literature review on rainfall variability

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

• View all journals
• My Account Login
• Explore content
• About the journal
• Publish with us
• Sign up for alerts
• Open access
• Published: 17 April 2024

Climate damage projections beyond annual temperature

• Paul Waidelich   ORCID: orcid.org/0000-0002-5081-776X 1 ,
• Fulden Batibeniz   ORCID: orcid.org/0000-0002-1751-3385 2 , 3 , 4 ,
• James Rising   ORCID: orcid.org/0000-0001-8514-4748 5 ,
• Jarmo S. Kikstra   ORCID: orcid.org/0000-0001-9405-1228 6 , 7 , 8 &
• Sonia I. Seneviratne   ORCID: orcid.org/0000-0001-9528-2917 2  

Nature Climate Change ( 2024 ) Cite this article

10k Accesses

294 Altmetric

Metrics details

• Climate-change impacts
• Environmental economics
• Projection and prediction

Estimates of global economic damage from climate change assess the effect of annual temperature changes. However, the roles of precipitation, temperature variability and extreme events are not yet known. Here, by combining projections of climate models with empirical dose–response functions translating shifts in temperature means and variability, rainfall patterns and extreme precipitation into economic damage, we show that at +3  ° C global average losses reach 10% of gross domestic product, with worst effects (up to 17%) in poorer, low-latitude countries. Relative to annual temperature damage, the additional impacts of projecting variability and extremes are smaller and dominated by interannual variability, especially at lower latitudes. However, accounting for variability and extremes when estimating the temperature dose–response function raises global economic losses by nearly two percentage points and exacerbates economic tail risks. These results call for region-specific risk assessments and the integration of other climate variables for a better understanding of climate change impacts.

Similar content being viewed by others

The economic commitment of climate change

New damage curves and multimodel analysis suggest lower optimal temperature

Decadal global temperature variability increases strongly with climate sensitivity

Projections of economic damage from climate change are key for evaluating climate mitigation benefits, identifying effects on vulnerable communities and informing discussions around adaptation needs, as well as loss and damage financing. On a global or country level, such assessments have focused on how projected changes in annual mean temperatures affect gross domestic product (GDP) 1 , 2 , 3 , 4 . However, the widespread losses in recent years driven by flooding and drought suggest that precipitation variability and extremes are similarly important 5 , 6 . Anthropogenic forcing is increasing the frequency and intensity of precipitation extremes and variability on multiple scales, altering daily temperature patterns and driving an overall increase in precipitation over land 7 , 8 . Continued global warming is expected to exacerbate these trends, potentially with uneven impacts across regions 5 , 9 , 10 . Therefore, including precipitation, variability and extremes can improve the precision, comprehensiveness and interpretability of climate change damage estimations 11 .

Economic damage from climate change can be assessed either bottom-up by quantifying, valuating and aggregating specific impacts (for example, crop failures or labour supply changes) or top-down by identifying the statistical relationship between observed climatic shifts and economic growth. While both approaches have advantages and limitations, top-down approaches usually neglect climatic shifts beyond annual temperature changes 12 . To address this shortcoming, recent studies have estimated the relationship between macrolevel income and a wider range of climatic indicators, such as total precipitation 13 , 14 , 15 , temperature variability 16 , 17 or temperature and precipitation extremes and anomalies 14 , 18 , 19 . However, these studies do not investigate how much the inclusion of these climate indicators alters previous economic assessments of climate change, which is highly relevant for policy-making and future adaptation. A notable exception is ref. 15 , which projects the effects of annual precipitation and temperature shifts on inequality. A comprehensive assessment of the projected economic impacts of intense periods of precipitation and temperature anomalies, however, is still missing.

In this study, we draw upon recent advances in estimating dose–response functions, which relate shifts in various climate indicators (total precipitation, temperature variability, precipitation anomalies and extremes) to GDP changes 14 . Combining these functions with projections from 33 models of Coupled Model Intercomparison Project Phase 6 (CMIP6), including two large ensembles, we investigate how including these indicators affects the understanding of future economic impacts at different global warming levels. Variability and extremes introduce substantial climatic and associated economic uncertainties and we conduct a variance decomposition to determine the main uncertainty drivers. Furthermore, we explore how including variability and extremes in empirical regressions alters the dose–response function for annual mean temperature, which the extant literature has estimated controlling only for annual precipitation 1 , 2 , 4 , 20 , 21 .

Projecting GDP impacts for six climate indicators

Compared to annual temperature, future changes in precipitation and temperature variability under climate change are subject to high uncertainties 8 , 22 , 23 . To capture these uncertainties, we use projections from various CMIP6 models 10 to analyse four climate indicators besides annual mean temperature and annual precipitation: (1) day-to-day temperature variability (how much daily temperatures deviate from monthly means); (2) extreme precipitation (the annual sum of precipitation on days with exceptionally high precipitation exceeding the historical 99.9th percentile); (3) monthly precipitation deviation (how much monthly precipitation deviates from historical means throughout the year); and (4) the number of ‘wet days’ with precipitation above 1 mm d −1 . These indicators have been linked to forcing from GHGs 24 , 25 as well as to income growth using a global sample 14 , 16 . Considering all indicators in one coherent estimation framework is crucial because variability and extremes correlate strongly with annual temperature, total precipitation and each other (Supplementary Fig. 3 ). Therefore, combining dose–response functions from different estimations risks double-counting impacts. Notably, our coherent estimation framework 14 does not model damage from droughts and heat waves. Therefore, we include heat in a complementary analysis, whereas we find no significant statistical link to economic growth for droughts, potentially due to limited spatial and temporal resolution and impact heterogeneity across regions (Supplementary Appendix F ).

We illustrate our approach for the example of extreme precipitation impacts on economic output for New York State at +3°C of global warming (Fig. 1 ). On the basis of how a given CMIP6 model and scenario project the respective climate indicator (Fig. 1a ), we compare the GDP impacts in a given year against the average impacts if the climate were to remain constant at levels of a recent baseline period (Fig. 1b ) 2 , 15 . For each model and scenario, the baseline period is the 41-year period during which global warming equals the historical warming between 1979 and 2019 (+0.84 °C), which is the climatic baseline used for estimating the dose–response functions used here ( Methods ) 14 . We then repeat this procedure for different CMIP6 models and potential damage parameter estimates based on statistical uncertainty and aggregate results to the national level. This yields a distribution of GDP impacts for each country featuring all years in each model and scenario associated with the same global warming level (Fig. 1c ). Therefore, the main sources of uncertainty in our GDP impact distribution for a given global warming level and territory are (1) internal variability for the same CMIP6 model because the magnitude of extremes can vary strongly from year to year and, for large ensembles, across model runs, (2) statistical uncertainty in the dose–response functions and (3) projection differences between CMIP6 models.

a , Projected extreme precipitation under SSP3-7.0 for an example model run (ACCESS-CM2, black) and other CMIP6 model runs (grey). Vertical lines denote the baseline period (blue) and the +3 °C global warming window (brown). b , Dose–response function for extreme precipitation (black line) and 95% confidence interval (grey area). Coloured dots and the blue diamond represent extreme precipitation levels from a and the baseline period average. The red error bar illustrates the difference between the dose–response function for an example year (2061) and the baseline average, which equals the projected damage for this year. Dose–response function values are transformed from the original logarithmic changes to percentage of GDP by exponentiating and subtracting one ( Methods ). c , Distribution of projected extreme precipitation damage at +3 °C by CMIP6 model. Boxplot centre, hinges and whiskers denote median, upper/lower quartiles and upper/lower deciles, respectively. For the CESM2-LE and MPI-ESM1-2-LR large ensembles, the +3 °C global warming-level window varies across ensemble members.

