Study area, refaunation and earlier butterfly surveys
The Milovice military training range (50.26N, 14.89E, altitude 200–250 m a.s.l., mean annual temperature 8–9 °C, annual precipitation 500–600 mm) (Figure 1) was established in 1904, originally on 34.6 km2. It was subsequently used by all armies that operated on Czech territory, gradually expanding its area to 40 km2. The last users were the Soviets, who operated an air force base and headquarters here for the former Czechoslovakia until 1991. The natural setting is the gently rolling Středočeská Tabule Plain formed by Mesozoic carbonate-rich sandstones, siltstones and claystones, and covered by brown soils, rendzinas and carbonate rich sands. Woodlands dominated by Quercus petraea, Pinus sylvestris and Betula pendula are interspersed by finely grained mosaics of shrublands, grasslands and early successional vegetation that developed on former farmlands (mainly wheat, vegetables and diary family farms) and were utilised for training troops for over 80 years (Cizek et al. 2013).
Following the cessation of military use, parts of the open training fields were developed (golf course, amusement park, industrial zone), while three large areas were proclaimed a Site of European Community Importance (SCI) Milovice-Mladá. The Central site (local toponym: Pozorovatelna, hereinafter “C”, 50.254N, 14.881E) has been partly managed by conservation grazing by fenced sheep, while the Northern (Traviny, “N”, 50.278N, 14.883E) and Southern (Pod Benáteckým vrchem, “S”, 50.241N, 14.886E) sites remained unmanaged, except for occasional disturbance of S by armoured vehicles practiced by military history enthusiasts and for domestic cattle grazing in a corner of N in 2014–2016. Much of all three sites had suffered succession-driven homogenisation of the once diverse vegetation mosaic by competitively dominant grasses (mainly Calamagrostis epigejos and Arrhenatherum elatius), ruderal forbs and shrubs (mainly Crataegus, Prunus and Rosa).
The site S (2015–2017, 40 ha; 106 ha since 2018) has been grazed since spring 2015 by ≈35 Exmoor ponies (hereinafter “horse”) and ≈20 Tauros cattle (hereinafter “aurochs”). Since spring 2016, ≈35 horses and ≈20 wisents graze the site N (125 ha) (Figure 1). Both S and N are thus year-round cross-grazed by horses and big bovids (aurochs or wisent) living in naturally structured social units, i.e. mixed sex/age harems/herds. To provide variable management regimes, both temporally and permanently ungrazed plots of various sizes (units to tens of hectares) are present both within and outside the grazing reserves at any given time. The animals receive no supplementary feeding and no medication, except for strictly determined individual cases, and predators enter the sites freely (Jirku et al. 2018). Wolf, as a re-expanding apex predator, is not present yet, but its colonisation is expected. To control grazing intensity, facilitate gene-flow and avoid social stress, two to three year-old surplus animals are transferred to similar projects in the Czech Republic and abroad.
The first modern butterfly survey of the area was conducted immediately after the cessation of military use. Matouš (1994) published a commented list of species, treating the entire military range as a single locality. Fifteen years later, in 2009, the training fields S, C and N were surveyed in a semiquantitative manner (Čížek et al., 2013). The current monitoring of the refaunation impact, launched in spring 2016, thus represents the third survey.
Current butterfly monitoring
We set 16 rectangular plots (50 x 200 m) at both refaunated (n = 7) and neglected (n = 9) sections of N and S sites (n = 8 each) (Figure 1). From spring 2016 onwards, the plots were visited five times each year (May, early June, late June, July, August) to cover seasonal aspects of butterfly assemblages. The recording followed the timed survey protocol by Kadlec et al. (2010), appropriate for heterogenous environments with temporally changing locations of butterfly resources, such as flower patches. Each visit to a plot lasted 30 min, all butterfly species present were recorded using a net when necessary and taking vouchers of species not recognisable in the field. Abundances were recorded, using rounded numbers for species seen in large abundances. We also recorded the closest hour, cloudiness (3-points ordinal scale, from clear sky – 1 to overcast – 3), wind (Beaufort scale 1 – 4, i.e., calm to gentle breeze), and nectar supply (0 – no flowers within the plot, 1 – flowers scarce but present, 2 – flowers moderately abundant, 3 – flowers abundant). We restricted the visits to the highest butterfly activity period (10 AM – 4 PM) and to weather suitable for butterflies, randomising their sequence with respect to time of day. A single round of visits took 2–3 consecutive days.
