Modeling Diversity of Vascular Plants

Contents of this page

Species distribution/envelope models
Dynamic global vegetation models



Modeling approaches hve primarily been used to predict future climate chagne impacts on biodiversity and vegetation, in particular (Lovejoy & Hannah j2005).
In principle, two kinds of approaches were applied towards this end: (1) bioclimatic species distribution or species envelope models of the empirical, statistical type (Townsend Peterson et al. 2005) and (2) dynamic global vegetation models of the mechanistic, process-based ecosystem model type (Betts & Shugart 2005).
Both approaches have their strengths and weaknesses. Those will be adressed in the next paragraphs.


Species distribution/envelope models

These statistical models have commonly been developed within landscape monitoring programs (Heikkinen et al. 2006). They relate current species distribution or, in the case of species envelope models, current species presence/absence to environmental factors measured at the monitoring sites. Climate change factors predicted for the region are then used to compute future species distribution maps from the model (Townsend Peterson et al. 2005).
Since the 1990s a shift from multivariate statistical methods, predominantly Canonical Correspondence Analysis (CCA), towards modern data mining techniques, such as Classification and Regression Trees (CART), Multivariate Adaptive Regression Splines (MARS) and Random Forests, took place (Franklin 1995, Heikkinen et al. 2005). Data mining techniques were first introduced by Prasad & Iverson (2000) driven by the desire to automatize model choosing and model building, respectively. Even more methods, both parametric and non-parametric, were applied in comparative studies which sought to find the best species envelope model for the given data (Heikkinen et al. 2006).
Distribution models offer a unique tool for rapid analysis of the relationship between abundance and governing environmental factors for numerous individual species (Prasad & Iverson 1999-2009, Prasad & Iverson 2000, Heikinen et al. 2006). Although this represents the main advantage of the approach, the large variation of model predictions discovered in comparative studies reminds us of the many limitations and uncertainties the technique entails (Heikinen et al. 2006).
The models assume that current distributions are in equilibrium with the local climate (Heikkinen et al. 2006), to begin with. Certainlly this is questionable on the background of the rapid changes reported for many, if not all communities (compare Lovejoy & Hannah 2005 and the section Central European Plants)on this website).
In principle, the development of empirical models with high predictive accuracy relies heavily upon careful and proper selection of environmental variables (Heikkinen et. 2006). Given the limited knowledge about how the numerous variables interact with climate and affect real communities it is hardly conceivable that a satisfying variable selection will ever be reached. It may rather be preferrable to fit empirical models on virtual communities run from a process-based and, thus, transparent model with proven local validity.
However, many of the threat factors mentioned in section Central European Plants are neglected in this kind of models. Tthere are even climate change factors such as rising atmospheric CO2 level, which are key to the predictions made by dynamic ecosystem models (Betts & Shugart 2005), but have never been included in bioclimatic distribution models (Heikkinen et al. 2006).
Furthermore, the modeling scheme suffers from an overall lack in methodolgy to integrate species interaction (compare Jongman, ter Braak & van Tongeren 1995, Legendre & Legendre 1998, Leps & Smilauer 2003, Heikkinen et al. 2006). Although it is possible to incorporate effects of single species removal in multivariate data analysis (see Leps & Smilauer 2003, case study 4, pp. 214ff for an example), biomanipulation is indeed prohibited in monitoring programs. Unlike, in virtual communities biomanipulation is easy to conduct, although computatively expensive.
To end the list of disadvantages: Whether a species will reach and "finally" occupy the environmental tolerance region predicted by the model is dependent primarily on the extent to which the dispersal capabilitiy of that species is hindered by landscape fragmentation (Heikkinen et al. 2006). So far, this critical issue is entirely ignored by the modeling scheme.
Therefore, bioclimatic distribution/envelope models need to be applied only when users of models have a thorough understanding of their limitations and uncertainties (Heikkinen et al. 2006).


