Study site and future climate treatment
Our study site is located at the Bad Lauchstädt Field Research Station, Bad Lauchstädt, Germany (51° 22060 N, 11° 50060 E), which belongs to the Helmholtz Centre for Environmental Research–UFZ. Long-term mean annual precipitation in the area is 489 mm and the mean annual temperature is 8.9 °C (ref. 32). During 2018 and 2019, Europe experienced a record-setting drought that was especially severe in 2018 (refs. 33,34); the mean annual precipitation at our study site in 2018 and 2019 was 254 mm and 353 mm, respectively, whereas 2017 was a more normal year, with a mean annual precipitation of 403 mm. Mean annual temperatures were above average: 2017, 10.5 °C; 2018, 10.8 °C; 2019, 11.2 °C (data from the weather station at the Bad Lauchstädt field station). The soils in the study area are fertile Haplic Chernozem type32,35.
Our eDiValo experiment was conducted in the GCEF, which was designed to investigate climate change effects under different land-use scenarios32. We used 10 ‘extensively’ used pastures of the GCEF in our experiment; that is, 384-m2 (16 × 24 m) areas of grassland (hereafter called ‘pastures’) that were grazed by a flock of 20 sheep 2–3 times each year. Grazing was implemented as short-time high-intensity grazing events, each lasting 24 h (ref. 32). This type of high-intensity but short-term grazing is considered better in maintaining species richness as it gives plants more time to recover between grazing events36. It is also a recommended management type for nature conservation areas in Germany37. Vegetation in the pastures was species-rich grassland vegetation that is typical of drier regions of central Germany32,38. The whole GCEF was fenced to exclude native large mammalian herbivores (for example, deer); however, European hare (Lepus europaeus), wood mice (Apodemus sylvaticus) and voles (Microtus arvalis) are common at the site.
Our experimental design was originally intended to test the dependence of light competition on nutrient and herbivory under current and future climatic scenarios. Although we included both climate treatments in our data, climate was never significant for richness and Shannon diversity, either alone or in interaction with other factors, and our focus was therefore on the other treatments. Five of the above random pastures received future climatic treatment which was based on different dynamic regional climate models for Germany, all predicting an increased mean temperature by approximately 2 °C year-round, strongly decreased summer precipitation and slightly increased spring and autumn precipitation (https://www.regionaler-klimaatlas.de/) (ref. 32). Passive night-time (after sunset and before sunrise) warming through the use of roller blinds attached to the GCEF roof and eastern and western wall structures was used to increase the air temperature. In each spring (1 March–31 May) and autumn (1 September–30 November), future climate plots received 110% of the ambient rainfall and in the summer (1 June–31 August), they received 80% of the ambient rainfall. The precipitation treatment was adjusted weekly and compensated for a possible night-time reduction in rainfall due to temperature treatment. A detailed description of the future climate treatment is provided in a previous report32.
Fertilization, herbivore exclusion and light addition
We first tested whether adding light can offset the negative effect of fertilization on plant diversity. In May 2017, we established a full-factorial experiment of fertilization and light addition. Within each 10 pastures (5 in ambient climatic conditions, 5 in future climatic conditions), we established 4 plots of 1.4 × 1.4 m, separated by a 1-m buffer zone (hereafter called ‘blocks’), in total 40 plots and 10 blocks. At the time the experiment was established, vegetation in the whole experimental area (that is, in a block of 4 plots and the surrounding 1-m area) was trimmed to a height of 5 cm to make conditions uniform and the whole area was temporarily fenced to let the experiment establish and fertilization effects develop. The temporary fence was removed in August when the herbivore exclusion treatment was started. Therefore, there was no grazing by sheep in the experimental plots in the summer of 2017. Two randomly chosen plots received fertilizer treatment and two were controls. For the former (fertilizer-treatment plots), slow-release granular NPK fertilizer (a mixture of Haifa Multicote 2 M 40-0-0 40% N; Triple Super Phosphate (TSP) 45% P205; and potassium sulfate fertilizer 50% K2O, 45% SO3) was added twice per growing season, in a total of 10 g N, 10 g P and 10 g K per m² (see ref. 3 for a similar protocol that is used in grasslands worldwide). In 2017, the first fertilization was done at the beginning of June right after establishing the experiment and the second fertilization was done at the beginning of July. In the subsequent years, the first fertilization was done at the beginning of the growing season (late March–April) and the second fertilization was done in June. In 2019, two previously unfertilized plots were accidentally fertilized and were thereafter treated as fertilized plots.
