8. Discussion
8.1.2 Small-scale heterogeneous forest structure
The small-scale heterogeneous forest structure can be observed in four ways: by the development of the structural-complexity-index SCI, the develop- ment of the Shannon-index SI, by the age ranges of canopy trees, and by the number of development phases. Large age ranges of canopy trees sug- gest that those trees emerged from tree cohorts with different ages and sizes in close proximity to each other. Increasing SCI and SI values over the simulation period indicate a horizontally and vertically divers forest struc- ture, which can be termed heterogeneous. The SCI values exceed observed indices from the Serrahn beech forest (Schr¨oter et al., 2012) by far, which might stem from the absence of tree regeneration in the inventory of the mentioned study. The interplay of tree mortality and crown plasticity leads to long periods over which the canopy is interspersed with small gaps. These
gaps may close again, which might lead to the death of already established regeneration. As a consequence, the beech regeneration greatly differs in terms of age, height, and diameter, and this result is congruent with the find- ings from central Bohemian beech forests (B´ılek et al., 2014) and Ukrainian beech forests (Hobi et al., 2015a).
The question about the scale on which the heterogeneous forest struc- ture is observed may be difficult to answer, because the sizes of development phases of old-growth forests vary on the scale applied by the observer (Com- marmot et al., 2005). However, the characteristic mosaic pattern should be observable on an area of less than 1 ha (Tabaku, 2000; Piovesan et al., 2005; Alessandrini et al., 2011). The sizes of developmental phases in the sim- ulation experiment are smaller than 0.5 ha and refer to cohorts of trees (Fig. 43). The simulation with plastic tree crowns produced on average 9 detectable development phases of 156.25 m2 on a simulation area of 0.5 ha. This pattern is comparable to the results obtained from the model BEFORE (Rademacher et al., 2004), which clearly demonstrates that beech forests consist of a mosaic of areas with different developmental stages (on average 0.3 ha), although the grid pattern with cell sizes of 204 m2 impact the size of detectable phases, as phase sizes below cell size cannot be described. As the raster size of 12.5 x 12.5 m applied in this study can be occupied by one large beech tree alone, if the particular tree has a crown width of 20 m for example, then the scale of observation cannot be decreased further, because it is already at the scale of an individual tree. Thus, based on the findings from the sheer number of development phases, the forest structure can be described as small-scale heterogeneous.
The dynamic raster search with which development phases have been detected in this study is in contrary to Tabaku (2000), who applied a fixed raster grid. Thus, the number of phases in the simulation cannot be easily compared to this field observations, as the dynamic raster search may arti-
ficially increase the number of phases. On the other side, a fixed raster may underestimate the total number of detectable phases.
A very important mechanism that drives the mosaic pattern is the mor- tality submodel. As described in the model description (see section 2.2.9), the tree senescence, which means the process of decreasing tree vitality and subsequent tree death as irregular mortality, is based on the tree dimension, particularly the tree height. Tree senescence is initiated with a mortality index drawn from a normal distribution. Thus, trees may die as single indi- viduals or in groups depending on their mortality index, height, and position. However, empirical studies strongly suggest that structures and processes in old-growth beech forest are driven by gap dynamics that in turn rely on dis- turbance events, mostly storms (Piovesan et al., 2005; Nagel & Diaci, 2006; Trotsiuk et al., 2012). Single or groups of trees are destroyed which lead to the observed gap patterns by Hobi et al. (2015b), with large-scale distur- bances being very rare. Thus, it can be assumed that tree deaths caused by wind disturbance events occur in clusters. This death of tree groups is only mimicked in the BEEP model by trees which mortality indices, dimension, and positions are close to each other.
It can be argued that tree morality in the simulation must occur in clusters of groups of trees due to the above-mentioned development phases, which is similar to the observed gap patterns in old-growth beech forests (Hobi et al., 2015b). However, the mortality model used in the simulation experiment may not come up to the observed patterns of beech tree mortal- ity, because it does not take into account important drivers of tree mortality, such as drought stress (H¨ulsmann et al., 2016). This can only be improved by the incorporation of adequate empirical or mechanistic mortality mod- els based on long-term data derived from monitoring plots in near-natural beech forests. Therefore, the BEEP model is able to reproduce a small-scale heterogeneous forest structure, but the results are strongly influenced by the
applied mortality submodel. Thus, it remains uncertain, if the small-scale pattern could have been produced by other model formulation in the mortal- ity submodel. The sensitivity analysis already showed that forest structure result are very sensitive to assumptions made in the mortality submodel. However, the hypothesis that the BEEP model is able to reproduce a small- scale heterogeneous forest structure can be confirmed.