrub colonization since they are no longer grazed by sheep (see Adriaens et al. 2007). Whereas secondary succession is one explanation for the decrease in light and heat indicator plants of calcareous grasslands, eutrophication by N-deposition is another. Eutrophication may cause a large increase in the biomass of Brachypodium pinnatum (a fairly common species in Thiérache), which can result in competitive exclusion of other plant species (Bobbink et al. 1998).

changes in woodlands

Like the open habitats, nitrogen indicator and competitive species increased strongly in woodland, indicative of eutrophication. Canopy interception of atmospheric N-deposition causes eutrophication of woodlands and has been frequently observed in the herb layer (e.g. Kirby & Thomas 2000; Diekmann & Falkengren-Grerup 2002), often in combination with soil acidification on poorly buffered soils which was reflected in Thiérache by the decline of species with high soil pH trait values. Acidophilous light-demanding species, however, also declined in Thiérache because conifers planted after World War II were on open acid mires or heaths.

The decline of light and heat indicator plants in woodlands deserves special attention. Riomet often mentioned the localized occurrence in forests of rare, often light demanding, plants on acid (e.g. Genista pilosa, Erica cinerea, Euphrasia micrantha, Pedicularis sylvatica, Vaccinium myrtillus, V. vitis-idaea) or basic substrates (e.g. Crepis praemorsa, Helleborus foetidus, Limodorum abortivum, Verbascum phlomoides), but these species declined or went extinct. Although they were explicitly mentioned as occurring in one or more of the forests, they have their typical habitat outside woodlands. Almost all forests in Riomet’s time were managed as coppice-with-standards with a rotation between 20 and 30 years (Brayer 1825; Catrin 1870). The coppice-with-standards system has been maintained until the end of the 1970s, after which 85% of the forests have been converted into high forest systems; only 12% are still managed as coppice-with-standards and 3% as coppice. Traditional forest uses, such as litter raking, forest grazing by live stock and collection of bark for tanning, which were still widespread in the 19th century, have now disappeared completely. Traditional crafts (e.g. charcoal burners, clog makers, glass makers) and early iron industries and glass factories that depended on forest resources also disappeared (Doyen 2003). Undoubtedly, these practices maintained forest ecosystems in a much more open state than nowadays by creating clearings in the forest interior. Furthermore, after World War II, the region has experienced a continuous increase in roe deer densities, reaching up to 40 individuals per 100 ha (Office national de la chasse, unpublished data), which might have caused relative abundance changes in woodland herbs (cf. Kirby & Thomas 2000; Corney et al. 2006). Thus, despite being stable entities in the landscape, the face of woodlands changed considerably. Stehlik et al. (2007) also noted increased extinction of light demanding species in deciduous forests in the Swiss lowland. Decocq et al. (2005) compared the herbaceous layer in a Thiérache forest between parts still managed as coppice-with-standards and parts converted into a selectively cut high forest system. Many so-called ‘ancient’ woodland species were associated with the traditional management system. Even more so, before 1900, woodlands probably also served as refuges for non-woodland species more than they do now. In an increasingly dynamic and hostile agricultural matrix, this seems an important, partly new ecological function for woods (Peterken & Francis 1999).

extinction in thiérache and comparison across contrasting landscapes

Of 959 species recorded by Riomet, 19% are presumably extinct in Thiérache. Especially, historically rare species were prone to extinction, a pattern which seems to be consistent among the few studies that have tested this (Robinson et al. 1994; Duncan & Young 2000; Lavergne et al. 2006).

To compare the extinction rate of native species in Thiérache with other contrasting landscapes, we surveyed the literature for studies of floristic change which used complete historical floras (listed in Appendix S3). We accounted for the temporal and spatial extent of the studies by calculating the percentage species loss per year and, using a linear regression, related this to the size of the study area. The percentage species loss per year calculated for 11 studies were negatively related (Fig. 5) with log-transformed size of the study area (F = 10.4, P = 0.010, r² = 0.531). The intercept and slope equalled 0.392 and –0.073, respectively. The negative relationship between extinction rates and size of the study area was expected because area buffers against extinction (e.g. May et al. 1995). Differences between these contrasting landscapes may be explained by the degree of environmental degradation and the strength of conservation actions (see Gaston 2005). Walker and Preston’s study (2006) on floristic change in Bedfordshire and Northamptonshire seemed most comparable to our study. Arable fields cover more than half the county areas, resulting in less grassland (10–17%) and forest (6–9%) in comparison with Thiérache. Simply considering the numbers of extinct species, almost twice as many species were lost in Thiérache (186) compared with the two counties (both lost 94 species). After standardization, the percentage loss of species per year in Thiérache (0.22) was still higher than for the counties (average 0.14). Perhaps earlier environmental degradation in Bedfordshire and Northamptonshire caused species losses before the main periods of historical data collection, or the numbers of historically rare species – which are most prone to extinction – have been lower than in Thiérache (see Thompson & Jones (1999): the numbers of scarce plants lost since 1970 from British vice-counties were positively related to the pre-1970 number of scarce plants). However, such explanations seem unlikely given the similarities of these rural landscapes. Moreover, two other aspects make the higher extinction rate in Thiérache paradoxical: (i) both counties, especially Bedfordshire, showed more signs of urbanization, and (ii) the sum of arable land and grassland was fairly constant for Thiérache (1820: 76%, 1999: 73%), whereas Bedfordshire and Northamptonshire saw a reduction from approximately 85% in 1866 to 65% in 2000. An important difference between the regions is the relatively long tradition of nature conservation in both counties (for instance, there are 16 and 14 local nature reserves, and one and two national nature reserves, in Bedfordshire and Northamptonshire, respectively), which might have prevented or delayed local extinction of many plants. Conversely, no nature conservation organizations are active and no protected areas exist in Thiérache. Although little is known about the effectiveness of protected areas (Gaston et al. 2006), its success should depend on the resilience of the landscape to disturbance or mismanagement which is inversely related to surrounding land use intensification (Bengtsson et al. 2003). As a rural depopulated landscape, Thiérache is considered less important compared with other parts of Picardie that are undergoing strong human pressure, for example, the suburbs of Paris. Unfortunately, protected areas in the latter landscapes may be less successful as they are islands in a matrix of otherwise intense land use.

Conclusions

Human population density decrease, no urbanization and a relatively stable landscape configuration did not spare the rural region of Thiérache from strong floristic losses over a period of 89 years (19% of vascular plant species went regionally extinct). We suggest that protected semi-natural areas and conservation management may delay or prevent historically rare species from extinction, but also, that the success of such a strategy depends on the intensity of surrounding environmental change. We plead for a landscape approach of biological conservation. Eutrophication, agricultural intensification, changes in forest management and secondary succession were strong actors of qualitative environmental changes. They caused trait-specific and predictable declines of plant species groups, irrespective of their historical rarity. On a relative scale, woodlands were better buffered against change than open habitats, but the absolute number of lost species remained high due to decline of habitat quality, within-habitat dynamics and loss of structural and microhabitat diversity.

To the extent that our results apply elsewhere, we should expect similar floristic loss in other landscapes despite an overall stable landscape configuration. We conclude that not only areas undergoing strong habitat loss should be targeted for conservation priorities, but also landscapes suffering from qualitative environmental changes, which might only reveal themselves after relatively long periods of time