Source data

Global results.

We examine the impact on global GDP for all indicators combined, as well as the separate impacts from annual temperature, annual precipitation and the four variability and extremes indicators (Fig. 2a ). Global GDP is estimated to be 3.2% lower (lower/upper decile: 1.2–5.4%) at +1.5 °C of global warming, compared to a world with no further climate change beyond recent levels. At +3 °C, global GDP decreases by 10.0% (5.1–14.9%). When disaggregated by climate indicator, global impacts are strongly determined by the annual mean temperature changes, which account for a GDP reduction of 10.0% at +3 °C. This estimate is consistent with recent top-down studies focusing exclusively on damage from annual temperature changes and projecting impacts of 7–14% of GDP per capita loss by the end of the century at global warming levels of more than +4 °C (refs. 4 , 8 , 20 ); with other top-down studies estimating damage even higher 2 as a result of their assumption that temperature changes impact growth trajectories persistently 12 , 26 , 27 . For context, a 10% reduction exceeds the GDP loss of the COVID-19 pandemic when global growth plummeted from +2.6% in 2019 to −3.1% in 2020 or the effect of the global financial crisis in 2009 when global output shrunk by −1.3% (ref. 28 ). Importantly, recent research suggests that damage attributed to annual temperature covers heat wave impacts at least partially 18 . Indeed, when disentangling the two, we find that almost half of annual temperature damage can be attributed to heat extremes (Supplementary Appendix F ).

a , Points and the error bars centred around them denote the mean and upper-to-lower-decile range, respectively. ‘Variability and extremes’ comprises the four indicators in b . Labelled grey horizontal lines denote example growth rates in real GDP 28 . b , Same as a , with ‘variability and extremes’ impacts disaggregated by indicator. c , Variance decomposition for the combined GDP impacts of all climate indicators and for indicator-specific impacts. Indicator-specific decompositions are feasible because impacts in the underlying regression model are additive and hence can be projected separately. For absolute variances and coefficients of variation, see Supplementary Figs. 8 and 9 . d , Same as c , with ‘variability and extremes’ disaggregated by indicator.

By contrast, increases in annual precipitation in many areas have a small positive impact on global GDP (0.2% at +3  ° C warming) and the combined impact of the variability and extremes indicators remains centred around zero. While this suggests a lack of signal, disaggregating projections by individual indicators reveals otherwise (Fig. 2b ). At +3 °C, extreme precipitation reduces global GDP by 0.2% (0.1–0.4%), with 99% of our impact distribution indicating economic losses as extreme precipitation increases around the globe (Supplementary Fig. 5 ). Notably, these impacts are much lower than annual temperature damage. This is somewhat expected because extremes have a lower temporal and spatial correlation. Therefore, aggregation from daily, location-specific events to annual indicators and country-level projections reduces signals more strongly compared to annual mean temperature 13 , 14 . However, a 0.2% GDP loss due to extreme precipitation globally for an average year still represents a tenth of the damage caused by the catastrophic 2022 floods in Pakistan, estimated at 2.2% of GDP 29 . Global GDP losses from extreme precipitation are compensated, on average, by temperature variability reductions in higher latitudes (+0.1% of global GDP at +3 °C) 24 , 30 and fewer wet days (+0.2%). However, only 63% and 74% of the impact distribution imply global economic gains for these indicators, respectively. Monthly precipitation deviations, on average, add to global GDP losses (0.2% at +3 °C), but uncertainty ranges remain centred around zero.

To explore uncertainty drivers, we decompose the variance in GDP impacts from each climate indicator into statistical dose–response function uncertainty, climate model uncertainty and internal variability (Fig. 2c ). For annual temperature damage, uncertainty is primarily driven by the dose–response function, particularly at higher global warming levels. By contrast, impact uncertainty for annual precipitation and variability and extremes is dominated by internal variability. Additional analyses focusing on the two large ensembles in our sample suggest that this stems primarily from interannual variation within model runs rather than differences across ensemble members (Supplementary Figs. 15 and 16 ). Moreover, disagreement between CMIP6 models plays either a comparable or a more substantial role than dose–response function uncertainty for these additional indicators (except for monthly precipitation deviation) and is particularly pronounced for day-to-day temperature variability and the number of wet days (Fig. 2d ). Notably, the share of climate model uncertainty decreases in global warming for annual temperature impacts but not for variability and extremes because even for a stronger global warming signal, GDP impact projections do not converge between models.

Country-level results

Because global aggregates risk masking heterogeneities across regions, we further investigate the combined country-level GDP impacts from all six climate indicators at +3  ° C of warming (Fig. 3a ). All countries face GDP losses, in line with recent evidence that climate change might not benefit cooler countries economically 20 . Impacts are more severe in the Global South and highest in Africa and the Middle East, where higher initial temperatures make countries particularly vulnerable to additional warming. Considering the combined GDP impact of all four variability and extremes indicators reveals a clear North–South divide (Fig. 3b ). While for higher latitudes, the decrease in temperature variability and, for some countries, wet days (Supplementary Fig. 5) mitigates overall GDP damage, variability and extremes exacerbate GDP losses in most parts of the Global South. However, these effects vary substantially across the full distribution of projected impacts for each country.

a , Mean GDP impact at +3 °C of global warming for sovereign countries (other territories in dark grey) considering all six indicators in c , using shapefiles from ref. 42 . b , Same as a but only considering the bottom four ‘variability and extremes’ indicators in c . c , Mean GDP impact ( x axis) and share of the impact distribution agreeing with the sign of the mean ( y axis) for sovereign countries by World Bank income group (colour) and the global economy (grey diamond) at +3 °C. Middle income comprises lower- and upper-middle-income countries for conciseness. Dashed horizontal lines denote thresholds for 66% and 90% likelihood following IPCC terminology 8 .

Annual temperature is the only indicator where negative impacts arise for at least 90% of our impact distribution for all countries (upper dashed line in Fig. 3c ), including substantial impacts in several colder countries partially because the temperature dose-response function deployed here implicitly features damages from higher inter-annual temperature variability as well (Supplementary Appendix C). Annual precipitation increases benefit most countries on average, but for many countries, less than two-thirds of the distribution supports the sign of expected impacts (lower dashed line). For day-to-day temperature variability, we find a clear divide between relatively certain gains for a few high-income countries and less certain, smaller losses for many lower-income countries as a result of variability increases in lower latitudes 24 . While extreme precipitation increases in most regions, projected damages are highest and least uncertain for middle- and high-income countries in higher latitudes such as China and the United States 31 . In contrast, low-income countries are more likely to face losses from shifts in precipitation deviation and wet days, but high uncertainties limit the conclusions that can be drawn.

