Towards a more transparent use of the potential natural vegetation concept – an answer to Chiarucci et al.
In this paper, the concerns of Chiarucci et al. (2010) regarding use of the potential natural vegetation (PNV) concept are addressed, as voiced in the forum section of the Journal of Vegetation Science. First, we rectify some unfounded expectations concerning PNV, including a relationship with prehuman vegetation and phytosociology. Second, we point out issues that pose considerable challenges in PNV and require common agreement. Here, we address the issue of time and disturbance. We propose to use the static PNV concept as a baseline, a null model for landscape assessment and in comparisons. Instead of changing the PNV concept itself, we introduce a new term, potential future natural vegetation (PFV) to cover estimations of potential successional outcomes. Finally, we offer a new view of PNV with which we intend to make the use of PNV estimates more transparent. We formalize the PNV theory into a partial cause-effect model of vegetation that clearly states which effects on vegetation are factored out during its estimation. Further, we also propose to assess PNV in a probabilistic setting, rather than providing a single estimate for one location. This multiple PNV would reflect our uncertainty about the vegetation entity that could persist at the locality concerned. Such uncertainty arises from the overlap of environmental preferences of different mature vegetation types. Thus reformulated, we argue that the PNV concept has much to offer as a null model, especially in landscape ecology and in site comparisons in space and time.
Potential Natural Vegetation
Chiarucci et al. (2010) have recently questioned the validity and usefulness of the potential natural vegetation (PNV) concept in this journal forum. They took up a debate on the subject that started in the Journal of Biogeography. The starting papers in these forums (Carrión & Fernandez 2009; Chiarucci et al. 2010) raised fundamental concerns about pitfalls in the PNV concept. Here, we argue that dismissal of PNV in vegetation ecology and related scientific fields is not justified. Our aims are to: (1) eliminate unfounded expectations concerning PNV; (2) point out issues that are problematic and need common agreement in the field; and (3) offer a new view of PNV that may help to make the use of PNV estimates more transparent, i.e. clarify the process of estimation. We do not intend to repeat the original definition (Tüxen 1956) or the later specifications of the concept (Kowarik 1987; Härdtle 1995; Moravec 1998), since many preceding forum papers have done this, and the issue can also be traced back in the literature. To keep our arguments focussed, we also refrain from reiterating earlier discussions (among others, Leuschner 1997; Zerbe 1998; Mucina 2010) in so far as they are not directly related to the critique of Chiarucci et al. (2010).
Unfounded Expectations Concerning PNV
First, we emphasize that PNV differs from the last prehuman vegetation, and disregarding this has led to unfounded criticism (Carrión & Fernandez 2009). Although Chiarucci et al. (2010) do not explicitly endorse this common misunderstanding, they do refer to palaeoecological reconstructions that seem to reveal inconsistencies between the expected vegetation before and after human occupation. Thus, they suggest that palaeoecological reconstructions are relevant for the assessment of PNV. While this may hold under certain specific circumstances, deviations are just as common (Moravec 1998). A difference between the prehuman vegetation and PNV can be due to human transformation or natural changes in the physical environment, such as climate change. Due to these effects, environmental conditions have likely changed since the first human occupation and this has had an effect on natural vegetation that can be expected at a specific location.
One of the major criticisms of Chiarucci et al. (2010) of PNV is that it ‘can hardly be defined because of methodological problems.’ We believe that methodological problems should not lead to dismissal of a concept. If a researcher finds the currently applied methodology inappropriate, a new methodology can always be developed (as Farris et al. (2010) also argued). It is also not justified to say that PNV has been coupled exclusively to phytosociology. Although phytosociologists conducted most of the European PNV estimations, the PNV concept has been used outside Europe, where different estimation methods were used. Thus, early PNV estimates in the USA, for example, relied on knowledge of succession and the climax concept (Küchler 1964; Rowe 1977; Rzedowski 1990). More recently, several modelling approaches have been used throughout the world (Brzeziecki et al. 1993; Tichy 1999; Kelly et al. 2006; Liu et al. 2009; Zampieri & Lionello 2010). This development shows that where relevés and syntaxa are considered inadequate, other data sources can be utilized and target vegetation can be defined differently.
Challenges in Grasping PNV
Other major criticisms of Chiarucci et al. (2010) relate to ecosystem dynamics. The PNV definition applied so far by ecologists is deliberately static. PNV in this static form provides a baseline for assessing the vegetation potential of an area (Crumpacker et al. 1988; Grabherr et al. 1997; Rosati et al. 2008; Aguilar et al. 2010), and this is why earlier maps are still being reproduced in improved/digital formats (Küchler 1993; Vaca et al. 2011). Even when we acknowledge uncertainty in its estimation, PNV is still the best available null model which allows us to grasp the vegetation that could be present in a landscape (Ricotta et al. 2002). It is important to understand how much of this potential is expressed in the actual vegetation, especially if human-influenced landscapes are compared, whether in space or in time (Crumpacker et al. 1988; Carranza et al. 2003; Aguilar et al. 2010; Hickler et al. 2012; Wang et al. 2011). PNV also serves as a null model for vegetation management, including forestry (cf. site maps of Ellenberg 1967 used in forestry; Barnes et al. 1982; Gégout & Houllier 1996; Kelly et al. 2006), nature conservation (Crumpacker et al. 1988; Rosati et al. 2008; Aguilar et al. 2010) and modelling of climate change effects (Thonicke & Cramer 2006; Hickler et al. 2012; Wang et al. 2011).