Statistical analyses
For the past-present comparison, we visualised the patterns defined by species presences/absences recorded by Matouš (1994), Cizek et al. (2013) and the current monitoring, the latter collated across the four years, using the correspondence analysis (CA), an unconstrained ordination appropriate for 1/0 data, in CANOCO, v. 5.0 (Ter Braak and Smilauer 2013). We computed four variants of CAs: 1) based on three “samples” defined by the three consecutive surveys; 2) differentiating records from the locations N, C and S (possible using Cizek et al. (2013) and the current data), thus obtaining six “samples”; and 3+4) as in the previous two cases, but after exclusion of migrant and arboreal species.
We interpreted the CA results by three sets of the constituent species traits (Table 1, Appendix 1): (a) life history traits, mostly associated with feeding modes, dispersal and population structure, as compiled for Central Europe by Bartonova et al. (2014); (b) climatic niche traits, compiled by Schweiger et al. (2014) on the basis of species ranges in Europe and known to contribute to population trends (Essens et al. 2017); and (c) conservation attributes describing the distribution and Red-list status in the Czech Republic. We used the CANOCO option “explanation of species scores for functional traits”. This analysis, a multivariate version of the fourth-corner approach (Legendre et al. 1997; Dray et al. 2014), relates the species ordination scores from the CA ordination to trait values of the species, testing for strengths of the relationship using redundancy analysis (RDA), a multivariate version of linear regression (Ter Braak and Smilauer 2013). We analysed the three sets of traits separately, using the forward selection process to attain best-fitting traits combinations.
To compare numbers of butterfly species and individuals recorded during the current monitoring, we used 2-way analysis of variance with factors year (4 levels) and management (2 levels, refaunation vs. neglect, the latter including the plots grazed by cattle in 2016–17). Cumulative numbers of species and summed numbers of individuals across the five yearly visits were the dependent variables.
To study the current composition of butterfly assemblages, we used canonical correspondence analysis (CCA), a constrained ordination method relating the species composition of samples to external predictors and testing the relationships of species composition to predictors using the Monte Carlo test with 999 permutations, again in CANOCO. We reflected the temporal dependency in our data using a hierarchical split-plot permutation design, permuting the individual plots randomly, and the 20 subsequent visits per plot as mutually dependent cyclic shifts. We first ran separate tests for all possible nuisance covariables, i.e. year (both as 4-levels factor and as a linear value), site (N vs. S), hour (as factor and 2nd-degree polynomial), weather (a combination of cloudiness and wind), nectar and plots position (forward-selected from latitude, longitude, their polynomials and interaction).
For the pivotal effect of refaunation itself, we used two different codings, aiming to answer two slightly different questions. Refaunation (3-levels factor: refaunation, cattle and neglect) aimed to disclose the effect of wild ungulates, while ungulates (5-levels: horse, aurochs, wisent, cattle and neglect) aimed to decipher effects of the three ungulate animals. We also tested for military vehicle effect (2-levels factor tanks). We developed the models by systematically adding the covariates that had significant effects in the single-term CCAs to the refaunation and ungulates models, until we reached models that significantly explained the distribution of monitored butterflies while being stringently controlled for nuisance effects.
Analogously to the past-present comparison, we interpreted the final CCA current monitoring models by species traits, relating the CCA scores to the three sets of traits and using forward selection to select the best-fitting traits combinations.