Overall impact of including variability and extremes

The results in the previous sections seemingly suggest that including variability and extremes in GDP impact projections exacerbates disparities between higher- and lower-income countries (Fig. 3 ) but does not substantially alter the implications of climate change for global GDP (Fig. 2 ), particularly since annual temperature damages capture heat wave impacts to some extent already. However, providing an apples-to-apples comparison with the recent climate economics literature requires calculating damage on the basis of the current ‘status quo’ approach, which (1) projects only damage from annual temperature changes and (2) estimates the relationship between income growth and annual temperature controlling only for annual precipitation 1 , 2 , 4 , 20 , 21 . When comparing the resulting global GDP impacts following this status quo methodology to our approach, which (1) projects damage for all six indicators and (2) controls for all our climate indicators when estimating the temperature dose–response function (Fig. 4a ), we find that including variability and extremes increases global damage at +3 °C of global warming by 1.8 percentage points (10.0% instead of 8.2%).

a , Dots represent mean values of the GDP impact distribution, while boxplot centre, hinges and whiskers denote median, upper/lower quartiles and upper/lower deciles, respectively. b , GDP impact of a +1 °C increase in the annual temperature of a territory for different initial temperature levels ( x axis) using mean marginal effects (lines) with 95% confidence intervals around them (shaded area). Estimated using Supplementary equation ( 22 ) and the regression models in Supplementary Table 7 (columns 1, 2 and 5). No confidence interval shown for ‘+ Temperature variability’ for visual conciseness. c , Difference in mean GDP impacts between our main approach and the status quo approach at +3 °C for sovereign countries (other territories marked in dark grey), using shapefiles from ref. 42 .

The main reason for this increase is that controlling for variability and extremes, instead of only for annual precipitation, increases the estimated effect of mean temperature changes (Fig. 4b ). The marginal GDP impact of a +1 °C rise in annual temperature increases by more than 0.5 percentage points irrespective of the initial temperature level when all climate indicators are included as control variables (red line). Most of this effect is driven by including temperature variability (dotted line), which leads to higher impacts, particularly for colder regions. Therefore, the positive impacts of temperature variability in Fig. 3 obscure that, in fact, including this parameter leads to higher global damage since it disentangles potential benefits of reduced variability from the negative effects of temperature increases. As a result, including all climate indicators exacerbates GDP impacts across the globe (Fig. 4c ).

Exposure to tail risks

Aside from average impacts and the uncertainty around them, prudent risk management by policy-makers also requires information about tail risks. Therefore, we examine the percentage of the present global population living in countries that have a non-negligible chance (at least 5%) of suffering from damage exceeding different thresholds at different global warming levels (Fig. 5a ), both for our main approach (solid line) and the status quo approach (dotted line). Even at +1.5 °C, tail risks are substantial, with 99% of the global population living in countries with a non-negligible risk of suffering GDP damage of 5% or higher if all climate indicators are included. Notably, including variability and extremes increases tail risks considerably (Fig. 5b ). While under the status quo, 54% of the global population is projected to face damage of at least 15% with a likelihood of at least 5% at +3  ° C of warming, this increases to 68% of the population when variability and extremes are included. The share of the global population facing catastrophic impacts of 20% or higher with a 5% chance rises from 4% to 17%. However, disaggregating these results by individual climate indicators (Supplementary Fig. 6 ) highlights that the increase in global exposure to catastrophic climate change damage is primarily driven by higher temperature damage if underlying regression models control for more climate indicators than just annual precipitation and less by the direct impacts of these indicators on global GDP.

a , Share of the current global population living in countries whose projected GDP impacts for a given warming level (colour) exceeds the respective threshold ( x axis) for at least 5% of the GDP impact distribution, based on the status quo approach (dotted lines) and our main approach using all climate indicators (solid lines), respectively. The grey arrow and text annotation provide a reading example. b , Selected values from a .

Taken together, our results illustrate that projecting top-down damage of variability and extremes exacerbates global disparities further. Aggregate GDP loss projections, however, remain primarily driven by the impacts of mean temperature changes, which seem to cover economic losses due to heat waves at least partially 18 —an important finding for climate–economy modelling that complementary assessments of economic damage should corroborate to disentangle different impact channels better. As a result, overall uncertainty in GDP losses is dominated by the temperature dose–response function. However, substantial climatic uncertainties still limit the understanding of direct impacts by variability and extremes, particularly for low-income countries, which are expected to suffer the most but exhibit the largest uncertainties.

For scholars studying the economic effects of climate change, our results suggest a potential downward bias in temperature damage estimates by not disentangling the impacts of changes in temperature means and temperature variability. Future studies estimating temperature dose–response functions should test how including variability and extremes indicators linked to economic growth alters their findings. Notably, such biases could also be caused by other climate indicators not explicitly considered here and their direction and magnitude are likely to vary by location 32 . Furthermore, since the signal clarity is highest for extreme precipitation, this indicator seems most suitable to be included in climate–economy calculations, such as the social cost of carbon.

While our results rest on strong empirical foundations and a wide range of state-of-the-art climate models, there are several reasons why actual GDP impacts may exceed our projections. First, while the temperature dose–response function seems to include heat impacts at least partially, the dose–response functions used here do not explicitly cover important climate extremes, most notably droughts 33 . Second, to be conservative, we abstract from the possibility that climatic shifts do not only change GDP growth in a given year but alter a country’s long-run growth trajectory persistently. While such persistence in GDP losses remains empirically debated 1 , 2 , 14 , 21 , 34 , it would increase damage estimates substantially 26 , 27 . Third, aggregation across time and space is more likely to reduce signals in precipitation patterns because of their lower spatial and temporal correlation compared to annual mean temperature 13 , 14 . For these reasons, our results should be seen as an important first step, but they certainly do not exclude the possibility of larger GDP losses. Furthermore, econometric-based dose–response functions such as the ones used here have several limitations; for example, the risk of conflating weather impacts with climatic shifts or the extrapolation of impacts to warming levels that go far beyond historical observations, with the implicit assumption that adaptation remains at historically observed levels 35 , 36 . In addition, specification questions can further exacerbate socioeconomic uncertainties 21 and uniform dose–response functions for aggregate GDP can mask heterogeneities between countries, sectors and income segments 15 . Moreover, considering impacts in percentage of GDP implicitly assigns lower weights to poorer regions within countries that are disproportionately exposed to climate change risks 37 . Lastly, the distributions presented here might underestimate true climatic uncertainties for at least three reasons: (1) measurement imperfections in the reanalysis data underlying the dose–response functions, particularly for precipitation and lower-income regions 32 , 38 ; (2) using single runs for most CMIP6 models may underestimate tail risk events (Supplementary Appendix E ); and (3) not all CMIP6 models, despite representing the current frontier of global climatic projections, fully capture future changes in temperature variability and precipitation 24 , 25 , 39 , 40 , 41 .

