The 11 sites encompass a large amount of climatic variation. Not only do they differ in terms of total rainfall, temperature (both given in ), solar radiation, etc., but they also vary considerably in the seasonality and periodicity of these climatic parameters. The first axis of the PCA on site climatic parameters (75·7 % of the variation) was largely associated with a gradient of aridity coupled with site temperature and growing degree-days (). PCA axis 2 (22·1 % of the variation) was overwhelmingly linked to annual rainfall, separating two patterns underlying low aridity: high rainfall and moderate temperatures year-round, as opposed to true mountain sites with modest rainfall but low evaporation due to low temperatures during rainy periods. The projections of the sites on the two axes make it possible to separate the following groups of sites: (1) a group of Mediterranean sites with high temperatures and aridity (GR-LAG, PT-MER and IS-KDE: site abbreviations given in ) – FR-HGM is linked with this group by high mean annual temperature but is distinguished by higher rainfall and lower aridity; (2) a group of colder sites but still with relatively low rainfall and medium values of aridity index (GE-MNP, CZ-OHR, SE-BAL); (3) a group of sites with similar temperatures to the above, but characterized by high rainfall and thus low aridity (SC-SUT, FR-ERC); and (4) a group of true mountain sites (FR-LAU, NO-BER) characterized by low temperatures, a short growing season, low aridity and medium rainfall.
Fig. 1. Principal component analysis on climatic data: mean annual rainfall (Rainfall), mean annual temperature (Temp), annual growing degree-days (GDD, and an aridity index for each site. Annual PET and solar radiation were removed after a covariance analysis (more ...)
Based on these results we chose to use synthetic climate indices such as the aridity index and projections of sites on PCA axis 1 () as covariates for analyses of response of traits and ecosystem properties to land use.
Disturbance indices of VISTA treatments reflect the wide variety of land-use systems that characterize marginal landscapes across Europe, as shown by the return interval and intensity of disturbances for the least and most used treatments in each site (). The site with the highest disturbance was a short-rotational crop/fallow field in Scotland (SC-SUT: short return interval with the highest intensity), and the lowest was a pasture in Sweden (SE-BAL: longest return interval without disturbance) abandoned for more than 60 years. Aggregated return time of disturbances (A) and biomass removal by disturbance (B) show that the GE-MNP, IS-KDE, PT-MER and SE-BAL sites all span a considerably large range on the disturbance gradient, because they cover situations from continuing, moderately intense management to advanced secondary succession. The magnitude in disturbance regimes is more restricted at the NO-BER, FR-ERC, FR-LAU and FR-HGM sites. Among these, FR-ERC treatments are highly disturbed, whereas the abandonment sequence at FR-HGM belongs to the lower end of the total VISTA disturbance gradient. All other sites are intermediate with respect to variation in either disturbance frequency or biomass loss.
Fig. 2. (A) Return interval (log scale) and (B) intensity (expressed as a percentage of maximum standing biomass) of disturbance for the least (open columns) and most (shaded columns) intensively used treatments in each site. In A, a value of 0 indicates a return (more ...)
The DCA () revealed significant turnover at the family level across all the sites and plots for the first two axes of the ordination (axes lengths of 6·58, 5·88, 2·82 and 2·55, respectively). These first two axes also explained a substantial part of the variation (percentage of total inertia explained of 10·9 and 8·1 compared with 3·5 and 2·8 for axes 3 and 4, respectively). The ordination clearly showed that many of the sites were relatively similar in cover at the family level with a dominance of Poaceae, as well as Apiaceae, Asteraceae, Rosaceae, Caryophyllaceae and Plantaginaceae. At high values on Axis 1 were the Norwegian sites (NO-BER) characterized by substantial cover of Ericaceae and Salicaeae, as well as the late successional Swedish sites with high values for the Ericaceae. Low values on Axis 1 were associated with Mediterranean woodland and scrub-dominated plots, including the GR-LAG and PT-MER sites dominated by Cistaceae and Fagaceae (Axis 1 scores of approx. 3).
