Online Resource methodological details

1.1. Determination of light availability

To determine the proportion of visible sky in our plots, we took hemispherical photos under overcast conditions in the summer of 2014 using the HemiView camera system (DeltaT Devices). We divided the plot area into four quadrants and took three photos with different exposures within each quadrant. Subsequently, we selected one photo per quadrant with the best exposure, i.e., not under- or overexposed according to the histogram, and then applied the shadow/highlight function in Adobe Photoshop to maximize the light spectrum. We then extracted the blue channel to obtain the best separation of sky and non-sky (according to Frazer et al. 2001; Jelaska et al. 2006) and determined the threshold edge value, i.e., the brightness value that separates sky and non-sky, using the software SideLook (Version 1.1.01). Using this processed image file, we determined the proportion of visible sky in the skymap sectors above a zenith angle of 75° using the HemiView software and subsequently calculated the arithmetic mean of the visible sky values from the four photos.

1.2. Determination of soil conditions

We collected eight soil samples per plot (0-10 cm depth) in September 2014. The soil samples were mixed together plot-wise, air-dried and sieved (2 mm mesh).We analyzed the samples in the laboratory to determine the total carbon, total nitrogen, phosphorus, potassium, calcium, and magnesium contents and the pH value. The carbon and nitrogen contents were determined using a dry combustion method (ISO 10694 and ISO 13878). Phosphorus and potassium (per 100 g of soil) were determined photometrically using the double calcium lactate method at pH 3.6. The calcium content was measured using an extractant solution of ammonium acetate in combination with acetic acid and hydroxypropionic acid. To determine the magnesium content (per 100 g of soil) we used atom absorption spectrometry after extraction using a calcium chloride solution. Finally, the soil pH was determined in 0.01 M calcium chloride using a glass electrode following ISO 10390.

1.3. Adaptation of resurvey date to account for phenological shifts

We accounted for phenological shifts by first calculating the annual sums of hours (cumulative degree hours, CDH) with a temperature above 5 °C, i.e., the temperature at which plant metabolic processes start (Lindsey & Newman 1956), for each year from 1947 to 2013 using temperature data from two meteorological stations (Marnitz and Neuruppin, data available at Then, for every year and for a range of CDH between 20 °Ch and 50000 °Ch, i.e., the CDH that may be reached from the beginning until the end of the vegetation period, we calculated the days where a specific cumulative degree hour (in 20 °Ch steps) was reached. Using these data, we predicted the corresponding days of the year for which a specific CDH was reached in 1960 and 2014 using linear first-order autoregressive models. The CDH thresholds between April 1 and July 31 were reached between six and 17 (Marnitz station) or 14 and 30 (Neuruppin station) days earlier in 2014 compared with 1960. We averaged the values from the two stations and scheduled the resurvey dates based on the shifts that we found.

1.4. Determination of regional ancient forest indicator species

We determined the regional ancient forest species using the species distribution data from our previous studies in this area (Kolk & Naaf 2015; Naaf & Kolk 2015). We used the distribution data from 441 herb-layer species across 224 forest patches (104 ancient and 110 post-agricultural) and identified the regional ancient forest species by fitting generalized linear models (GLMM) with the glmer function in the lme4 R package (Bates et al. 2015), using a binomial link function and patch age (ancient vs. post-agricultural) as a predictor of species occurrence. Species with a significantly higher probability of occurrence in the ancient forests were classified as ancient forest species.

1.5. Selection of the optimization algorithm to identify winner/loser species

To identify winner and loser species, we used the glmer function in the lme4 package (Bates et al. 2015) using the Laplace approximation, Wald Z-test and Bobyqa algorithm. We used the Bobyqa optimization algorithm instead of the default Nelder Meat algorithm because this algorithm deals better with convergence issues (Mind the Brain Blog 2015). Convergence problems emerge when the optimization algorithm cannot reach the maximum improvement, i.e., the function does not converge to zero. This can be caused by omitting important random or fixed factors but may also result from other unknown reasons. This topic is currently debated among mathematicians and statisticians (stats.stackexchange 2015). Although we used the Bobyqa algorithm there were still convergence issues in some cases. Thus, we fitted all models using both algorithms (Nelder Meat and Bobyqa) and checked for differences in the fitted estimates, a method that is also proposed by the gmler help file lmerControl {lme4} (Bates et al. 2015). The differences in the coefficients were very small and the direction of the effects did not differ between the methods; thus, we can assume that our results are robust, although the coefficients can only be observed as an approximation of the real values.


Frazer GW, Fournier RA, Trofymow JA, Hall RJ (2001) A comparison of digital and film fisheye photography for analysis of forest canopy structure and gap light transmission. Agr Forest Meteorol 109:249-263

Jelaska SD, Antonic O, Bozic M, Krizan J, Kusan V (2006) Responses of forest herbs to available understory light measured with hemispherical photographs in silver fir-beech forest in Croatia. Ecol Model 194:209-218

Kolk J, Naaf T (2015) Herb-layer extinction debt in highly fragmented temperate deciduous forests - Completely paid after 160 years?. Biol Cons 182:164-172

Lindsey AA, Newman JE (1956) Use of official weather data in spring time - temperature analysis of an Indiana phenological record. Ecology 37:812-823

Naaf T, Kolk J (2015) Colonization credit of post-agricultural forest patches in NE Germany remains 130-230 years after reforestation. Biol Cons 182:155-163

Unpublished material and webpages:

Bates D, Maechler M, Bolker B, Walker S (2015) lme4: Linear mixed-effect models using Eigen and S4. R package version 1.1-8,

Mind the Brain Blog (2015) [last access: 23.09.2015]

stats.stackexchange (2015) [last access: 23.09.2015]