Nevertheless, our study represents a key improvement in top-down damage projections, highlights the risks of omitting climate indicators beyond annual temperature, either as impact channels or control variables, and identifies the most promising fields for additional research. Building on our work, researchers could integrate further climate indicators, such as droughts, into a comprehensive assessment of climate change impacts. Aside from improvements in climate modelling, particularly for developing countries, this would also require more empirical studies to robustly identify the link between economic growth and different climatic extremes, ideally combined with an improved temporal or spatial resolution 17 . In addition, future research should assess the impact of controlling for these extremes on temperature dose–response functions and enhance the understanding of potential adaptation mechanisms and the persistence of GDP losses, ideally by consistently estimating and implementing persistence for each climate indicator under consideration.

Climatic data

Daily temperature and precipitation projections are taken from 33 CMIP6 models under two low-emission scenarios (Shared Socioeconomic Pathways SSP1-1.9 and SSP1-2.6) and one high-emission scenario (SSP3-7.0) to calculate bias-corrected, annual climate indicators for the 1850–2100 period. Owing to computational constraints, we use only one realization for most model–scenario pairs. However, to explore the role of intra-ensemble variation, we include 30 realizations of MPI-ESM1-2-LR and all 100 realizations of the CESM2-LE large ensemble under SSP3-7.0, which provides us with a total of 199 model-realization–scenario pairings (Supplementary Tables 1–3 ). Time series switch between historical scenarios and the respective Representative Concentration Pathway (RCP)–SSP pair in 2015 and are regridded onto a common 2.5° × 2.5° longitude–latitude grid using conservative remapping 43 .

Consistent with our source of empirically calibrated dose–response functions, which relies on ERA5 data 14 , we calculate annual average temperature T , annual total precipitation RA as well as four climate indicators using the equations listed below before downscaling and regridding the annual indicators from 2.5° to 0.25° (the grid resolution of ERA5). Notably, the indicators used here have been motivated and subjected to various robustness checks by previous studies 14 , 16 .

Day-to-day temperature variability:

where T x ,d,a,t is the temperature for grid cell x of day d of month a in year t and D a   ∈  {28, 30, 31} is the number of days in the respective month a . \({\bar{T}}_{\rm{x,a,t}}\) denotes the mean temperature in month a of year t for the respective grid cell.

Extreme precipitation:

where R x,d,t is the precipitation of grid cell x on day d of year t , I () is an indicator function and R x,99p9,base denotes the 99.9th percentile of daily precipitation in grid cell x over a historical baseline period.

Number of wet days with precipitation exceeding 1 mm d −1 :

Grid-cell-level annual climate indicators are then aggregated to the subnational region level (ADM1) using the geospatial data from the Database of Global Administrative Areas (GADM, v.3.6) and area weighting.

Monthly precipitation deviation, which we calculate only at the ADM1 level and not at the grid-cell level, consistent with ref. 14 :

where R i, a ,t denotes precipitation totals in month a of year t for a given ADM1-level region i . Variables denoted by a bar represent averages across the baseline period, either for the full year or for a specific month, while σ i,a,base denotes the month-specific standard deviation across the baseline period for region i . As for all other climate indicators, region-level monthly precipitation R i,a,t is derived from grid-cell-level values based on area weighting.

For the baseline-dependent climate indicators \(\hat{\rm{RD}}\) and RM, our source of dose–response functions 14 uses 1979–2019 as the historical baseline period, during which global warming relative to pre-industrial levels over 1850–1900 averaged +0.84 °C according to Berkeley Earth data (the best estimate for the observed warming and, in a previous version, used in the IPCC AR6; ref. 8 ). However, the 1979–2019 time period can differ climatically across CMIP6 models, which warm at very different paces 44 . To maintain consistency and ensure that all climate indicators are based on the same baseline in terms of global warming, we, therefore, identify the corresponding 41-year window during which global warming relative to pre-industrial levels over 1850–1900 averages +0.84 °C for each climate model-realization and scenario. Then, we use the +0.84 °C window to calculate all values with a ‘base’ subscript in equations ( 2 ) and ( 4 ). Warming-level windows for each model-realization–scenario pairing are calculated following the approach by ref. 10 and shown in Supplementary Tables 1–3 . However, percentile-based indicators, such as our extreme precipitation measure, can exhibit artificial jumps at the end of the baseline period, causing potential frequency increases and discontinuities outside this period 10 , 45 , 46 . To correct this, we use the bootstrap resampling procedure developed by ref. 46 , estimating the percentile applied to each year in the baseline period on the basis of the remaining 40 years in the baseline period and then using the average across these percentiles for all years outside the baseline period. Mathematical expressions for the bootstrap resampling procedure and the calculation of global warming levels, as well as more information on the suitability of CMIP6 and ERA5 data to assess variability and extremes, are provided in Supplementary Appendix A .

Bias correction

We bias-correct annual climate indicators using the change factor method 47 , which adds model-projected changes compared to a reference period to the corresponding reference period average of an observational dataset. For any climate indicator C out of the six indicators considered here, bias-corrected values are obtained as follows:

where C x,t,m,r,s represents the raw climate indicator output of climate model m ’s realization r under scenario s in year t for grid cell x . \({\bar{C}}{\,\!}_{\rm{x},{\rm{ref}}}^{\rm{ERA5}}\) and \({\bar{C}}_{\rm{x},{\rm{ref}},m,r,s}\) represent the reference period average in ERA5 and for the climate model run, respectively. As a reference period, we use 1950–1990, during which global warming averaged +0.38 °C according to Berkeley Earth. Therefore, \({\bar{C}}_{\rm{x},{\rm{ref}},\rm{m,r,s}}\) is calculated for the 41-year global warming-level window corresponding to +0.38 °C (for the specific global warming-level windows, see Supplementary Tables 1–3 ). We bias-correct each annual indicator separately and, for the monthly precipitation deviation, apply the change factor method to the underlying monthly precipitation amounts. As a reference period, we use 1950–1990 because of its low influence of anthropogenic forcing and, to avoid impossible values, further impose zero lower bounds for all non-negative climate indicators and an upper bound of 365 for the number of wet days. In addition, the bias-corrected monthly precipitation deviation in some selected cases yields values that are one or two orders of magnitude above the maximum in our raw CMIP6 data. To address these outliers, we cap bias-corrected monthly precipitation deviation on the basis of the highest values observed for the raw CMIP6 data for up to +3 °C of global warming (~11.6; Supplementary Table 4 ), which affects only observations beyond the 99.993th percentile of our distribution.

Bias-correcting annual climate indicators ensures the highest consistency for each indicator with the ERA5 data used to estimate dose–response functions by ref. 14 (Supplementary Figs. 1–3 ). However, it can lead to inconsistencies between the different climate indicators derived from the same daily temperature or precipitation and, as outlined above, to outlier values in a few cases. As a robustness check, we apply the change factor method to the underlying daily temperature and precipitation values of an example model run instead, which increases the computational burden of bias correction considerably but leaves our conclusions unchanged (Supplementary Fig. 4 ).

