Each grid box shows the correlation for the quarters with the highest mean rainfall

Most of the production increases supporting these surpluses may occur in Eastern and Southern Asia and Northern America, where our modeling suggests 47% and 28% of new production will occur as a consequence of 25 to 35% increases in yields. Our projected yields in Eastern Asia and Northern America reach 7,500 kg ha−1 . Yields of this magnitude assume further innovation and increasing petrochemical inputs, and may not be technically feasible . However, regionally, a continuation of recent trends that include vast disparities of access to food will probably expose hundreds of millions more people to chronic food insecurity, even if increasing cereal demands as a result of biofuels and increased consumption are ignored. With a 2030 population of about 2.1 billion, Southern Asia will face substantial food availability challenges. Our 2030 projections suggest a per capita cereal production of 193 kg per person per year . This value is slightly greater than our arbitrary subsistence threshold of 190 but substantially less than the 2007 value of 231. Hence, while our theoretical ‘food balance’ suggests sufficiency, real conditions will probably result in chronic food shortages for large segments of this diverse region who have negligible purchasing power. By 2050, our theoretical food balance suggests that regional cereal production might be adequate for only 90% of the population, leaving a shortfall equivalent to the amount required by 373 million people. Substantial water scarcity intensified by anthropogenic increases in air temperature and evaporation will further hamper agricultural expansion. Central Asia appears likely to face challenges similar to those of Southern Asia. Eastern and Western Africa,blueberry plant size where cereal crops provide the majority of calories, will face substantial and increasing food security challenges.

Per capita cereal production in Eastern Africa may decrease from a low 131 kg per person per year in 2007 to a very low 84 kg per person per year in 2030. This decline almost triples the theoretical food imbalance from -96 million in 2007 to -277 million people in 2030. This corresponds to 32% of the total population in 2007 and 56% of the population in 2030. Western Africa faces a similar, albeit more modest, decline in per capita production . Our theoretical food balance suggests that this could expose about 61 million people, or 14% of the population to chronic food shortages. This analysis suggests that Africa and Asia will experience continuing decreases in food availability and security. Rapidly growing populations and increasing temperature will place further demands on scarce water supplies. Biofuels and rising demand by the global middle class will probably compete for global production, raising prices and reducing food access for rural and urban poor. Eighty-eight percent of the 2007–2030 population growth will occur in African and Asian countries which will be strongly influenced climatically by the rapidly warming Indian and Pacific tropical Oceans .What do global climate change models tell us about 21st century rainfall? The models , on average, suggest increases in tropical rainfall over the Indian Ocean and tropical Pacific Ocean . In these regions with very warm surface waters, there is a clear relationship between SSTs and tropical atmospheric dynamics. Future warming of the oceans appears likely to increase rainfall over the tropical Indian and Pacific basins. This increased oceanic rainfall will release large amounts of energy into the atmosphere, impacting global and regional circulations.

These impacts may be quantified using the 21st century climate change simulations to calculate the PC1 and IO climate indicators . In general, the areas with increasing precipitation correspond to the geographic footprint of both PC1 and IO, and the models examined suggest that both PC1 and IO will increase by 2050 . The global response, which corresponds strongly to warming in the central Pacific, appears to increase in all quarters. The IO warming appears much greater during March-April-May and December-January-February than JuneJuly-August or September-October-November . However, there is an inherent uncertainty in all these projections due to differences in model formulations, natural 10-year variations in the climate and the imperfect simulations of key processes, such as El Niño. To quantify this uncertainty the differences between the simulations can be examined and the 68% confidence intervals obtained from these differences evaluated . In summary, the models appear to agree on substantial increases in the PC1 and IO indicators, implying associated changes in the Indian and Pacific Oceans circulations, but there is still a high level of uncertainty as to the size of the changes. When using climate change simulations, it is important to realize how poorly the models used in the IPCC assessments represent rainfall over land. The average seasonal correlation between 1980–2000 observed and modeled rainfall was examined .Multi-model ensemble estimates were made for each model, the correlations estimated, and then averaged across the models. In general, areas over the tropical oceans fit well with the climate models and have good correlations. Brown boxes denote areas where the IPCC models tend to perform very poorly, with correlations of less than 0.3. Dark green areas are reasonably skillful . In the Indian and Pacific Oceans, these areas also tend to be areas with substantial increases in rainfall predicted . However, over almost all land areas these evaluations suggest very small correlation coefficients.

