Robust Modelling of the Impacts of Climate Change on the Habitat Suitability of Forest Tree Species. de Rigo, D.; Caudullo, G.; San-Miguel-Ayanz, J.; and Barredo, J. I. Publications Office of the European Union.
Robust Modelling of the Impacts of Climate Change on the Habitat Suitability of Forest Tree Species [link]Paper  doi  abstract   bibtex   
[::] In Europe, forests play a strategic multifunctional role, serving economic, social and environmental purposes. However, their complex interaction with climate change is not yet well understood. [::] The JRC PESETA project series proposes a consistent multi-sectoral assessment of the impacts of climate change in Europe. [::] Within the PESETA II project, a robust methodology is introduced for modelling the habitat suitability of forest tree species (2071-2100 time horizon). [::] Abies alba (the silver fir) is selected as case study: a main European tree species often distributed in bioclimatically complex areas, spanning over various forest types and with multiple populations adapted to different conditions. [::] The modular modelling architecture is based on relative distance similarity (RDS) estimates which link field observations with bioclimatic patterns. Robust management of uncertainty is also discussed. [Abstract] In Europe, forests play a strategic multifunctional role, serving economic, social and environmental purposes. However, forests are among the most complex systems and their interaction with the ongoing climate change - and the multifaceted chain of potential cascading consequences for European biodiversity, environment, society and economy - is not yet well understood. [\n] The JRC PESETA project series proposes a consistent multi-sectoral assessment of the impacts of climate change in Europe. Within the PESETA II project, a robust methodology is introduced for modelling the habitat suitability of forest tree species (2071-2100 time horizon). Abies alba (the silver fir) is selected as a case study: a main European tree species often distributed in bioclimatically complex areas, spanning over various forest types and with multiple populations adapted to different conditions. [\n] The modular modelling architecture is based on relative distance similarity (RDS) estimates which link field observations with bioclimatic patterns, projecting their change under climate scenarios into the expected potential change of suitable habitat for tree species. Robust management of uncertainty is also examined. Both technical and interpretation core aspects are presented in an integrated overview. The semantics of the array of quantities under focus and the uneven sources of uncertainty at the continental scale are discussed (following the semantic array programming paradigm), with an effort to offer some minimal guidance on terminology, meaning and methodological limitations not only of the proposed approach, but also of the broad available literature - whose heterogeneity and partial ambiguity might potentially reverberate at the science-policy interface. [Excerpt: Robust modelling of tree species habitat suitability] How resilient are the European forests? How robust is our understanding of the potential impacts of the changing climate on the future forests in the European continent? Recalling the aforementioned inherent complexity of forest systems, and the significant share of forested areas in Europe, an articulated response may be expected. [\n][\n] Forests in Europe span over a variety of types and even a basic broad characterisation requires multiple, quite different ecological domains to be considered, from the subtropical domain in the Mediterranean areas to the temperate and boreal domains in the northern areas of the continent. [\n] The literature on tree species distribution is huge. Unfortunately, terminology, meaning and methodological limitations of reported findings might be ambiguous - an ambiguity which could potentially reverberate at the science-policy interface. A confusion exists between distinct concepts such as the distribution range of a certain tree species, its frequency or probability of presence, its habitat suitability (and which kind of statistical operator is exploited to compute this suitability), the realised niche of the species versus its potential niche. Furthermore, available pan-European forest field data are collected with challenging harmonisation efforts from multiple regional (e.g. country-level) sources. Regional datasets are typically collected and organised independently, with uneven spatial density of sampling and uncertainty; and sometimes following nonhomogeneous definitions of the collected information. Therefore, semantic, modelling and data uncertainties characterise tree species distribution and suitability modelling at the European scale, requiring robust modelling strategies to mitigate their combined uncertainty. [\n][\n] This work mainly deals with the habitat suitability (HS) of a tree species, frequently linked to the bioclimatic conditions characterising the habitats under which the species is suitable to thrive. Here, a distinction should be made between: [::] the average HS - derived by considering the frequency of observed occurrences for a given bioclimatic pattern as a proxy for the corresponding fitness of the species - and [::] the maximum HS (MHS or survivability) - which equally considers also less frequent occurrences, so