One aspect of Chiarucci et al.'s criticism relating to ecosystem dynamics is the lack of temporal considerations in PNV estimations. We agree with Chiarucci et al. (2010) that PNV in its static form should not be used to assess successional outcomes; we advocate its use as a null model. Probably the reformulation of the original ‘coming into existence at once’ criterion (Tüxen 1956), which has raised so much debate, would help to make PNV more acceptable. In our opinion, it can be viewed as the vegetation that ‘would persist under the current conditions if it was already there.’ Importantly, this expression is consistent with the original concept and leads to the same estimated PNV as the version of Tüxen. Furthermore, this formulation lends justification to the use of actual natural vegetation (established stands that persist without human intervention) in estimating PNV, which has been typical practice in expert-based as well as in formalized methods. Consequently, sites where the actually observed vegetation is not likely to persist should be excluded from the estimation of PNV. It is in this respect that vegetation history records can assist in decision-making as to whether or not to include certain states of vegetation when calibrating PNV models.
Although we propose retaining PNV as a static baseline, we agree with Chiarucci et al. (2010) that providing an estimate of potential vegetation dynamics and forecasting its outcome at a site is a fascinating research topic and that advances in this direction are most welcome. Indeed, examples of such research already exist: Hickler et al. (2012) and Wang et al. (2011) simulated future distribution of PNV zones in Europe and China, respectively, at a coarse scale. However, these estimations are beyond PNV and deserve a separate name and definition, e.g. potential site dynamics or potential future vegetation states, as Hickler et al. (2012) also suggested. We propose to differentiate this potential future natural vegetation from PNV by using the acronym ‘PFV’ for the former. However, it should also be kept in mind that the inclusion of time remains a great challenge, since over time, not only vegetation succession proceeds but also environmental conditions change, thus altering succession routes. Furthermore, climate change scenarios involve a high level of uncertainty and variability (Meehl et al. 2007), not to mention future land-use change predictions (e.g. Rounsevell et al. 2006). Thus, instead of expecting actual PNV to express all aspects of possible vegetation development, in our opinion it should be preserved as a null model and be viewed as a baseline, analogous to Odum's (1971) approach of treating the idea of climax as ‘a basic yardstick available for comparison.’
Chiarucci et al. (2010) also dedicate much space to other aspects of ecosystem dynamics, namely disturbances and human influences (large mammals, management, invasions), which they think PNV should cover. Until now, PNV has been defined and used as a null expectation, which excludes any current human impact on vegetation. In contrast, the concept of potential replacement vegetation (PRV) has been introduced to cover the disturbed vegetation corresponding to a PNV unit (Chytrý 1998). PNV estimation in the strict sense considers all natural disturbances (e.g. fires lit by lightning, extant large mammals), but not human-induced disturbances. Notwithstanding, irreversible anthropogenic changes to the environment (as a result of past human interference) are taken into account when estimating PNV (for detailed reasoning see Härdtle 1995; Moravec 1998). This is quite straightforward for past transformations of geomorphology and the lowering of water tables due to human intervention. Similarly, the extinction of large mammals is a lasting effect of human presence that should be accounted for in estimating current PNV. On the other hand, plant invasions clearly reflect human influence on vegetation; thus invaded vegetation will most often belong to PRV as defined in Chytrý (1998). The question of invasions is inevitably linked to how we treat dispersal barriers. Dispersal limitation has so far only been considered implicitly in PNV estimates; we therefore propose to make this decision transparent, too. Based on existing experience and examples, we conclude that PNV should not be constrained by within-region dispersal limitation, i.e. it should be assumed that all species in the regional species pool are actually able to reach all sites of a region. This is also practical, since in heavily transformed landscapes much of the woody vegetation would be excluded from PNV if within-region dispersal constraints were taken into account. On the other hand, coarse-scale biogeographic limitations must be taken into account, e.g. regardless of site conditions, it does not make sense to consider Nothofagus forests as PNV in the Northern hemisphere. Disregarding cross-region dispersal limitation presents no advantage for PNV as a null model, which serves for comparison with existing, future or past vegetation.
As ongoing debates indicate that no general agreement on these challenges has yet been reached among ecologists, for the sake of transparency, publications on PNV should make assumptions about disturbances and invasions explicit.