Dynamic global vegetation models

Since the pioneering work of the US-american ecologists Eugene P. Odum (Odum EP, 1995) and Howard T. Odum (Odum HT, 1994) in quantifying ecosystem matter and energy flows, ecosystem ecology and, with it, dynamic ecosystem models have developed into the mainstream of the ecological discipline.
The ecosystem paradigm became the cornerstone of the International Geosphere Biosphere Program (IGBP) and the climate change research conducted therein. The dynamic ecosystem models applied in the program have evolved into the dynamic global vegetation models (DGVMs) anongst others.
Dynamic global vegetation nmodels are process-based ecosystem models for the prediction of transient climate change impacts on the natural vegetation cover of large-scale grids (3 deg) around the world. In contrast to biome models they incorporate aspects of vegetation dynamics by allowing for competition among Plant Functional Types (PFTs) on the grid (Betts & Shugart 2005). PFTs, thus, are the elementary units of this model type. DGVMs vary in the number of Plant Functional Types, but commonly the PFTs used are more or less confined to vegetation types characterising a biome. PFTs in use are evergreen and deciduous forest, broad- and needleleaf forest and grassland of C3 or C4 photosynthetic pathway, to name a few. In principle, Plant Functional Types behave differently in competition due to their distinct process-related parameters.
In DGVMs plant phenological and physiological processes are consistently dependent on one or more climate change factors, such as temperature, precipitation, humidity, atmospheric CO2 level and sunlight. The processes are either simulated mechanistically or empirically in the models.
Main processes included in DGVMs are:
(1) Photosynthesis: Since photosynthesis is derived from lower scales, i.e. leef biochemistry, its description in the models is purely mechanistic.
(2) Autotrophic respiration: Respiration, in contrast, is modelled empirically.
(3) Stomatal behavior: The same is true for stomatal conductance.
(4) Transpiration: Transpiration is rather a mixture, since it is dependent on stomatal conductance, but relies on first physical principles otherwise.
(5) Allocation: Used allocation schemes, that is the fraction of the gained carbon/nitrogen translocated to foliage, wood or roots, are rather empirically derived.
(6) Phenology: In all models the annual vegetation cycle is enclosed by phenological events, namely budburst and leaf fall. The timing of these events, however, is based on PFT-specific measurements.
(7) Ecosystem processes: Litter, being a fraction of the standing biomass each year, is then made available to decomposition and nutrient cycling, which both are influenced by soil temperature and moisture, respectively.
In the end, these processes complete the ecosystem's carbon and nitrogen balance.
So far, six DGVMs have been developed: IBIS, SDGVM, TRIFFID, VECODE and HYBRID, LPJ (Cramer et al. 2001). While the first four models simulate competition among PFTs rather on the aggregate population level, HYBRID and LPJ have shifted somewhat in paradigm by modeling competition among individual trees.
Now, after having summarized the characteristics of this kind of model: What are typical predictive results DGVMs produce?
Cramer et al. (2001) investigated the response of the six existing dynamic global vegetation models to an IPCC warming scenario with rising CO2 concentration and climate change combined. The two main findings for the next century were (1) an expansion of deciduous forests in Siberia, North America and Europe despite underlying opposite precipitation trends and (2) a dieback of tropical forests and their conversion to C4 savanna due to more severe drought. Interestingly, no statement is made with respect to the tundra.
On the background of the last two sections you may wonder, as did Betts & Shugart (2005), whether the long-distance shifts in PFT ranges are related to biodiversity at all. This is particularly true in regions such as Europe, where natural vegetation was repelled to few, if any locations (Ellenberg & Strutt 2009). Basically, DGVMs with PFTs as their elementary unit are not able to get in touch with concerns about threatening of species. This is a critical issue, since we know from the past that species have responded to climate change in an individualistic manner with respect to direction and speed of migration (Hannah, Lovejoy & Schneider 2005, Overpeck, Cole & Bartlein 2005). Hence, although it is understandable that DGVM-developers have to restrict level of detail incorporated in the models, the situation is unfortunate.
In principle, the following issues shall be blamed responsible for this drawback:
The top-down approach inherent to ecosystem models as well as the reliance on huge systems of differential equations bear a fundamental burden to face complexity. Likewise, complexity is hard to master with procedural programming languages these models are written in.



Fig. 5: Graphical Summary Bjiodiversity Modeling

As the graphical summary of Fiture 5 illustrates current biodiversity modeling concepts contribute little to our understanding of plant diversity loss.
Both kinds of models do not simulate how nutrient load interacts with management intensification and climate change factors. They rather focus on climate change impacts alone.
Due to their many disadvantages it is questionable whether distribution/envelope models can be extended to simulate this kind of interaction when applied to real communities.
The spatial scale and the focus on natural vegetation hinder DGVMs to become biodiversity models. Besides that, dynamic ecosystem models per se come with the burden of a philosophical and technical background that prevents them to break the "PFT threshold" and reach the species level.
From the foregoing I deduce an urgent need to grow tte next generation biodiversity models from roots other than ecosystem models. This is further detailed in the next sections.


You are here: Modeling Biodiverity -> Background -> Home

Local Navigator: Λ Top <- Previous -> Next

Sub-Page Navigator: Biodiversity <- Central-Eurpean Plants <- Modeling Biodiversity -> Future Biodiversity Models -> Online Databases -> Universal Scaling Laws

Whole-Page Navigator: Home <- Background -> UIBM Development -> Virtual Experiments -> Outlook: Virtual Ecology

Copyright Dec. 2009 Dr. U. Grueters

(C) 2006 - All rights reserved

Print this page