Prioritizing Climate Change Adaptation Needs for Food Security in 2030

See allHide authors and affiliations

Science  01 Feb 2008:
Vol. 319, Issue 5863, pp. 607-610
DOI: 10.1126/science.1152339


Investments aimed at improving agricultural adaptation to climate change inevitably favor some crops and regions over others. An analysis of climate risks for crops in 12 food-insecure regions was conducted to identify adaptation priorities, based on statistical crop models and climate projections for 2030 from 20 general circulation models. Results indicate South Asia and Southern Africa as two regions that, without sufficient adaptation measures, will likely suffer negative impacts on several crops that are important to large food-insecure human populations. We also find that uncertainties vary widely by crop, and therefore priorities will depend on the risk attitudes of investment institutions.

Adaptation is a key factor that will shape the future severity of climate change impacts on food production (1). Although relatively inexpensive changes, such as shifting planting dates or switching to an existing crop variety, may moderate negative impacts, the biggest benefits will likely result from more costly measures including the development of new crop varieties and expansion of irrigation (2). These adaptations will require substantial investments by farmers, governments, scientists, and development organizations, all of whom face many other demands on their resources. Prioritization of investment needs, such as through the identification of “climate risk hot spots” (3), is therefore a critical issue but has received limited attention to date.

We consider three components to be essential to any prioritization approach: (i) selection of a time scale over which impacts are most relevant to investment decisions, (ii) a clear definition of criteria used for prioritization, and (iii) an ability to evaluate these criteria across a suite of crops and regions. Here, we focus on food security impacts by 2030: a time period most relevant to large agricultural investments, which typically take 15 to 30 years to realize full returns (4, 5).

We consider several different criteria for this time scale. First is the importance of the crop to a region's food-insecure human population [hunger importance (HI)]. Second is the median projected impact of climate change on a crop's production by 2030 (indicated by C50), assuming no adaptation. For this analysis, we generate multiple (i.e., 100) projections of impacts based on different models of climate change and crop response, in order to capture relevant uncertainties. The projections are then ranked, and the average of the 50th and 51st values are used as the median. A third criterion is the fifth percentile of projected impacts by 2030 (where C05 indicates the fifth value of the ranked projections), which we use to represent the lower tail or “worst case” among the projections. Finally, we consider the 95th percentile of projected impacts by 2030 (where C95 indicates the 95th value of the ranked projections), which we use to represent the upper tail or “best case” among the projections.

We first identified 12 major food-insecure regions, each of which (i) comprise groups of countries with broadly similar diets and agricultural production systems and (ii) contain a notable share of the world's malnourished individuals as estimated by the Food and Agriculture Organization (FAO) (Table 1; see fig. S1 for details on regions). For each region, we computed the HI value for each crop by multiplying the number of malnourished individuals by the crop's percent contribution to average per capita calorie consumption [see supporting online material (SOM) Text S1 and table S1]. A hunger importance ranking (HIR) was then generated by ranking the HI values for all crop-by-region combinations. Rice, maize, and wheat contribute roughly half of the calories currently consumed by the world's poor and only 31% of the calories consumed by those in sub-Saharan Africa, illustrating the importance of considering additional crops in food security assessments. The use of projected malnourished populations in 2030 rather than current population values had a very small influence on the rankings (table S2).

Table 1.

Regions evaluated in this study and selected summary statistics. Countries within each region are indicated in the SOM.

View this table:

Several options exist for evaluating climate change impacts across a suite of crops and regions (SOM Text S2). We used data sets on historical crop harvests (6), monthly temperatures and precipitation, and maps of crop locations to develop statistical crop models for 94 crop-region combinations spanning the 12 study regions (see SOM Text S3; results summarized in Table 1). Of these combinations, 46% (43) exhibited a statistically significant model (P < 0.05), and 22% (21) had a model R2 of at least 0.3. As seen in the examples for wheat in South and West Asia (fig. S3), in some cases the model's strength came primarily from a (typically negative) temperature effect on yield, whereas, in other cases, a (typically positive) rainfall effect provided most of the explanatory power.