GDP impacts

Dose–response functions for subnational economic growth are taken from ref. 14 , which jointly estimates the impact of all six indicators on income per capita growth. The resulting dose–response functions for each climate indicator are shown in Supplementary Fig. 7 and the underlying regression is reproduced in Supplementary Table 7 , column 5. Importantly, this regression model has been subjected to comprehensive robustness checks, such as using alternative datasets and variable definitions, controlling for region-specific time trends or assessing seasonal heterogeneities 14 . Mathematically, the specification of the regression can be summarized as

while the full model is written out in Supplementary equation ( 5 ). Here, g i,t denotes the economic growth of ADM1-level region i in year t , measured as the first difference of the log-transformed gross regional product per capita 48 . α i , δ t and ϵ i,t denote fixed effects and the error term and h C is the estimated dose–response function specific to climate indicator C i,t where \(C\in \{T,{\rm{RA}},\widetilde{T},\hat{{\rm{RD}}},{\rm{RD}},{\rm{RM}}\}\) . For instance, for annual precipitation RA, the relationship with economic growth is estimated as a quadratic relationship such that

where \({\beta }_{1}^{{\rm{RA}}}\) and \({\beta }_{2}^{{\rm{RA}}}\) are the respective regression coefficients.

To calculate the impacts of climate change, we compare annual economic impacts against the average impact during the historical baseline period for the same model–realization–scenario pairing, such that our impacts represent changes from a hypothetical scenario in which climate remains constant, following previous studies 2 , 15 . As a baseline period for GDP impacts, we again use the +0.84  ° C global warming-level window for a given realization r of climate model m and RCP–SSP pair s for consistency with the calculation of our climate indicators. Therefore, annual impacts, in percentage of GDP, of all climate indicators combined due to shifts relative to the baseline period are calculated as follows

where K is the 41-year model–realization–scenario-specific baseline period corresponding to +0.84 °C of global warming. Note that we exponentiate and subtract one to convert logarithmic changes to percentage of GDP, but impacts of different indicators and years are added and averaged in log scale. Individual GDP impacts of each climate indicator are obtained by using only the respective individual dose–response function in equation ( 8 ), instead of the sum across dose–response functions ∑ C h C ( C i,k ). Similarly, the combined GDP impacts of variability and extremes are calculated by summing only the dose–response functions for \(\widetilde{T}\) , \(\hat{{\rm{RD}}}\) , RD and RM in equation ( 8 ). More detailed mathematical expressions for all steps in equation ( 8 ) are provided in Supplementary Appendix C .

Importantly, the model specification by ref. 14 features annual temperature in first-differences compared to previous years and not in absolute levels:

To translate these regression coefficients into impact projections, we calculate cumulative impacts following ref. 4 , such that the dose–response function for annual temperature used in equation ( 8 ) reads as follows:

where k 0 denotes the first year in the baseline period K . As we discuss in Supplementary Appendix C , this procedure implicitly includes impacts of inter-annual temperature variability that persist over time.

For extreme precipitation \(\hat{{\rm{RD}}}\) , the dose–response function estimated by ref. 14 interacts extreme rainfall with the annual mean temperature T because the marginal impact of extreme precipitation is lower in warmer climates. Projecting this out under climate change, however, would make the strong assumption that global warming increases the resilience of countries to extreme precipitation worldwide. Because there is no evidence supporting such a positive feedback of warming and because the heterogeneity of extreme rainfall effects in ref. 14 is equally well-explained by the latitude of a country (see R2 and Adjusted R2 in Supplementary Table 4 of ref. 14 ), which is time-constant, we hold temperature in the interaction constant at the average level during the baseline period such that

When projecting damage of climate change, a core methodological choice is whether to assume that impacts affect GDP levels, such that the economy bounces back in the following year or whether to assume that a part of the damage persists and alters the long-run growth trajectory. Assuming persistence has a substantial impact on damage projections and the associated uncertainty 21 , 26 , 27 . Recent empirical analyses differ in methods and outcomes, with no consensus yet 1 , 2 , 12 , 21 , 34 . To be conservative, here we assume no persistence in implementing δ i,t , aside from any persistence implicit in the methodology by ref. 4 , and provide further mathematical expressions and discussions of damage persistence in Supplementary Appendix C . Furthermore, by holding historical dose–response functions and baseline periods for climate indicators constant, our approach rests on the common implicit assumption that future adaptation outcomes mirror historical ones 2 , 4 , 35 , in line with the mixed evidence on macro-economic adaptation so far 2 , 20 , 21 . In addition, we follow ref. 27 in equating relative GDP per capita impacts with relative GDP impacts (that is, assuming that any decrease in GDP per capita is caused by a climate change-induced reduction in economic output and not by an increase in population).

Spatial aggregation of GDP impacts

We aggregate the GDP impacts calculated via equation ( 8 ) from the subnational detail (ADM1) to the country level (ADM0) using GDP weighting. For GDP weights, we use 2010 GDP data downscaled to a 0.5° grid by ref. 49 . To deal with 105 outlier grid cells with raw GDP data exceeding US$10 20 , we apply a ceiling at $10 10 , which is the next highest grid-cell GDP value in the dataset. Note that we hold this intracountry income distribution constant across all years and SSPs. This simplification stems from the SSPs not directly informing spatial intracountry GDP per capita distributions and also prevents our results from being driven by changes in the weighting scheme over time rather than climatic changes, which is standard practice in the literature 2 . To calculate GDP impacts at the global level, we weigh each country i with its share in global GDP in year t as per the respective SSP. Since the SSP database does not provide GDP growth trajectories for a few sovereign countries, namely Andorra, Liechtenstein, Kosovo, Nauru, North Korea, San Marino, South Sudan and Tuvalu, these economies are not represented in our damage projections for the global economy.

GDP impact distribution

To capture dose–response function uncertainty, we draw 1,000 estimates for the dose–response function parameters \({\beta }_{1}^{{\rm{RA}}},{\beta }_{2}^{{\rm{RA}}},{\beta }_{1}^{\rm{T}},\ldots\) jointly from the multivariate Gaussian distribution estimated by ref. 14 (main specification, standard errors clustered at the country level). Applying each Monte Carlo draw to each of the 199 model–realization–scenario pairings provides us with 199,000 different impact projection pathways for each territory. For each model-realization–scenario pairing, we then identify the 20-year window corresponding to a global warming level of +1 °C, +1.5 °C, +2 °C, +3 °C and +4 °C, respectively, following the approach by ref. 10 (for the specific global warming-level windows, see Supplementary Tables 1–3 . This provides us with a conditional distribution of GDP impacts for a given territory and warming level. Aside from reducing the importance of individual RCP–SSP scenarios, conditioning results on global warming levels also reduces the need to omit or down-weight ‘hot models’ in CMIP6, which project too much warming 44 . Since not all models reach all warming levels and to prevent the two large ensembles from dominating our results, we weight models inversely such that each CMIP6 model has the same sampling probability for each warming level following ref. 18 . All summary statistics of the distribution (means, percentiles, variances and so on) are calculated using these CMIP6 model weights.