To manipulate light, 1.4 × 1.4-m plots were further divided into two subplots, 0.7 m × 1.4 m each, and one of these was randomly assigned to the light-addition treatment, resulting in 80 subplots (Fig. 1). We installed two 120-cm-long and 3.5-cm-wide recently developed LED lamps (C65, Valoya) parallel to each other and at a 28-cm distance from each other to each light-addition subplot. To increase light for the small understory plants that are the most likely to suffer from competition for light, we installed the lamps 10 cm above the smallest plants. The lamps were gradually uplifted over the course of the growing season to follow the growth of the smallest plants. As our light-addition treatment was intended to mimic natural sunlight (that is, making a gap in a dense vegetation and allowing the sunshine in), we chose the spectrum of the lamps to include all wavelengths of sunlight, including small amounts of ultraviolet and infrared. Each lamp added roughly 350–400 µmol and did not alter the air or aboveground soil surface temperature (Fig. 1b), which is an improvement on previous studies12. Each year, we added light during the active growing season: the lamps were switched on early in the spring (March–April), when temperatures were clearly above zero, and switched off and removed when temperatures dropped close to zero in November–December and aboveground plant parts had died and formed litter. Each day, the lamps were set to switch on two hours after sunrise, and to switch off two hours before sunset, and when the temperature exceeded 28 °C to prevent overheating. We did not install unpowered lamps to unlighted plots because our modern, narrow LED lamps caused minimal disturbance (see below) and no heating (Fig. 1b), and because unpowered lamps would have added an artefact in that they create shade that does not occur when the lamps are on in lighted plots.
At the end of August 2017, after running the fertilization–light-addition experiment for one growing season, we expanded the experiment by implementing the herbivore exclusion treatment in a full-factorial combination with the other treatments. Two of the previously established 1.4 m × 1.4-m plots, one with and one without the fertilization treatment, were randomly allotted to the herbivore (sheep) exclusion treatment and fenced with rectangular metal fences of 1.8 m × 1.8 m, 82 cm height and 10 cm mesh size. At the same time, the temporary fence established in May 2017 was removed from around the whole experimental area, allowing the grazing of sheep in unfenced plots. The fences did not exclude mice, voles and hares. For the time of each grazing event, lamps in grazed subplots were removed and switched off in the ungrazed subplots. Uplifting the lamps from grazed plots did not cause disturbance because vegetation in grazed plots was always short and did not reach above the lamps. Inside exclosures, lamps were always kept in place during the growing season, and plants could freely grow around and above them.
Plant community and trait sampling
In July 2017, we established 50 cm × 50-cm permanent quadrats in every subplot for plant community sampling. We visually estimated the per cent areal cover for all species occurring in the quadrats, and litter cover, from the beginning of June to mid-June 2019, when the vegetation was at its peak biomass. The 2017 sampling happened later, in mid-July, because vegetation in all plots and surrounding areas was trimmed to a height of 5 cm at the time of the establishment of the experiment at the end of May, and it took later for vegetation to reach its peak biomass. In 2018, the effects of drought were devastating, and most plants had senesced or died before the planned sampling date; we therefore omitted the year 2018. At the beginning of each growing season—that is, when the lamps were installed and switched on—there was very little live biomass in the plots, and the maximum height of existing plants was approximately 5 cm (in all plots). During the peak biomass the maximum plant height was up to approximately 1 m; however, it varied greatly between the treatments and was especially low in grazed plots. All vegetation surveys were done by the same trained and experienced person with a minimum estimate threshold of 0.1%. We used plant cover data to calculate species richness and Shannon diversity.
