Page 93 - Climate Change and Food Systems

climate change projections received from climate inputs and in the crop models themselves. Like GCM models, crop models reduce complexity
by simplifying some aspects of the plant growth process at lower levels of the plant growth hierarchy. Parameter values that describe sub- processes may not correctly represent these bio- physical processes (Sinclair and Seligman, 2000). While they can be validated with historical crop yield data, climate change inputs may push crop models beyond the ranges of their applicability.
Tao et al. (2008) addresses uncertainty with respect to climate shock inputs derived from
GCMs. In a study of future rice yields at six stations in China, they apply Monte Carlo techniques to
20 climate scenarios to develop probabilities for climate change impacts. These results are used as inputs into calibrated DSSAT/CERES crop models, with automatic changes in irrigation and fertilizer applications, to generate probabilistic outcomes
for rice yields with respect to uncertainties in future climate projections. For instance, including CO2 fertilization effects, they find that the length of the rice growing season will be reduced with 100 percent probability for increases in mean global warming
of 1, 2, and 3C above 1961-1990 levels. Mean yields across stations change by -10.3 percent, -16.3 percent and -19.2 percent, respectively for temperature increases of 1, 2, and 3C.
Tao et al. (2009b) explore uncertainty in both climate model inputs and in the MCWLA crop model’s parameters in a study of maize production in two agricultural provinces in China. The crop model assumes endogenous adaptive responses by the farmer, with planting delayed until planting conditions are met, or until the planting window closes. They address climate uncertainty by using ten climate scenarios from five GCMS under SRESs A1F and B1, and 60 sets of crop model parameters, yielding 18,000 simulation yield results. Their probability distribution of results for the Henan province describe mean yield impacts of -10 percent, -16 percent and -24 percent during the 2020s, 2050s, and 2080s, respectively, as
a percent of 1961–1990 yields, with 95 percent probability intervals for each period of (-29, +16),
(-46, +24), (-93, +20). Their most important finding is that climate change scenarios contribute more to uncertainty in projected yield impacts than their crop model parameters.
Advances in achieving representative, large- area crop models and in quantifying the uncertainty related to climate change projections and to crop simulation model parameters strengthens the foundations that these models can provide for economic analyses. Yet a number of areas remain to be developed in future crop yield simulation modelling. In Rivington and Koo’s (2010) survey, crop simulation modellers identified the most important of these to be a reduction of parameter uncertainty, by improving our understanding of the effects of extreme heat and intra-seasonal climate variability on plant growth processes and yields, and the identifying threshold levels that lead to crop failure. Only half of the models represented in their survey accounted for the fertilization effects
of elevated CO2, and the question of how large its yield benefits might be is still open. Applications of crop models to more crops that have a large role
in diet in developing countries is also needed, in addition to the current focus on cereals, maize and rice.
Hertel and Lobell (2014) point out that the majority of crop models used in IAMs miss out on several climate-linked processes such as CO2 fertilization, effects of heat stress on grain set and leaf senescence, and pest and disease pressures (Howden, et al., 2007; Rosenzweig et al., 2000) and generally suffer from the lack of development and testing in extreme climate conditions
(White et al., 2005, 2011).
Statistical crop yield models
Statistical yield models are the second, and more common, type of model used to provide crop yield shocks in economic impact analyses, in large part because of their compatibility in spatial scale. Time series or cross-section estimations, or the two combined, describe empirical relationships between observed crop output or yield and projected changes in temperature, rainfall and other climate variables, usually on a monthly or
chapter 3: economic modelling of climate impacts and adaptation in agriculture: a survey of methods, results and gaps
73