Variance decomposition

To decompose the observed variance in our conditional global GDP impact distribution for a given warming level, we adapt the approach by ref. 50 based on the partitioning method by ref. 51 . First, we carry out projections using point estimates for all dose–response function parameters (abstracting from dose–response function uncertainty) and calculate internal variability as the average across CMIP6 models of each model’s variance of global GDP impacts in a given global warming-level window. For CMIP6 models with only a single run in our analysis, this within-model variance stems from climatic differences between different scenario–years, whereas for the two large ensembles, it also includes differences between ensemble members. Correspondingly, we calculate climate model uncertainty as the variance between the mean global GDP impact of CMIP6 models for a given global warming level. Lastly, we calculate dose–response function uncertainty as the variance across dose–response function Monte Carlo draws of the mean GDP impact for each Monte Carlo draw (that is, an average across all CMIP6 models and scenario–years in the respective global warming-level window). Mathematical expressions for each variance component are provided in Supplementary Appendix D . Notably, this approach rests on the commonly made assumption that variance drivers are orthogonal, thus abstracting from interaction terms 52 . As a robustness check, we use an alternative approach by ref. 3 (Supplementary Appendix D ), which does not affect our conclusions.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Data availability

CMIP6 temperature and precipitation indicators are available on the ETH Zurich CMIP6 repository 53 . CESM2 large ensemble outputs are available at https://www.earthsystemgrid.org/dataset/ucar.cgd.cesm2le.output.html . Tx5d and PDSI values from ref. 18 (used in Supplementary Appendix F ) are available at https://github.com/ccallahan45/CallahanMankin_ExtremeHeatEconomics_2022 (ref. 54 ). The historical climate data and the economic growth data to estimate the dose–response functions from ref. 14 are available from ref. 55 . ERA5 reanalysis data are available at https://www.ecmwf.int/en/forecasts/datasets/reanalysis-datasets/era5 . Source data are provided with this paper. All additional data are publicly available from ref. 56 .

Code availability

All scripts used to conduct the analysis and create the figures are publicly available from ref. 56 .

Dell, M., Jones, B. F. & Olken, B. A. Temperature shocks and economic growth: evidence from the last half century. Am. Econ. J. Macroecon. 4 , 66–95 (2012).

Article   Google Scholar  

Burke, M., Hsiang, S. M. & Miguel, E. Global non-linear effect of temperature on economic production. Nature 527 , 235–239 (2015).

Article   CAS   Google Scholar  

Hsiang, S. et al. Estimating economic damage from climate change in the United States. Science 356 , 1362–1369 (2017).

Kalkuhl, M. & Wenz, L. The impact of climate conditions on economic production: evidence from a global panel of regions. J. Environ. Econ. Manag. 103 , 102360 (2020).

Seneviratne, S. I. et al. in Climate Change 2021: The Physical Science Basis (eds Masson-Delmotte, V. et al.) 1513–1766 (Cambridge Univ. Press, 2021).

Newman, R. & Noy, I. The global costs of extreme weather that are attributable to climate change. Nat. Commun. 14 , 6103 (2023).

Pendergrass, A. G., Knutti, R., Lehner, F., Deser, C. & Sanderson, B. M. Precipitation variability increases in a warmer climate. Sci. Rep. 7 , 17966 (2017).

Masson-Delmotte, V. et al. (eds) Climate Change 2021: The Physical Science Basis (Cambridge Univ. Press, 2021).

Bathiany, S., Dakos, V., Scheffer, M. & Lenton, T. M. Climate models predict increasing temperature variability in poor countries. Sci. Adv. 4 , eaar5809 (2018).

Batibeniz, F., Hauser, M. & Seneviratne, S. I. Countries most exposed to individual and concurrent extremes and near-permanent extreme conditions at different global warming levels. Earth Syst. Dynam. 14 , 485–505 (2023).

Burke, M. et al. Opportunities for advances in climate change economics. Science 352 , 292–293 (2016).

Piontek, F. et al. Integrated perspective on translating biophysical to economic impacts of climate change. Nat. Clim. Change 11 , 563–572 (2021).

Damania, R., Desbureaux, S. & Zaveri, E. Does rainfall matter for economic growth? Evidence from global sub-national data (1990–2014). J. Environ. Econ. Manag. 102 , 102335 (2020).

Kotz, M., Levermann, A. & Wenz, L. The effect of rainfall changes on economic production. Nature 601 , 223–227 (2022).

Palagi, E., Coronese, M., Lamperti, F. & Roventini, A. Climate change and the nonlinear impact of precipitation anomalies on income inequality. Proc. Natl Acad. Sci. USA 119 , e2203595119 (2022).

Kotz, M., Wenz, L., Stechemesser, A., Kalkuhl, M. & Levermann, A. Day-to-day temperature variability reduces economic growth. Nat. Clim. Change 11 , 319–325 (2021).

Linsenmeier, M. Temperature variability and long-run economic development. J. Environ. Econ. Manag. 121 , 102840 (2023).

Callahan, C. W. & Mankin, J. S. Globally unequal effect of extreme heat on economic growth. Sci. Adv. 8 , eadd3726 (2022).

Holtermann, L. Precipitation anomalies, economic production and the role of ‘first-nature’ and ‘second-nature’ geographies: a disaggregated analysis in high-income countries. Glob. Environ. Change 65 , 102167 (2020).

Kahn, M. E. et al. Long-term macroeconomic effects of climate change: a cross-country analysis. Energy Econ. 104 , 105624 (2021).

Newell, R. G., Prest, B. C. & Sexton, S. E. The GDP–temperature relationship: implications for climate change damages. J. Environ. Econ. Manag. 108 , 102445 (2021).

Orlowsky, B. & Seneviratne, S. I. Elusive drought: uncertainty in observed trends and short- and long-term CMIP5 projections. Hydrol. Earth Syst. Sci. 17 , 1765–1781 (2013).

Pfahl, S., O’Gorman, P. A. & Fischer, E. M. Understanding the regional pattern of projected future changes in extreme precipitation. Nat. Clim. Change 7 , 423–427 (2017).

Kotz, M., Wenz, L. & Levermann, A. Footprint of greenhouse forcing in daily temperature variability. Proc. Natl Acad. Sci. USA 118 , e2103294118 (2021).

Kotz, M., Lange, S., Wenz, L. & Levermann, A. Constraining the pattern and magnitude of projected extreme precipitation change in a multimodel ensemble. J. Clim. 37 , 97–111 (2024).

Moore, F. C. & Diaz, D. B. Temperature impacts on economic growth warrant stringent mitigation policy. Nat. Clim. Change 5 , 127–131 (2015).

Kikstra, J. S. et al. The social cost of carbon dioxide under climate–economy feedbacks and temperature variability. Environ. Res. Lett. 16 , 094037 (2021).