Detrended correspondence analysis (DCA) of relative abundance of higher plant families for each recorded plot. Only families with a weight of >2 % of the maximum family's weight are shown as the first eight letters of their full name.
Responses of plant traits to land-use change: an overview
Continuous traits (aggregated values) can be grouped into three categories, according to their sensitivity to treatments across sites (): (1) traits that respond to treatments in most sites – reproductive plant height, stem dry matter content, leaf phosphorus concentration and flowering time; (2) traits that respond in half to two-thirds of the sites – specific leaf area, vegetative plant height, leaf nitrogen concentration and leaf dry matter content; and (3) traits that respond in approx. one-third of the sites – seed mass and leaf carbon concentration. The REML analysis across the 11 sites highlighted strong and consistent responses across sites for all traits except seed mass. Direction of change was overall consistent for all traits with a significant response. Decreasing land use resulted at the community level in dominance by taller plants (increased vegetative and reproductive height), plants with more conservative leaf syndromes (decreased SLA, LNC and LPC, increased LDMC associated with increased StDMC), and delayed flowering phenology. These responses were overall strongly consistent across sites, and the few exceptions resulted from either canopy closure after woody colonization (e.g. Sweden, Greece) or specific soil conditions (water-logging in Germany).
Multivariate analyses conducted at four individual sites identified consistent patterns in variation in plant strategies. At the SC-SUT site, Pakeman and Small (2004)
showed that land use was a significant predictor of plant traits as abandonment of rotational cultivation led to an increase in species with more conservative nutrient use, had later flowering and larger seed sizes. However, species composition did not recover in the direction expected from analysis of unploughed areas. Overall, these results were repeated at the French alpine site (FR-LAU), where cessation of fertilization or of mowing resulted in the dominance by taller plants with more conservative nutrient use and later flowering, although seed mass did not vary across land-use treatments (Quétier et al., 2006
). A path analysis conducted on the data from the French Mediterranean site (FR-HGM) demonstrated that abandonment of vine cultivation led to the progressive replacement of species with small stature, high rates of resource capture and early flowering by species with the opposite characteristics, while seed mass did not show any significant trend with time after abandonment (Vile et al., 2006
). Finally, Lepš (1999)
demonstrated significant changes in species composition as a response both to fertilization and to cessation of mowing. Plant height was a very good predictor of species success under fertilization, as only tall plants were able to survive in competition.
Responses of ecosystem and soil properties to land-use change: an overview ()
Maximum above-ground live (AGBmass) and dead biomass (AGTotDead) were significantly different among treatments in ten out of the 11 sites, but the effects of treatments were not necessarily comparable for the two variables (). By contrast, owing to the very tight correlation between maximum AGBmass and ANPP within (data not shown) and across sites (r = 0·89, P < 0·001, n = 170), the effects of treatments on these two variables were identical. Treatments had significant effects on average community decomposition rate measured under standard conditions, when there was no interaction with environmental conditions, in six out of the 11 sites. This reflects a decline in litter quality (C. Fortunel et al., unpubl. res.), which is a major determinant of in situ decomposition rates (Swift et al., 1979; Lavelle, 1993). The results of the in situ experiment are more complex (data not shown), however, as changes in the decomposition environment within sites also play a major role.
There was also a very close relationship between soil C and N both within (not shown) and across (r = 0·87, P < 0·001, n = 194) sites, with the consequence that treatment effects were almost identical on the two variables (the SC-SUT site being the only exception). Interestingly, nitrogen limitation as detected by the NNI had little concordance with information given by soil N. Phosphorus availability to plants as detected by soil P-Olsen was significantly different among treatments in nine of the sites, and except for two sites (IS-KDE and FR-LAU), these differences were consistent with those found using the PNI.
Table 5. Summary of significant differences and direction of response to decreasing land use in selected ecosystem, community and soil properties among treatments (summarized in column 2) in the 11 sites of the VISTA project (results of within-sites one-way ANOVAs) (more ...)