This low level of skill makes analysis of simulations of ‘raw’ climate change rainfall problematic. Since the IPCC models tend to perform poorly over land and reasonably well over the oceans, this study adopted an alternative approach, based on hybrid-dynamicstatistical reformulations.Hybrid dynamic-statistical reformulations provide one potential way to overcome the limitations of global climate models. Instead of using the climate model precipitation directly, this analysis uses regression to relate changes at some location to large scale climate indicators . This is especially useful when there is good evidence linking changes in tropical oceanic rainfall and SSTs to terrestrial rainfall . Precipitation reformulations , based on the 1st and 2nd principal components of global precipitation suggest that substantial rainfall declines may occur over Central America, northern South America, Africa, and parts of Southern Asia, and Australia. For more detailed spatial analysis, regressions between African rainfall and PC1 and IO time-series may be used to downscale anticipated 21st century shifts in these climate forcings . The season with the highest mean rainfall was selected . Regression equations linking PC1 and IO to the local rainfall were then estimated. For most areas, these models explained 40–70% of the variance. For parts of sub-tropical Eastern Africa and Southern Africa near the Indian Ocean, increasing IO and PC1 values are associated with increasing aridity, warm anomalies in the south-central Indian Ocean and moderate-to-strong El Nino Southern Oscillation ,plant raspberry in container typically associated with below normal MAM or DJF rainfall . These historical relationships, combined with projected increases in the IO and PC1 indicators , suggest continued declines in rainfall across southern Ethiopia, Somalia, Kenya, northern Tanzania, southern Mozambique and southern Zimbabwe. While considerable uncertainty remains, it appears plausible, and even likely, that portions of Zimbabwe, Mozambique, Tanzania, Kenya, Somalia and southern Ethiopia may experience greenhouse gas induced rainfall reductions over the next 40 years. Therefore, if warming of the Pacific and Indian Ocean continues, as suggested by climate change models , anthropogenic drought appears likely to impact one of the most food insecure regions of the world. Our conclusions are generally in agreement with the most recent 4th IPCC finding that semiarid Africa may experience large-scale water stress and yield reductions by 2020 . Our work, however, avoids the direct use of terrestrial precipitation simulations due to their low accuracy . Focusing on downscalings of climate forcing diagnostics , however, suggests further drying, especially for Eastern Africa, where the IPCC report suggests that precipitation will increase.

Future expansion of this work into Asia could help confirm the potential decline in the Asian monsoon suggested by our global reformulations .In Africa, the trends determining food security are complex. Selected agricultural, food aid and population statistics for 18 semiarid food insecure countries in Western Africa, Eastern Africa and the eastern part of Southern Africa include combined data for Ethiopia and Eritrea as they were united before 1993. Geographic variations between these three regions play a strong role in their level of agricultural self-sufficiency. In 2005, the Western African countries had, on average, three times as much harvested area as Eastern Africa . Per capita harvested areas for southeastern portions of Southern Africa are only slightly higher than those for Eastern Africa . There are also considerable differences in per capita harvested area between the countries in each of these regions. For these countries, harvested area largely determines national cereal production totals.6 Over the period 2001– 2005, the relatively food secure Sahelian countries 7 have percapita agricultural capacity values above 190 kg person−1 year−1 . Over the same period, the southeastern Africa and Greater Horn countries had agricultural capacity values of 122 and 99 kg person−1 year−1 , respectively. Seed and fertilizer inputs were limited. In 2005, fertilizer inputs were typically below 20 kg ha−1 in these low productivity zones. Low yield growth combined with declining per capita harvested area has led to decreases in per capita agricultural capacity . Because of increases in population, these food insecure countries in Eastern, Southern and Western Africa have experienced, respectively, 18, 22, and 28% reductions in per capita harvested area between 1979 and 2005. Between 1979 and 2005, fertilizer increased in the Sahel and Greater Horn and declined in eastern Southern Africa. Of the four main users of fertilizer in 2005, Kenya had increased its fertilizer use from 21 to 67 kg ha−1 . Zambia, Zimbabwe, and Swaziland saw substantial reductions from the early 1980s. In these semiarid countries, a strong dependence on rainfed smallholder farming practices results in quasi-linear relationships between seasonal rainfall, grain yields, and food deficits. Hence, the agricultural capacity multiplied by rainfall is strongly related to per capita production. The inverse of this measure is related to food aid. For each country, the food imbalance measure was regressed against 1979–2005 WFP humanitarian assistance. This gives a pragmatic means of translating changes in rainfall, cropped area, seed use and fertilizer use into an index of potential food aid requirements, supported empirically by historical aid figures. Due to the low per capita production, the resulting model performed well at a regional scale for Eastern and Southern Africa but was less accurate for the Western African countries . Agricultural sufficiency may also be expressed as a theoretical food balance, based on an assumed annual cereals requirement of 190 kg per capita. Changes in the theoretical food balance agree strongly with changes in WFP food aid,8 explaining 70% and 85% of their variance at national and regional scales. Combining observed 1994–2003 agricultural capacity trends with our projected rainfall tendencies, this model can be used to project 2000 to 2030 food aid requirements . We show historical WFP aid figures, historical model aid figures, and results from four sets of aid projection scenarios. The first scenario assumes that recent trends in population, rainfall, crop area, seed use, and fertilizer use continue for the next 30 years. The second scenario is the same, but with the change in rainfall inferred from our 1950–2005 Indian Ocean regressions and 21st century climate simulations. The third scenario assumes that precipitation levels will remain similar to those observed today. The fourth is an ‘agricultural growth’ scenario, in which observed rainfall trends continue, but per capita food availability is assumed to increase by 2 kg per person per year. These results suggest that the interaction between drought and declining agricultural capacity may be explosive, dangerous and costly, with annual aid totals increasing by 83% by 2030. The ‘observed’ versus ‘projected’ trends differ primarily for the Sahel . The impact of climate change on the Sahel is keenly debated, and our analysis explicitly ignores influences from the Atlantic Ocean. Current agricultural capacity and rainfall trends will probably produce a 60% increase in food aid expenditures in the next two decades, and will probably lead to a 43% increase in food insecurity in Africa. These figures are significant because food aid is an indicator of many related problems including child malnutrition, as well as declines in health, productivity and economic growth .