World Development Indicators: GDP (constant 2015 US$) (World Bank, accessed 3 May 2023); https://databank.worldbank.org/source/world-development-indicators

Pakistan Floods 2022: Post-Disaster Needs Assessment (Government of Pakistan, 2022); https://www.undp.org/pakistan/publications/pakistan-floods-2022-post-disaster-needs-assessment-pdna

Screen, J. A. Arctic amplification decreases temperature variance in northern mid- to high-latitudes. Nat. Clim. Change 4 , 577–582 (2014).

de Vries, I. E., Sippel, S., Pendergrass, A. G. & Knutti, R. Robust global detection of forced changes in mean and extreme precipitation despite observational disagreement on the magnitude of change. Earth Syst. Dynam. 14 , 81–100 (2023).

Auffhammer, M., Hsiang, S. M., Schlenker, W. & Sobel, A. Using weather data and climate model output in economic analyses of climate change. Rev. Environ. Econ. Policy 7 , 181–198 (2013).

Naumann, G., Cammalleri, C., Mentaschi, L. & Feyen, L. Increased economic drought impacts in Europe with anthropogenic warming. Nat. Clim. Change 11 , 485–491 (2021).

Bastien-Olvera, B. A., Granella, F. & Moore, F. C. Persistent effect of temperature on GDP identified from lower frequency temperature variability. Environ. Res. Lett. 17 , 084038 (2022).

Dell, M., Jones, B. F. & Olken, B. A. What do we learn from the weather? The new climate–economy literature. J. Econ. Lit. 52 , 740–798 (2014).

Auffhammer, M. Quantifying economic damages from climate change. J. Econ. Perspect. 32 , 33–52 (2018).

Byers, E. et al. Global exposure and vulnerability to multi-sector development and climate change hotspots. Environ. Res. Lett. 13 , 055012 (2018).

Dinku, T. in Extreme Hydrology and Climate Variability (eds Melesse, A. M. et al.) 71–80 (Elsevier, 2019).

Kornhuber, K. et al. Risks of synchronized low yields are underestimated in climate and crop model projections. Nat. Commun. 14 , 3528 (2023).

Schwarzwald, K., Goddard, L., Seager, R., Ting, M. & Marvel, K. Understanding CMIP6 biases in the representation of the Greater Horn of Africa long and short rains. Clim. Dynam. 61 , 1229–1255 (2023).

Ayugi, B. et al. Comparison of CMIP6 and CMIP5 models in simulating mean and extreme precipitation over East Africa. Int. J. Climatol. 41 , 6474–6496 (2021).

South, A. rnaturalearthdata: World vector map data from Natural Earth used in ’rnaturalearth’. R package version 0.1.0 https://github.com/ropenscilabs/rnaturalearthdata (2017).

Jones, P. First- and second-order conservative remapping schemes for grids in spherical coordinates. Mon. Weather Rev. 127 , 2204–2210 (1999).

Hausfather, Z., Marvel, K., Schmidt, G. A., Nielsen-Gammon, J. W. & Zelinka, M. Climate simulations: recognize the ‘hot model’ problem. Nature 605 , 26–29 (2022).

Zhang, X. et al. Indices for monitoring changes in extremes based on daily temperature and precipitation data. WIREs Clim. Change 2 , 851–870 (2011).

Zhang, X., Hegerl, G., Zwiers, F. W. & Kenyon, J. Avoiding inhomogeneity in percentile-based indices of temperature extremes. J. Clim. 18 , 1641–1651 (2005).

Tabor, K. & Williams, J. W. Globally downscaled climate projections for assessing the conservation impacts of climate change. Ecol. Appl. 20 , 554–565 (2010).

Wenz, L., Carr, R. D., Kögel, N., Kotz, M. & Kalkuhl, M. DOSE—global data set of reported sub-national economic output. Sci. Data 10 , 425 (2023).

Murakami, D., Yoshida, T. & Yamagata, Y. Gridded GDP projections compatible with the five SSPs (shared socioeconomic pathways). Front. Built Environ. 7 , 760306 (2021).

Schwarzwald, K. & Lenssen, N. The importance of internal climate variability in climate impact projections. Proc. Natl Acad. Sci. USA 119 , e2208095119 (2022).

Hawkins, E. & Sutton, R. The potential to narrow uncertainty in regional climate predictions. Bull. Am. Meteorol. Soc. 90 , 1095–1108 (2009).

Lehner, F. et al. Partitioning climate projection uncertainty with multiple large ensembles and CMIP5/6. Earth Syst. Dynam. 11 , 491–508 (2020).

Brunner, L., Hauser, M., Lorenz, R. & Beyerle, U. The ETH Zurich CMIP6 next generation archive: technical documentation. Zenodo https://zenodo.org/doi/10.5281/zenodo.3734127 (2020).

Callahan, C. W. & Mankin, J. S. Globally unequal effect of extreme heat on economic growth. GitHub https://github.com/ccallahan45/CallahanMankin_ExtremeHeatEconomics_2022 (2022).

Kotz, M., Levermann, A., & Wenz, L. Data and code for the publication ‘The effect of rainfall changes on economic production’. Zenodo https://zenodo.org/doi/10.5281/zenodo.5657456 (2021).

Waidelich, P., Batibeniz, F., Rising, J., Kikstra, J. S., & Seneviratne, S. Scripts and data for ‘Climate damage projections beyond annual temperature’. Zenodo https://zenodo.org/doi/10.5281/zenodo.10465253 (2024).

Download references

Acknowledgements

We thank M. Hauser, J. Krug and I. de Vries, as well as participants at the NAVIGATE-ENGAGE Summer School 2023 and the 16th IAMC Annual Meeting 2023 for valuable comments. This work was supported by the European Union’s Horizon 2020 research and innovation programme, European Research Council under grant agreement no. 948220, project no. GREENFIN (P.W.) and by the Natural Environment Research Council under grant agreement no. NE/S007415/1 (J.S.K.).

Open access funding provided by Swiss Federal Institute of Technology Zurich.

Author information

Authors and affiliations.

Climate Finance and Policy Group, ETH Zurich, Zurich, Switzerland

Paul Waidelich

Institute for Atmospheric and Climate Science, ETH Zurich, Zurich, Switzerland

Fulden Batibeniz & Sonia I. Seneviratne

Oeschger Center for Climate Change Research, University of Bern, Bern, Switzerland

Fulden Batibeniz

Climate and Environmental Physics, Physics Institute, University of Bern, Bern, Switzerland

School of Marine Science & Policy, University of Delaware, Newark, DE, US

James Rising

Centre for Environmental Policy, Imperial College London, London, UK

Jarmo S. Kikstra

Grantham Institute for Climate Change and the Environment, Imperial College London, London, UK

Energy, Climate and Environment (ECE) Program, International Institute for Applied Systems Analysis, Laxenburg, Austria

You can also search for this author in PubMed   Google Scholar

Contributions

All authors conceived the study. F.B. prepared the underlying climate data and conducted the bias correction. P.W. and J.R. performed the data aggregation and developed the projection methodology. P.W. carried out the impact projections, analysed and visualized the results and wrote the manuscript. All authors reviewed the manuscript.

Corresponding author

Correspondence to Paul Waidelich .

