Page 94 - Climate Change and Food Systems

climate change and food systems: global assessments and implications for food security and trade
annual basis. To analyse climate-change effects on output or yields, projected changes in temperature and rainfall from GCMs replace the historic climate data and new output or yields are calculated.
The key advantage of statistical crop
models is that adaptive farmer behaviour is implicitly described in the observed climate-yield relationship, for given land values. This advantage is tempered, however, in that time-series models describe adaptive management changes over
the entire period of analysis, without identifying specific adaptive strategies or responses to short- term climate extremes. Other advantages are
that statistical models can be estimated at local, regional and even global levels, and at annual and intra-annual scales. Compared to small-area crop simulation models, their larger scale and temporal dimensions are more compatible with the scale of inputs from climate models, which can eliminate the need for downscaling. Their robustness also can be evaluated based on statistical measures, and their data input requirements are considerably less than those of small-area crop simulation models.
Similar to crop simulation models, recent innovations in statistical crop modelling have focused on quantifying uncertainty. Tebaldi and Lobell (2008) were among the first to analyse uncertainty in both climate change projections and the climate/yield relationships in their crop models. Using three relatively strong global crop models
of barley, maize and wheat, with respective R2s
of 0.65, 0.47 and 0.41, the authors first develop probabilistic projections of combinations of temperature and precipitation changes based on SRE A1B from 20 GCMs. A sample of 1000 pairs
of temperature/productivity changes are used as predictors in the crop yield models to generate 1000 results. They then carry out a bootstrap analysis
of the yield results to describe uncertainty in the estimated model relationships between climate and crop variations. They find mean changes in global yields (without adaptation) between 2030 and 1980- 1999 of +1.6 percent (wheat), -14.1 percent (maize) and -1.8 percent (barley) and report their 95 percent probability intervals.
Lobell et al. (2008) build on these techniques
in an analysis of climate-induced yield changes
in food-insecure regions, in which they estimate statistical crop yield models to identify priority areas for adaptation investments. Using data on historical crop harvests, monthly temperature
and precipitation, and maps of crop locations, Lobell et al. estimated 94 statistical crop models
of grains, oilseeds, pulses, sugar and cassava crops in 12 food-insecure regions. To project yields under future climate conditions, the authors utilize outputs for SRESs A1B, A2 and B1from 20 GCMs that participated in the CMIP3. To address the uncertainty in both climate projections and the crop models, Lobell et al. use a Monte Carlo procedure to estimate a probability distribution of production changes in 2030. Based on their analysis, they identify hot spots for region and crop combinations where climate change is likely to have negative impacts, along with corresponding uncertainties about their projection, which vary widely by crop. Their analysis describes South Asia and Southern Africa as two regions that are likely to face the most severe outcomes for some of their crops.
Schlenker and Lobell (2010) carry out a similar analysis of climate change and crop production that addresses potential model errors in both inputs from GCMs and in the crop models. They apply
a panel data analysis to FAO crop yield data to estimate country-level statistical crop yield models for five crops in Sub Saharan Africa, based on temperature and precipitation data from 16 GCMS under the A1B scenario for 2046-2065. They carry out 1000 bootstrap runs for each of the16 GCMs models, reporting yield results in terms of the distributions of 16,000 impacts. The mean impacts on production describe substantial potential yield losses by the middle of the century of between -1 and −22 percent, relative to 1961-2000.
A second frontier in statistical crop modeling is their strengthened capability to capture short- term, intra-seasonal variability, which is found
to be important in explaining seasonal yields. Rowhani et al. (2011) compare the effects of intra- seasonal, monthly variability with inter-seasonal climate variability on rice, sorghum and maize
74