The crop temperature sensitivities estimated by the statistical models were compared with corresponding values from previous studies that relied on established process-based models within the same regions (SOM Text S4). Our statistical estimates generally overlapped the lower end of the range of previous estimates, indicating that impacts estimated by the statistical models may be considered conservative but in reasonable agreement with estimates from process-based approaches.

To project climate changes for the crop regions, along with their uncertainties, we used output from 20 general circulation models (GCMs) that have contributed to the World Climate Research Programme's Coupled Model Intercomparison Project phase 3 (WCRP CMIP3) (7). Median projections of average temperature change from 1980–2000 to 2020–2040 were roughly 1.0°C in most regions, with few models projecting less than 0.5°C warming in any season and some models warming by as much as 2.0°C (Fig. 1A). In contrast to the unanimous warming, models were mixed in the direction of simulated precipitation change. All regions had at least one model with positive and one model with negative projected precipitation changes, with median projections ranging from about –10% to +5% (Fig. 1B). Some relevant tendencies of current GCMs, as noted in (8), are toward precipitation decreases during December to February (DJF) in South Asia and Central America, precipitation decreases in June to August (JJA) in Southern Africa, Central America, and Brazil, and precipitation increases in DJF in East Africa.

Fig. 1.

Summary of projected (A) temperature (°C) and (B) precipitation (%) changes for 2030 (the averages from 2020 to 2039 relative to those from 1980 to 1999) based on output from 20 GCMs and three emission scenarios. Gray boxes show DJF averages and white boxes show JJA averages. Dashed lines extend from 5th to 95th percentile of projections, boxes extend from 25th to 75th percentile, and the middle vertical line within each box indicates the median projection.

We estimated a probability distribution of production changes for 2030 (the average from 2020 to 2039 relative to that from 1980 to 1999) for each crop using a Monte Carlo procedure that propagated both climate and crop uncertainties (9). To facilitate comparison between crops and regions, we expressed production changes for all crops as a percentage of average values for 1998 to 2002. The impact projections are summarized in Fig. 2.

Fig. 2.

Probabilistic projections of production impacts in 2030 from climate change (expressed as a percentage of 1998 to 2002 average yields). Red, orange, and yellow indicate a HIR of 1 to 30 (more important), 31 to 60 (important), and 61 to 94 (less important), respectively. Dashed lines extend from 5th to 95th percentile of projections, boxes extend from 25th to 75th percentile, and the middle vertical line within each box indicates the median projection. Region codes are defined in Table 1.

For simplicity, we consider three general classes of projections. First, several projections (e.g., Southern Africa maize and wheat) are consistently negative, with an estimated 95% or greater chance that climate changes will harm crop production in the absence of adaptation (C95 < 0). These cases generally arise from a strong dependence of historical production variations on temperature, combined with projected warming large enough to overwhelm the uncertain impacts of precipitation changes.

Second, there are many cases with large uncertainties, with model impacts ranging from substantially negative to positive (e.g., South Asia groundnut, Southern Africa sorghum). These cases usually arise from a relatively strong dependence of historical production on rainfall, combined with large uncertainties in future precipitation changes. More precise projections of precipitation would therefore be particularly useful to reduce impact uncertainties in these situations. Large uncertainties also arise in some cases (e.g., cowpea in East Africa) from an estimated production response to historical temperature that is strongly negative but also highly uncertain.

Finally, there are many cases characterized by a narrow 90% confidence interval of impacts within ±5% of zero. In a few cases, such as wheat in West Asia, this reflects a strong effect of historical rainfall variations (fig. S1), combined with a relatively narrow range of rainfall projections during the growing season (Fig. 1; West Asia wheat is grown in DJF). In most cases, such as cassava in West Africa, the narrow confidence intervals result from a relatively weak relationship between historical production and growing-season climate. Therefore, we can only say that the likely impacts appear small, given the current data sets and models used to describe crop responses to climate. In cases with low model R2, approaches other than the FAO-based regression models used here may be more appropriate.