as to detect where the species can survive irrespective of whether it would be dominant or secondary within a certain bioclimatic habitat. [\n] The first definition is the typical one implicitly considered by most HS applications, even because computing it is faster (if not the only possible option) with several available tools. The latter definition is the one on which this work focuses. Since the maximum HS is also based on available field observations (which are by definition limited to the realised niche, as altered by the anthropic influence), the estimated maximum HS of a species should not be confused with its potential niche - which may be a challengingly elusive concept from a data-driven perspective. Nevertheless, maximum HS might support the assessment of the areas whose bioclimatic conditions may allow a given species to survive (including conditions observed less frequently, thus more robust to the typically uneven sampling of the data available at regional or wider scale); and the potential spatial shift of these areas under changing climate scenarios. [\n][\n] HS is sometimes exploited as proxy information for crudely approximating the current distribution range (if not even the probability of presence) of the species. Extrapolating this approximation to future climate change scenarios might be tempting. However, there are several reasons why some geographic areas might be bioclimatically suitable for a species even if the species is not observed there. Among them, three may be mentioned, because of their policy relevance: [::ecological competition:] although the species may be bioclimatically suitable if considered in isolation, under natural conditions other taxa might prevail because of their higher fitness in taking advantage of the local bioclimatic conditions. In this case, the species may simply be unable to survive the competition with the other taxa, or it may instead be severely limited in its potential dominance, thus locally resulting as a secondary or rare species. Under these circumstances, plantations or managed forest stands (where natural competition is artificially limited) may enable the species to exploit its full habitat suitability. [...] [::Dispersal limitation,] distance from the borders of the current distribution: an otherwise suitable area may not yet have been colonised by the species. The changing bioclimatic patterns under climate change may amplify the impact of dispersal limitation, so as for areas expected to become suitable bioclimatically to remain not colonised by the species under focus. [...] [::Anthropogenic elimination:] for example, eradication of the species as a consequence of systematic anthropic interventions to favour other taxa which may be locally more convenient (e.g. from an economy perspective). [...] [\n] These three phenomena have in common a straightforward mechanism with which the habitat suitability influences the species distribution. The maximum extent of the HS acts as a logical constraint for the actual species distribution, so that the latter is strictly included within the maximum HS. Therefore, while a future expansion of the maximum HS might not automatically imply an expanding distribution range, a future MHS contraction would likely affect the distribution range by imposing a contraction to it. [\n][\n] Discussing on robustness of habitat suitability modelling under climate change scenarios, another key source of uncertainty deserves to be mentioned. Projections on potential future climate scenarios predict broad areas of Europe to possibly shift towards geoclimatic patterns which are far from any currently observed pattern in Europe. This wide shift introduces an intrinsic source of modelling uncertainty due to climate-driven extrapolation. Elementary models which are based on relatively small subsets of the information on the climate signal may be able to limit the impact of the forced extrapolation, although at the price of a more simplistic description of the biophysical conditions (higher modelling uncertainty). Unfortunately, this might be a reassuring overoptimistic consequence of the low-dimensionality of the simplified climate signal considered. [\n] The approach proposes in this study is robust even in exploiting a rich set of bioclimatic predictors (not to oversimplify the climate signal) while transparently highlighting the extent of extrapolation, which is intrinsically computed by the underpinning mathematical methods. The proposed modelling architecture to estimate maximum HS is based on the relative distance similarity (RDS) approach, which estimates a dimensionless score of how similar/dissimilar the bioclimatic patterns of a tested area are compared with the available species observations. [\n] [...]
@book{derigoRobustModellingImpacts2017,
  title = {Robust Modelling of the Impacts of Climate Change on the Habitat Suitability of Forest Tree Species},
  author = {de Rigo, Daniele and Caudullo, Giovanni and San-Miguel-Ayanz, Jesús and Barredo, José I.},
  date = {2017-03},
  publisher = {{Publications Office of the European Union}},
  location = {{Luxembourg}},
  doi = {10.2760/296501},
  url = {https://doi.org/10.2760/296501},
  abstract = {[::] In Europe, forests play a strategic multifunctional role, serving economic, social and environmental purposes. However, their complex interaction with climate change is not yet well understood.