In May–June 2020, we measured plant height (centimetres), SLA (leaf area in square millimetres per milligram of dry mass), foliar C:N (based on the per cent C and N in plant leaves) and LWC (leaf water content as 1,000 − LDMC (the ratio of leaf dry mass to saturated fresh mass), expressed as milligrams per gram39) for most species occurring in the experimental plots, and complemented the trait data from the TRY Plant Trait Database40,41,42 (v.5.0; https://www.try-db.org/TryWeb/Home.php) and for one species one trait value from another source9. The trait data were collected from seven to ten individuals per species from the study site or close areas; the collection and handling followed standard protocols39. We chose these traits because they are widely documented to be associated with responsiveness to soil nutrients, herbivory and light9,26,27,43,44,45,46. We used all traits as, although they partially reflect similar ecological adaptations (for example, leaf economics spectrum43), they could also potentially reflect independent and distinctive processes, and differently mediate the responses of species to our treatments. For example, SLA and LWC in our dataset correlated weakly (r2 = 0.16), but were to a greater extent uncorrelated (Extended Data Table 6), and could function differently, for example, in light capture and drought tolerance26,39. In 2017, our trait data covered on average 97.7–98.6% of the total cover in the plots, the value slightly differing depending on the trait as we did not have all traits for all species. Our own trait collections covered on average 96.6–97.6% and TRY data covered on average 0.9–2% of the total cover. In 2019, the whole trait data covered on average 99.5% of the total cover in the plots, again slightly depending on the trait. Our own trait collections covered on average 94.2–96.5% and TRY data covered on average 2.7–5.3% of the total cover.
Abiotic environmental measurements
We measured several soil and other environmental properties from the experimental plots. Light availability (photosynthetically active radiation; PAR) in unlighted and lighted (under lamps) subplots was measured using LI-190R and LI-250A meters (LI-COR), approximately 7–10 cm under the lamps and 15–20 cm above ground level. We measured light availability from the same distance to the ground in unlighted plots. Measurements of light availability were done in mid-July 2020 on three consecutive cloudless days around noon. Note that in grazed plots, light levels between lighted and unlighted plots are more similar than inside exclosures (Fig. 1), because herbivores keep the vegetation short, and natural sunlight can therefore reach under the lamps where the light measurements were taken. Air temperature and humidity were recorded from unlighted and lighted (under lamps) subplots using loggers (HOBO MX2301A, Onset Computer Cooperation) that were installed approximately 7 cm under the lamps and to the same height from the ground in unlighted plots, and were replicated under different combinations of fertilization, herbivore exclusion and light addition in ambient climatic conditions three times (n = 3). The logger data were collected in May 2019 before the effects of drought were visible.
Statistical analysis
We analysed our data in two steps. First, to test whether competition for light mediates the effect of fertilization on diversity, we analysed the effects of fertilization and light and their interaction on species richness and Shannon diversity using data from 2017, when the herbivore exclusion treatment had not yet been implemented. We also analysed the effects of treatment on total vegetation cover and litter cover. We fit LME models in which diversity (species richness and Shannon diversity), total cover and litter cover, each in their own model, were explained by fertilization, light addition and their interaction (fixed variables). All treatments were categorical variables with two levels (treated and untreated). In each model, subplot was nested within plot, which was nested within block (nested random variable). We simplified the models using the anova() function for model comparison in the nlme and lme4 packages in R (ref. 47) (on the basis of log likelihood ratio tests; P ≥ 0.05; Extended Data Table 2). This was done to uncover the significance of the main effects and interaction terms, to avoid overparametrization47,48 and to provide model-derived parameter estimates for the figures (Extended Data Table 5). However, we also provide full model results that are qualitatively similar to the results of simplified models (Extended Data Tables 3 and 4); therefore, model choice did not affect our conclusions. Climate treatment was included in all original models but was never significant for richness and diversity, and was not considered further. Total cover and litter results for 2017 are reported in Extended Data Figs. 1a,b and 3a). As there was heterogeneity in the variance structure between treatments, we used the varIdent() function in the nlme package in R to allow each treatment combination to have a different variance. Model fit was inspected using model diagnostic plots in the package nlme. In the full design with climate included, the number of replicates per treatment combination was ten.