Based on the above projections, we identified a small subset of crops that met different prioritization criteria (Table 2). First, crops were separated into groups of “more important” (HIR = 1 to 30), “important” (HIR = 31 to 60), and “less important” (HIR = 61 to 94). Within each category, we identified crops below three thresholds: the first corresponding to instances where at least 5% of the models predicted greater than 10% loss of production (C05 < –10%), the second to where at least half the models projected greater than a 5% production loss (C50 < –5%), and the third to where at least 95% of the models predicted some production loss (C95 < 0%).

Table 2.

Crop priority lists based on different criteria. C05 = 5th percentile of projected impacts (5th lowest out of 100 projections); C50 = 50th percentile (median); C95 = 95th percentile. Results are shown only for the HIR = 1 to 30 and HIR = 31 to 60 categories.

View this table:

Although several crops met more than one of these criteria, such as maize in Southern Africa and rapeseed in South Asia, the varying estimates of uncertainty for different crops, in general, resulted in noticeable differences when prioritizing crops on the basis of the three different thresholds (Table 2). For example, a relatively weak relationship was found between values at the two tails of the projection distributions—C05 and C95—across all crops (fig. S4). This result indicates a need to explicitly consider uncertainty and risk attitudes when setting priorities, which is an issue that has received limited attention (10).

Because attitudes toward risk differ, and given that impact projections for some crops are more uncertain than those for other crops, various institutions might derive different priorities from the results in Table 2. For example, one set of institutions might wish to focus on those cases where negative impacts are most likely to occur, in order to maximize the likelihood that investments will generate some benefits. By this criterion (C95 < 0%), South Asia wheat, Southeast Asia rice, and Southern Africa maize appear as the most important crops in need of adaptation investments.

Others might argue that adaptation activities that do not account for worst-case projections will be inadequate in the face of low-probability, high-consequence climate impacts: that is to say, investments should target those crops and regions for which some models predict very negative outcomes. A different subset of crops is identified for this criterion (C05 < –10%), with several South Asian crops, Sahel sorghum, and (again) Southern Africa maize appearing as the most in need of attention.

Either of these risk attitudes could be applied with an explicit regional focus. For a sub-Saharan African institution interested in investing where negative impacts are most likely to occur [where median impact projections are substantially negative (C50 < –5%) or where most climate models agree that negative impacts are likely to occur (C95 < 0%)], priority investments would include Southern Africa maize, wheat, and sugarcane, Western Africa yams and groundnut, and Sahel wheat.

Despite the many assumptions and uncertainties associated with the crop and climate models used (SOM Text S5), the above analysis points to many cases where food security is clearly threatened by climate change in the relatively near-term. The importance of adaptation in South Asia and Southern Africa appears particularly robust, because crops in these regions appear for all criteria considered here (Table 2). The results also highlight several regions (e.g., Central Africa) where climate-yield relationships are poorly captured by current data sets, and therefore future work in this regard is needed to inform adaptation efforts.

Impacts will likely vary substantially within individual regions according to differences in biophysical resources, management, and other factors. The broad-scale analysis presented here was intended only to identify major areas of concern, and further studies at finer spatial scales are needed to resolve local hot spots within regions. Consideration of other social and technological aspects of vulnerability, such as the existing adaptive capacity in a region or the difficulty of making adaptations for specific cropping systems, should also be integrated into prioritization efforts. Although we do not attempt to identify the particular adaptation strategies that should be pursued, we note that, in some regions, switching from highly impacted to less impacted crops may be one viable adaptation option. In this case, the identification of less impacted crops is another valuable outcome of a comprehensive approach that simultaneously considers all crops relevant to the food-insecure.

Supporting Online Material

SOM Text S1 to S5

Figs. S1 to S5

Tables S1 to S3


References and Notes

View Abstract

Navigate This Article