Uncertainty

Uncertainty talkEnd-users, stakeholders, and the broader scientific community have made it clear that projected climate impacts are not helpful unless related uncertainties are described and assessed. For its assessment of future food security, AgMIP’s transdisciplinary framework produces cascading uncertainty passed from an ensemble of climate simulations under several scenarios to an ensemble of crop model simulations, and then aggregated to force an ensemble of economic models.

AgMIP is developing methods to identify and track uncertainties throughout its framework, beginning with pilot investigations that pass the limits of the interquartile range of climate model projections into an ensemble of crop models, with the resulting interquartile range of crop results passed into economic models. Uncertainty owing to observational dataset errors are also tracked through the framework. In this way AgMIP can provide estimates of uncertainty at various phases of the impacts assessment process and pinpoint crucial bottlenecks, which will help prioritize future data collection and model improvement efforts. Uncertainty estimates of the full ensemble of AgMIP results are presented as cumulative distributions describing the probability of each outcome. In some cases this will be an empirical distribution, based on the finite number of models in an ensemble. In others this will be a continuous distribution, if, for example, parameter uncertainty is described by a normal distribution. These distributions are summarized by standard deviation or confidence intervals. It is important to emphasize that the level of uncertainty depends on the formulation of the prediction problem. For example, Wallach et al. (2012) found that uncertainty in predicting yield averaged over many climate scenarios was much smaller than the uncertainty in prediction for a given scenario. The realism of the AgMIP uncertainty estimates will be verified to the extent possible by comparing the probability distributions of hindcasts with historic data using confidence intervals and the Brier score.

Climate uncertainties have been widely explored by the IPCC (Solomon et al., 2007); however, AgMIP aims to further analyze the likelihood of extreme events that are of particular relevance to agriculture (e.g., droughts, heavy downpours, extreme heat and cold, and frost). Fewer studies have explored uncertainties introduced by the crop and economics models (e.g., Aggarwal, 1995; Monod et al., 2006; Challinor et al., 2009). AgMIP’s first research track will perform crop and economic model intercomparisons, allowing the quantification of uncertainties relating to uncertainties of soil, weather, and management inputs; uncertainties of model parameters; and uncertainties related to model formulation. In economic models, uncertainties include population and income growth rates, elasticity estimates, rate of technological development, and price shocks.

References

Aggarwal, P.J., 1995. Uncertainties in crop, soil, and weather inputs used in growth models: implications
for simulated outputs and their applications. Agricultural Systems. 48, 361-384.

Challinor, A.J., Wheeler, T.R., Hemming, D., Upadhyaya, H.D., 2009. Crop yield simulations using a perturbed
crop and climate parameter ensemble: sensitivity to temperature and potential for genotypic adaptation to climate change. Climate Research. 38, 117-127.

Monod, H., Naud, C., Makowski, D., 2006: Uncertainty and sensitivity analysis for crop models, in: Wallach, D.,
Makowski, D., Jones, J.W. (Eds.), Working with Dynamic Crop Models. Elsevier.

Solomon, S., Co-editors, 2007. Climate Change 2007: The Scientific Basis. Contribution of Working Group I
to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge University Press, Cambridge.

Wallach, D., Brun, F., Keussayan, N., Lacroix, B., Bergez, J.-E., 2012: Assessing the uncertainty when using a
model to compare irrigation strategies. Agronomy Journal. 104, 1274-1283.