A methodology to link changes in disturbance regime, plant traits and ecosystem properties
One of the methodologies that can be used to assess the effects of land-use change on the functional properties of the vegetation through plant traits and environmental variables consists in successive sets of general modelling analyses. We present here the case of leaf dry matter content and above-ground total dead biomass as an illustration. The approach consisted of first identifying the effects of climate and land-use factors on LDMC, and then testing whether land-use effects on AGTotdead may be explained by changes in LDMC and/or in climate.
Accumulation of dead material represents the net balance between rates of production and decomposition of dead matter. Decomposition has been shown to be slower in litter produced by plants with high leaf dry matter content (Garnier et al., 2004
; Kazakou et al., 2006
). Our hypothesis was therefore that a reduction of management pressure would cause litter to accumulate in communities as LDMC increased due to slower decomposition (Tables and ). We further investigated whether interactions with climate were detectable across sites.
Figure A shows that across sites, less intensive land use led to communities with higher LDMC-aggregated values (Wald statistic = 8·45, P
< 0·001). Within sites, differences were significant for five sites, but the direction of change was consistent for ten of the 11 sites. Aggregated LDMC tended to decrease with mean annual temperature (P
= 0·10) but not with aridity or the first axis of the PCA. The increase in LDMC in response to less intensive land use was greatest at warmest sites (significant interaction; Wald statistic = 21·73, P
< 0·001), possibly as a result of a higher intensity of land use increasing the dominance by annuals at the warmest (and especially Mediterranean) sites. At sites where the differences in LDMC among treatments were largest, dead material tended to accumulate in large quantities (B). The quantity of dead biomass was, in some plots, up to four or five times greater than the live standing biomass, with the largest differences evident in abandoned plots (e.g. FR-HGM, CZ-OHR, SE-BAL). As hypothesized, there was a positive relationship between aggregated-LDMC and above-ground dead biomass (). Five points appear as outliers in this relationship: three are from the CZ-OHR site (open circles) and two from the FR-HGM site (closed squares). In these two cases, the plots concerned are dominated by large tussock grasses with high LDMC leaves (Molinia caerulea
in the Czech site, Brachypodium phoenicoides
in the French site). The large accumulation of dead biomass in these plots is probably linked to the combination of high biomass production, substantial leaf turnover and particularly low rate of decomposition of the litter produced by these species. For example, we have shown that the rate of decomposition of the litter produced by B. phoenicoides
was the lowest among the species screened at the Mediterranean site (Kazakou et al., 2006
Fig. 4. (A) Community-aggregated values of leaf dry matter content (LDMC), and (B) above-ground total dead material (which includes standing dead biomass and litter fallen on the ground) at the time of peak biomass for the least (open columns) and most (shaded (more ...)
Fig. 5. Relationship between community-aggregated leaf dry matter content (LDMC) and above-ground dead biomass (AGTotdead) at the time of maximum standing biomass across ten sites of the VISTA project, using the least/most used treatments in each site. r values (more ...)
Across sites, both land use and LDMC significantly influenced total dead biomass (Wald statistics = 33·61 and 36·29, respectively, P < 0·001 in both cases). In the ‘most used’ plots, total dead biomass was significantly lower than in less disturbed plots, and plots with increasing LDMC had greater total dead biomass. A significant interaction term (P = 0·007) indicates that the higher the mean LDMC at a site, the greater the effect of utilization in reducing the amount of dead matter. However, no significant effects of mean annual temperature or any other climatic variables on total dead biomass were detected.
Together, these results suggest that owing to higher litter input rates, and/or lower decomposition rates, dead matter accumulation is greater in less disturbed conditions. Community aggregated LDMC provides a good indicator of these changes. This relationship emerges partly because both LDMC and litter inputs decline with increasing land use. A further potential mechanism is that litter produced by plants with a higher LDMC decomposes more slowly (Garnier et al., 2004
; Kazakou et al., 2006
), leading to a longer residence time for such litter in the plots. Furthermore, it is interesting to note that, across sites, the correlation between above-ground total dead biomass and LDMC is significant for least used plots (r
= 0·86, P
< 0·001) but not for most used plots. This is in part explained by the much lower range of variation in LDMC across most used plots.