Second, to include herbivore exclusion to the experimental design and to test whether competition for light mediates the effect of herbivore exclusion on diversity, and whether competition for light, herbivory and fertilization interact, we analysed the effects of herbivore exclusion, fertilization, light and their interactions on species richness and Shannon diversity using data from 2019. All treatments were categorical variables with two levels (treated and untreated). We also analysed the effects of treatment on total vegetation cover and litter cover. We fit similar models to those described above, except that herbivore exclusion was an additional fixed factor in the models. We simplified the models, used the varIdent() function to account for heteroscedasticity and checked the model fit using model diagnostic plots, as above. Climate treatment was included in all original models but was significant for litter cover only, and was not considered further. In the full design with climate included, the number of replicates per treatment combination was five.
To further assess which plant traits increased the probability of species benefiting from the addition of light, we first created a binary response variable: those species that increased from unlighted to lighted plots (that is, had a higher value in a lighted than an unlighted plot) were given a value of 1 and those that did not were given a value of 0. This response variable takes into account rare species that emerged or persisted in the lighted plots but were absent in the unlighted plots (that is, species gains and losses) and changes in small, subordinate species (those that are likely to benefit from light addition) with small but consistently trait-dependent changes in response to light. It is also in line with our species richness analyses, as species gains and losses ultimately determine richness responses. We did not use different indexes (for example, lnRR or RII) because these could not handle multiple zero values and species losses or gains (that is, species having zero cover in either unlighted or lighted subplots). Second, we fit GLME models with a binomial error structure (family = “binomial”, link = “logit”) in which a probability of a species increasing from unlighted to lighted plots was explained by categorical experimental treatments (fertilization, herbivore exclusion and their interactions), traits (SLA, height, LWC, foliar C:N), and interactions between the treatments and traits. Each trait was analysed in its own model as some of the traits were correlated (Extended Data Table 6), and to avoid overly complex models and overparametrization47,48. We included all species for which we had traits in the models. As we calculated the increase in cover from unlighted to lighted plots, our smallest experimental unit in trait analyses was a plot (not a subplot, unlike in other analyses). As there were several species in the same plots, we nested species within plots, and plots within blocks. We similarly simplified the models to include only significant variables (on the basis of χ2 tests; P ≥ 0.05). We did not include a crossed random effect for species in the models because the full models with a more complex random structure did not converge; however, when we refitted the simplified models with a crossed random effect for species, we found that the models converged (with scaled data) and that the significance of the effects remained qualitatively the same. Climate was included in all original models but was never significant. In addition, C:N and height did not predict the responsiveness of species to light in either year (P ≥ 0.13 for both); results are therefore not shown. In the full design with climate included, the number of replicates per treatment combination was five; however, the number of observations was greater (see Fig. 4 and Extended Data Fig. 4). To make sure that our results for SLA and LWC were not influenced by whether they were analysed in separate models or in the same model, or by the order in which they were in the models, we also performed analyses in which both SLA and LWC were included (in both orders). Results remained qualitatively similar and are not discussed further.
Furthermore, to check whether our trait results were driven primarily by species gains and losses or changes in abundance, we ran additional trait analyses for which we calculated the change in cover between lighted and unlighted subplots (cover in lighted subplot − cover in unlighted subplot), and analysed the ‘change’ with otherwise similar trait models to those described above, except that we used Gaussian error structure. With this index, which gives a disproportionate importance to the abundant species, we found that traits were poor predictors of changes in cover between lighted and unlighted plots (all interactions were non-significant, P > 0.05, except for a marginally significant C:N × fertilization interaction in 2017 that was no longer visible in 2019; results not shown; codes and data available in the Dryad repository). We also analysed presence–absence-based species losses and gains. In these models, each species was given a value of 1 when it was present in the lighted subplot but absent from the unlighted subplot; otherwise, these models were similar to the binomial trait models described above. These models produced, to a large extent, similar results to our models using the probability of increase in response to light as a response variable (results not shown; codes and data available in the Dryad repository). These additional analyses and results support using the probability of increase in response to light as our response variable, rather than abundance-based metrics, as it includes both gains and losses and abundance aspects, and is therefore a general test that is well suited to assessing species gains and extinctions and changes in subordinate species.
All statistical analyses were performed using R v. 4.0.0 (ref. 49). We used the nlme package (v.3.1.147) for LME models50, the lme4 package (v.1.1.23) for GLME models51, and the car package52 for P values (v.3.07).
Reporting summary
Further information on research design is available in the Nature Research Reporting Summary linked to this article.
Source: Ecology - nature.com