[::] The JRC PESETA project series proposes a consistent multi-sectoral assessment of the impacts of climate change in Europe.

[::] Within the PESETA II project, a robust methodology is introduced for modelling the habitat suitability of forest tree species (2071-2100 time horizon).

[::] Abies alba (the silver fir) is selected as case study: a main European tree species often distributed in bioclimatically complex areas, spanning over various forest types and with multiple populations adapted to different conditions.

[::] The modular modelling architecture is based on relative distance similarity (RDS) estimates which link field observations with bioclimatic patterns. Robust management of uncertainty is also discussed.

[Abstract]

In Europe, forests play a strategic multifunctional role, serving economic, social and environmental purposes. However, forests are among the most complex systems and their interaction with the ongoing climate change - and the multifaceted chain of potential cascading consequences for European biodiversity, environment, society and economy - is not yet well understood.

[\textbackslash n] The JRC PESETA project series proposes a consistent multi-sectoral assessment of the impacts of climate change in Europe. Within the PESETA II project, a robust methodology is introduced for modelling the habitat suitability of forest tree species (2071-2100 time horizon). Abies alba (the silver fir) is selected as a case study: a main European tree species often distributed in bioclimatically complex areas, spanning over various forest types and with multiple populations adapted to different conditions.

[\textbackslash n] The modular modelling architecture is based on relative distance similarity (RDS) estimates which link field observations with bioclimatic patterns, projecting their change under climate scenarios into the expected potential change of suitable habitat for tree species. Robust management of uncertainty is also examined. Both technical and interpretation core aspects are presented in an integrated overview. The semantics of the array of quantities under focus and the uneven sources of uncertainty at the continental scale are discussed (following the semantic array programming paradigm), with an effort to offer some minimal guidance on terminology, meaning and methodological limitations not only of the proposed approach, but also of the broad available literature - whose heterogeneity and partial ambiguity might potentially reverberate at the science-policy interface.

[Excerpt: Robust modelling of tree species habitat suitability]

How resilient are the European forests? How robust is our understanding of the potential impacts of the changing climate on the future forests in the European continent? Recalling the aforementioned inherent complexity of forest systems, and the significant share of forested areas in Europe, an articulated response may be expected.

[\textbackslash n][\textbackslash n] Forests in Europe span over a variety of types and even a basic broad characterisation requires multiple, quite different ecological domains to be considered, from the subtropical domain in the Mediterranean areas to the temperate and boreal domains in the northern areas of the continent.

[\textbackslash n] The literature on tree species distribution is huge. Unfortunately, terminology, meaning and methodological limitations of reported findings might be ambiguous - an ambiguity which could potentially reverberate at the science-policy interface. A confusion exists between distinct concepts such as the distribution range of a certain tree species, its frequency or probability of presence, its habitat suitability (and which kind of statistical operator is exploited to compute this suitability), the realised niche of the species versus its potential niche. Furthermore, available pan-European forest field data are collected with challenging harmonisation efforts from multiple regional (e.g. country-level) sources. Regional datasets are typically collected and organised independently, with uneven spatial density of sampling and uncertainty; and sometimes following nonhomogeneous definitions of the collected information. Therefore, semantic, modelling and data uncertainties characterise tree species distribution and suitability modelling at the European scale, requiring robust modelling strategies to mitigate their combined uncertainty.

[\textbackslash n][\textbackslash n] This work mainly deals with the habitat suitability (HS) of a tree species, frequently linked to the bioclimatic conditions characterising the habitats under which the species is suitable to thrive. Here, a distinction should be made between:

[::] the average HS - derived by considering the frequency of observed occurrences for a given bioclimatic pattern as a proxy for the corresponding fitness of the species - and

[::] the maximum HS (MHS or survivability) - which equally considers also less frequent occurrences, so as to detect where the species can survive irrespective of whether it would be dominant or secondary within a certain bioclimatic habitat.

[\textbackslash n] The first definition is the typical one implicitly considered by most HS applications, even because computing it is faster (if not the only possible option) with several available tools. The latter definition is the one on which this work focuses. Since the maximum HS is also based on available field observations (which are by definition limited to the realised niche, as altered by the anthropic influence), the estimated maximum HS of a species should not be confused with its potential niche - which may be a challengingly elusive concept from a data-driven perspective. Nevertheless, maximum HS might support the assessment of the areas whose bioclimatic conditions may allow a given species to survive (including conditions observed less frequently, thus more robust to the typically uneven sampling of the data available at regional or wider scale); and the potential spatial shift of these areas under changing climate scenarios.

