Page 153 - Climate Change and Food Systems

chapter 4: an overview of climate change impact on crop production and its variability in europe, related uncertainties and research challenges
In addition to quantifying uncertainties in climate projections and downscaling (see above, and Sections 1.2 and 2.2; Rötter et al., 2012a), several other sources of uncertainty must be acknowledged, including discrepancies in simulating climate impacts by different impact models. Such impact model-related uncertainty has been shown to constitute a considerable share of overall uncertainty in projections of
CC impacts on crop production (Palosuo et al., 2011; Rötter et al., 2012a; Asseng et al., 2013; Eitzinger et al., 2013). Asseng et al. (2013) even found that crop model uncertainty exceeded climate model uncertainty for the climate scenarios considered in that study. Earlier studies (e.g. Mearns et al., 1999) also highlighted the relatively high importance of crop model uncertainty. On
the other hand, other studies have found that uncertainties in climate projections have even greater importance (e.g. Iizumi et al., 2011).
There are basically three ways of evaluating uncertainty in models, including crop models:
1. The first approach – the traditional one – is through comparison of simulated and observed values, with the assumption that past errors are representative of
the uncertainty in future simulations;
this commonly done for crop models (Wallach et al., 2013). A major difficulty with this approach in the context of IAM of future climate effects is that past errors may not be representative of future errors.
2. A second approach is to evaluate the contribution of specific sources of error
to model uncertainty (Walker et al., 2003; Palosuo et al., 2011). The major effort in this respect has been to evaluate the effects of uncertainty in model parameter values on uncertainty in predictions, through sensitivity analysis or by using a Bayesian approach (deductive reasoning based on probabilistic outcomes) (for example, Wallach et al., 2012). While this approach makes it possible to evaluate errors for new conditions, it does
not take into account uncertainties in model formulation.
3. The third approach is through the use of ensembles of models (Rötter et al., 2011; Asseng et al., 2013). Renewed efforts have been made recently to use crop model intercomparisons to reveal uncertainties (e.g. in COST action 7347, AgMIP and MACSUR projects). Discrepancies between models
can be assumed to represent uncertainty in both model formulation and parameterization, specific to each context simulated. This approach is easily extended to cascades of models (for example, climate models feeding climate projections into crop models, and
the latter feeding relative yield changes into economic models), and therefore is particularly well suited to IAM. For example, Asseng et al. (2013) used ensembles of both GCMs and crop models to evaluate the uncertainty in future yields, and to apportion the overall uncertainty to contributions from uncertainty in GCMs and uncertainty in crop models. Recent studies by Tao et al. (2009) and Iizumi et al. (2011) combined an ensemble of GCMs with a Bayesian approach to model parameterization, in order to obtain multiple climate scenarios and model parameter combinations. This
was done in an attempt to determine the relative importance of uncertainties in CC impact assessments stemming from climate projections and crop models, respectively, and to analyze probabilities of yield outputs. Iizumi et al. (2011) found that the uncertainties of projected yield impacts for rice grown in different regions of Japan stemmed, in most cases, from climate projections, but that the relationship between crop model and climate projection uncertainty also varied considerably among regions.
7
The European Cooperation in Science and Technology (COST) Action 734 ’Impacts of Climate change and Variability on European Agriculture’, http://www.cost734.eu
133