The approach used by Alton et al 2011 is appropriate for the scope and intent of this paper - to demonstrate a weakness in the approach - but is not a comprehensive analysis of PFTs in models.
I think it is useful to question the use of PFTs but incorrect to conclude that Alton et al claim PFTs are not useful. Alton et al show that albedo, net shortwave radiation, and runoff are insensitive to PFT-specific values of four parameters, but they also find that net carbon balance is sensitive.
The study's conclusions are limited to the JULES model that is evaluated and the parameters that are allowed to vary (and thus be estimated by inversion). There are many different models of terrestrial ecosystem functioning, and many approaches to model parameterization. Furthermore, JULES has more than 50 parameters for each PFT, and only four were varied in this study. (This is the number of parameters set in JULES' fortran namelist under
This study also provides examples of how inversion using what are essentially 'free' parameters (analogous to 'flat priors' in a Bayesian context) has limitations. Fig 3 shows this, since the 'retrieved' (inverse-estimated) parameters are inconsistent with direct measurements of these parameters. This is more of a limitation of the inversion approach for estimating PFT-level parameters than of the utility of PFTs as an approximation of plant physiological diversity.
To answer your question - yes, there are alternative approaches to both model structure and parameterization. This is an active area of research (and one I am interested in). Here are some examples, in order of increasing complexity (and data and computational requirements).
- Global land surface models like JULES are necessarily more abstract (and use PFTs to represent a 'Biome').
- Represent groups of individuals that belong to a particular PFT and age class (This was first done in the Ecosystem Demography model (Moorecroft et al 2001, which has both evolved to ED2 and has been integrated into NCAR's Community Land Model.
- Vary parameters dynamically in both space and time. For example, Wang et al 2012 show that varying Vcmax in (vertical) space and time improve estimates of gross primary productivity in a very simple crop model.
- Individual Based Models (IBM's) represent individuals explicitly. These are reviewed by DeAngelis and Mooij 2005.
- Using probability distributions to reflect parameter variability in the system being studied. I am not aware that this has been attempted.
However, adding more degrees of freedom limits inference - too many unconstrained parameters make it difficult to parameterize the model, understand its output, or make general conclusions about how the world operates.
Remember "All models are wrong, but some are useful" - George Box