Species distribution models as a tool for forest management planning under climate change: risk evaluation of Abies alba in Bavaria

Abstract

Questions: How can SDMs be adopted as a tool for forest management planning? Based on presence-absence data, which modelling techniques are appropriate to determine species potential distribution for forest management planning under climate change? Do species distribution models (SDMs) agree with expert knowledge about species distribution and species traits? How can forest researcher deal with distribution data of a species whose distribution is heavily affected by human impacts?Location: Bavaria (Southern Germany). Methods:We used SDMs based on the Second National Forest Inventory from 2002 (4 Â 4 km grid) containing presence-absence data of tree species to identify species environment relationships ('Grinnellian niche'). As an example, the distribution of silver fir (Abies alba Mill.) was modelled. Site condition data of the plots were derived from solar radiation, climate and soil maps. Models applied were boosted regression trees (BRT) and generalised additive models (GAM). Model predictions were compared with an expert based evaluation of the potential natural vegetation and were run with a climate change scenario (WETTREG B1) to project future distribution of silver fir.Results: Both models discriminated well between presence and absence of silver fir but underestimated the potential distribution. The BRT model was more sensitive to local site conditions in the present data, but the GAM showed more generality. The truncated response curves and high uncertainties of predictions at the edge of the site spectrum indicated a low data density and that the data did not cover the whole niche space of silver fir. As indicated by validation with expert knowledge, the model output approached potential distribution by optimizing true positive predictions. The classification of SDMs output into risk classes allowed model evaluation and interpretation. Predictions of GAM and BRT under the climate change scenario showed high accordance and therefore, low uncertainty. Finally, large areas of Bavaria are described to have a high risk of silver fir cultivation in future.Conclusions: SDMs are especially interesting as a decision basis for forest management because some of the general limitations of static modelling approaches are not relevant in this context. Limitations of forest inventory data can be partially overcome by using information on the potential distribution of species. The transferability of the models to future scenarios strongly depends on the spectrum and range of the training data sets and the depicted functional relationships. In order to improve the models and reduce the uncertainties, we need to improve the soil data and cover the whole niche space of silver fir.

Assessing future suitability of tree species under climate change by multiple methods: a case study in southern Germany

“…This approach has sometimes been criticized for neglecting more complex interacting factors such as edaphic or topographic characteristics, genetic differences, and anthropogenic interference (Mátyás et al 2009, Zimmermann et al 2010, Austin & Van Niel 2011, Bertrand et al 2012, Chakraborty et al 2015. On the other hand, recent progress in statistical modelling now allows a species-specific determination of key environmental factors (Elith et al 2008, Falk & Mellert 2011) and the analysis of major climatic constraints on large scales (Pearson & Dawson 2003). While SDMs are currently the most important tool to assist species selection in forest management, they are still the subject of ongoing research and improvement (Mellert et al 2015).…”

Generalized additive models as an alternative approach to the modelling of the tree height-diameter relationship

“…Even this method, however, failed in several cases to choose a model without overfitting. Falk and Mellert (2011) denoted this method to be very good while adding that monitoring the ecological understandability of the resulting model was still necessary.…”

“…They are recommended for use in forestry models e.g. by Moisen and Frescino (2002), Wang et al (2005), Zhang et al (2008), Falk and Mellert (2011) and Yue et al (2012).…”

“…In forestry, this modelling technique can be encountered in a range of applications over the past few years, e.g. in the papers of Falk and Mellert (2011) and Kouba and Alados (2012), who dealt with the modelling of spatial dynamics of tree species, Wang et al (2005), Albert and Schmidt (2010) and Aertsen et al (2012), who discussed the site-specific production capacity, Robinson et al (2011) and Kuželka and Marušák (2014), who modelled the stem form or Byun et al (2013), who modelled the diameter growth of trees or Zhang et al (2008) and Schmidt et al (2011), who modelled exactly height curves of various tree species, but used only different spline functions without using the LOESS function. Based on previous statements, three key advantages suggest to prefer a GAM instead of a classic height curve: (i) higher flexibility of the model, (ii) robustness against outliers, (iii) no usage limitations as in the case of OLS regression (normality or independence of residuals).…”

Managing Protected Areas in Central and Eastern Europe Under Climate Change

“…For silver fir (Abies alba Mill. ), currently a rare species in Central Europe, Falk and Mellert (2011) present a risk evaluation based on different SDM's. This approach has also been used for vulnerability assessment of habitat-specific species in Natura 2000 habitats (Harley 2011).…”