[\textbackslash n][\textbackslash n] HS is sometimes exploited as proxy information for crudely approximating the current distribution range (if not even the probability of presence) of the species. Extrapolating this approximation to future climate change scenarios might be tempting. However, there are several reasons why some geographic areas might be bioclimatically suitable for a species even if the species is not observed there. Among them, three may be mentioned, because of their policy relevance:

[::ecological competition:] although the species may be bioclimatically suitable if considered in isolation, under natural conditions other taxa might prevail because of their higher fitness in taking advantage of the local bioclimatic conditions. In this case, the species may simply be unable to survive the competition with the other taxa, or it may instead be severely limited in its potential dominance, thus locally resulting as a secondary or rare species. Under these circumstances, plantations or managed forest stands (where natural competition is artificially limited) may enable the species to exploit its full habitat suitability. [...]

[::Dispersal limitation,] distance from the borders of the current distribution: an otherwise suitable area may not yet have been colonised by the species. The changing bioclimatic patterns under climate change may amplify the impact of dispersal limitation, so as for areas expected to become suitable bioclimatically to remain not colonised by the species under focus. [...]

[::Anthropogenic elimination:] for example, eradication of the species as a consequence of systematic anthropic interventions to favour other taxa which may be locally more convenient (e.g. from an economy perspective). [...]

[\textbackslash n] These three phenomena have in common a straightforward mechanism with which the habitat suitability influences the species distribution. The maximum extent of the HS acts as a logical constraint for the actual species distribution, so that the latter is strictly included within the maximum HS. Therefore, while a future expansion of the maximum HS might not automatically imply an expanding distribution range, a future MHS contraction would likely affect the distribution range by imposing a contraction to it.

[\textbackslash n][\textbackslash n] Discussing on robustness of habitat suitability modelling under climate change scenarios, another key source of uncertainty deserves to be mentioned. Projections on potential future climate scenarios predict broad areas of Europe to possibly shift towards geoclimatic patterns which are far from any currently observed pattern in Europe. This wide shift introduces an intrinsic source of modelling uncertainty due to climate-driven extrapolation. Elementary models which are based on relatively small subsets of the information on the climate signal may be able to limit the impact of the forced extrapolation, although at the price of a more simplistic description of the biophysical conditions (higher modelling uncertainty). Unfortunately, this might be a reassuring overoptimistic consequence of the low-dimensionality of the simplified climate signal considered.

[\textbackslash n] The approach proposes in this study is robust even in exploiting a rich set of bioclimatic predictors (not to oversimplify the climate signal) while transparently highlighting the extent of extrapolation, which is intrinsically computed by the underpinning mathematical methods. The proposed modelling architecture to estimate maximum HS is based on the relative distance similarity (RDS) approach, which estimates a dimensionless score of how similar/dissimilar the bioclimatic patterns of a tested area are compared with the available species observations.

[\textbackslash n] [...]},
  isbn = {978-92-79-66704-6},
  keywords = {*imported-from-citeulike-INRMM,~INRMM-MiD:c-14314400,~to-add-doi-URL,abies-alba,array-of-factors,artificial-neural-networks,bioclimatic-predictors,change-factor,climate-change,data-uncertainty,diversity,environmental-modelling,europe,extrapolation-uncertainty,featured-publication,forest-resources,free-scientific-knowledge,free-scientific-software,free-software,fuzzy,gdal,genetic-diversity,geospatial,geospatial-semantic-array-programming,gnu-bash,gnu-linux,gnu-octave,habitat-suitability,integration-techniques,mastrave-modelling-library,maximum-habitat-suitability,modelling-uncertainty,multiplicity,peseta-series,python,regional-climate-models,relative-distance-similarity,robust-modelling,semantic-array-programming,semantic-constraints,semantics,spatial-disaggregation,sres-a1b,supervised-training,unsupervised-training},
  options = {useprefix=true},
  pagetotal = {58 pp.}
}
Downloads: 0