Our Proposal to Increase the Transparency of PNV Estimations
Much of the critique of PNV comes from the use of phytosociology and expert knowledge for its estimation (Carrión & Fernandez 2009; Chiarucci et al. 2010). We believe that both of these serve the purposes of PNV well in certain situations, but we also think that the concept could benefit from making the process of estimation more transparent. To achieve this, we propose to formalize the original PNV concept in terms of a partial cause-and-effect model of vegetation, where the reversible anthropogenic conditions (co-variables) of its existence are factored out, while the natural site factors remain as predictors:
V = f (physiographic site factors: climate, bedrock, relief, etc.; dispersal limitation; human impact; natural disturbance; time).
PNV = f (physiographic site factors; biogeographic dispersal limitation; natural disturbance) | (time; local dispersal limitation; human impact on vegetation).
where, V denotes actual vegetation and | denotes partialling out, i.e. the removal of co-variable effects from the model (‘|’ borrowed from variance partitioning terminology; Legendre & Legendre 1998). This notation makes explicit which factors are used in vegetation modelling and which are disregarded in the construction of PNV. Should a researcher wish to define PNV differently, the equation lends itself to making such deviations transparent in any publication. We propose to focus on physiographic site factors as abiotic determinants of PNV, because this excludes those factors that depend on vegetation itself at fine scales (e.g. stand microclimate or humus form). This distinction comes from population ecology and niche modelling, where Hutchinson (1978) distinguished ‘scenopoetic’ environmental variables as those variables setting the abiotic scene for biotic interactions (Soberón 2010). Availability of new sources of information (digital elevation models, geological and soil maps, climate maps), geographic information systems and new modelling methods, such as predictive vegetation models (PVM; Franklin 1995; Ferrier & Guisan 2006), may provide PNV with a new foundation (as also suggested in Mucina 2010). Notwithstanding, even expert estimations, which may sometimes be more applicable for local decisions than mathematical models, can benefit from such a formalization.
In fact, there are already numerous examples of using PVM to estimate PNV (Brzeziecki et al. 1993; Tichy 1999; Kelly et al. 2006; Liu et al. 2009), even if many authors do not refer to the term PNV (Fischer 1990; some examples in Franklin 1995; Keith & Bedward 1999; Cawsey et al. 2002; Accad & Neil 2006). For example, much effort has been expended to model the total vegetation cover in parts of Australia (Keith & Bedward 1999; Cawsey et al. 2002; Accad & Neil 2006). Although these authors stated that they were modelling pre-European vegetation, they used actual undisturbed vegetation samples rather than historic sources for calibration – thus estimating PNV. We should note, however, that environmental changes since the establishment of Western civilization has a short history in Australia, thus it might be true that current PNV coincides with pre-European (though not prehuman, cf. aboriginals: McIntosh et al. 2009) vegetation.
However, many modern PNV models still impart false confidence in PNV estimations by giving a single outcome for one point in space (among others, Neuhäuslova & Moravec 1997; Mikyška 1968; Fischer 1990; Brzeziecki et al. 1993; Tichy 1999; Liu et al. 2009). This leads to criticism about determinism in PNV (Carrión 2010). Variations are found in environmental conditions under the same vegetation, partly due to stochasticity, and partly because conditions favouring two different vegetation units might overlap. Therefore, we propose assessing PNV in a probabilistic rather than a deterministic way, and allowing a range of mature (climax) vegetation types to constitute the PNV of a single point/site in space. Such a multiple probabilistic PNV estimation easily arises from PVMs constructed for individual vegetation types. It is important to stress, however, that a multiple PNV is different from supplying the range of vegetation types belonging to one successional sere leading to the same climax, and also from the polyclimax concept of Tansley (1935). The latter, as opposed to the monoclimax of Clements (1936), assumes several climax communities in the same climatic region that differ in local, mostly pedological, site conditions and therefore in the expected climax. Nevertheless, under one set of environmental conditions it assumes a single outcome, which differs from our approach.
Despite its greater complexity, multiple PNV may be attractive to nature managers in situations where a balanced evaluation of existing vegetation in terms of naturalness and diversity is required. Existing patches of rare vegetation types may not belong to the most probable types at each site, but will likely appear as part of a multiple PNV.
The PNV concept has much to offer as a null model and reference for comparisons drawn by researchers, landscape ecologists and managers. However, a clear assessment of its limits is necessary. PNV is different from prehuman vegetation and, as a baseline of current site conditions, cannot account for time. Advances in ecological modelling and the paradigm shift towards stochastic concepts call for a reconsideration of the deterministic nature of PNV. Besides retaining its static form, we propose a multiple probabilistic assessment of PNV, which offers deeper insights into the vegetation potential, both for comparative and management purposes. Further developments starting from the PNV concept will also be beneficial. These could extend the idea towards potential dynamics and potential future natural vegetation (PFV) through inclusion of disturbance and temporal aspects. However, these do not negate the usefulness of the null model type of static PNV and should not be confounded with that concept.
Imelda Somodi was supported by the Hungarian Scientific Research Fund (OTKA; grant no. PD-83522). We are also grateful to three anonymous referees and Jake Overton for useful comments on an earlier version of the manuscript.