Ethics declarations

Competing interests.

The authors declare no competing interests.

Peer review

Peer review information.

Nature Climate Change thanks Marshall Burke and the other, anonymous, reviewer(s) for their contribution to the peer review of this work.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary information

Supplementary information.

Supplementary Discussion (Appendices A–F), Figs. 1–19 and Tables 1–10.

Reporting Summary

Source data fig. 1.

Statistical source data.

Source Data Fig. 2

Source data fig. 3, source data fig. 4, source data fig. 5, rights and permissions.

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/ .

Reprints and permissions

About this article

Cite this article.

Waidelich, P., Batibeniz, F., Rising, J. et al. Climate damage projections beyond annual temperature. Nat. Clim. Chang. (2024). https://doi.org/10.1038/s41558-024-01990-8

Download citation

Received : 24 July 2023

Accepted : 20 March 2024

Published : 17 April 2024

DOI : https://doi.org/10.1038/s41558-024-01990-8

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

Quick links

• Explore articles by subject
• Guide to authors
• Editorial policies

Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

An official website of the United States government

The .gov means it’s official. Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

The site is secure. The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

• Publications
• Account settings

Preview improvements coming to the PMC website in October 2024. Learn More or Try it out now .

• Advanced Search
• Journal List
• Wiley-Blackwell Online Open

Climate variability and vulnerability to climate change: a review

Philip k thornton.

1 CGIAR Research Programme on Climate Change, Agriculture and Food Security (CCAFS), ILRI, PO Box 30709, Nairobi, 00100, Kenya

Polly J Ericksen

2 International Livestock Research Institute (ILRI), PO Box 30709, Nairobi, Kenya

Mario Herrero

3 Commonwealth Scientific and Industrial Research Organisation, 306 Carmody Road, St Lucia, Queensland, 4067, Australia

Andrew J Challinor

4 School of Earth and Environment, The University of Leeds, Leeds, LS2 9AT, UK

The focus of the great majority of climate change impact studies is on changes in mean climate. In terms of climate model output, these changes are more robust than changes in climate variability. By concentrating on changes in climate means, the full impacts of climate change on biological and human systems are probably being seriously underestimated. Here, we briefly review the possible impacts of changes in climate variability and the frequency of extreme events on biological and food systems, with a focus on the developing world. We present new analysis that tentatively links increases in climate variability with increasing food insecurity in the future. We consider the ways in which people deal with climate variability and extremes and how they may adapt in the future. Key knowledge and data gaps are highlighted. These include the timing and interactions of different climatic stresses on plant growth and development, particularly at higher temperatures, and the impacts on crops, livestock and farming systems of changes in climate variability and extreme events on pest-weed-disease complexes. We highlight the need to reframe research questions in such a way that they can provide decision makers throughout the food system with actionable answers, and the need for investment in climate and environmental monitoring. Improved understanding of the full range of impacts of climate change on biological and food systems is a critical step in being able to address effectively the effects of climate variability and extreme events on human vulnerability and food security, particularly in agriculturally based developing countries facing the challenge of having to feed rapidly growing populations in the coming decades.

Introduction

Climate change has many elements, affecting biological and human systems in different ways. The considerable spatial heterogeneity of climate change impacts has been widely studied; global average temperature increases mask considerable differences in temperature rise between land and sea and between high latitudes and low; precipitation increases are very likely in high latitudes, while decreases are likely in most of the tropics and subtropical land regions ( IPCC, 2007 ). It is widely projected that as the planet warms, climate and weather variability will increase. Changes in the frequency and severity of extreme climate events and in the variability of weather patterns will have significant consequences for human and natural systems. Increasing frequencies of heat stress, drought and flooding events are projected for the rest of this century, and these are expected to have many adverse effects over and above the impacts due to changes in mean variables alone ( IPCC, 2012 ).

In this review, we consider the possible impacts of changes in climate variability on biological and food systems, with a focus on the tropical and subtropical developing world, where the deleterious impacts of anthropogenic climate change are generally projected to be greatest. These less developed regions of the world already face an enormous food security challenge, with human populations rising unabated throughout the present century ( UNDESA, 2013 ). We start with a short consideration of the global importance and costs of climate variability and extreme events. We then briefly review some of the major impacts of climate variability and extremes on biological and agricultural systems at a range of scales, and on human health and nutrition. We then present some new analysis that seeks to link increases in climate variability with increasing food insecurity in the future, before considering the ways in which people deal with climate variability and extremes and how they may adapt in the coming decades. We conclude with a discussion of research gaps in relation to both the biophysical and the socioeconomic arenas and what needs to be done to better understand the impacts of climate variability on human vulnerability and food security, ultimately to increase the capacity of farmers in the tropics and subtropics to address climate variability and extreme events.

Climate change, climate variability and extreme events

Climate change is inevitably resulting in changes in climate variability and in the frequency, intensity, spatial extent, duration, and timing of extreme weather and climate events ( IPCC, 2012 ). Changes in climate variability and extremes can be visualized in relation to changes in probability distributions, shown in Fig. ​ Fig.1 1 ( IPCC, 2012 ). The top panel shows a shift of the entire distribution towards a warmer climate (a change in the mean), a situation in which more hot (and record hot) weather would be expected, along with less cold (and record cold) weather. The middle panel shows a change in the probability distribution of temperature that preserves the mean value, but involves an increase in the variance of the distribution: on average, the temperature is the same, but in the future, there would be more hot and cold (and record hot and cold) weather. The bottom panel shows the situation in which the temperature probability distribution preserves its mean, but the variability evolves through a change in asymmetry towards the hotter part of the distribution; here, we would see near constant cold (and record cold) weather, but increases in hot (and record hot weather).

The effect of changes in temperature distribution on extremes. Different changes of temperature distributions between present and future climate and their effects on extreme values of the distributions: (a) Effects of a simple shift of the entire distribution towards a warmer climate; (b) effects of an increase in temperature variability with no shift of the mean; (c) effects of an altered shape of the distribution, in this example a change in asymmetry towards the hotter part of the distribution. From IPCC (2012) .

Climate variability already has substantial impacts on biological systems and on the smallholders, communities and countries which depend on them. The importance of rainfall variability at the national level, for example, is illustrated in Fig. ​ Fig.2, 2 , which shows the relationship between annual rainfall variability and changes in the gross domestic product and agricultural gross domestic product for three countries of sub-Saharan Africa. In Fig. ​ Fig.2, 2 , interannual rainfall variability is expressed as the 12-month Weighted Anomaly of Standardized Precipitation (WASP), calculated from overlapping multimonth sums of standardized precipitation anomalies weighted according to the fraction of mean annual precipitation at the given time of year (from the data library of the International Research Institute for Climate and Society, iridl.ldeo.columbia.edu). This kind of close relationship is likely to be found for many tropical countries that depend heavily on agriculture as an engine for economic development.

The relationship between rainfall variability expressed as the 12-month Weighted Anomaly of Standardized Precipitation (WASP) and growth in GDP and agricultural GDP in three countries in sub-Saharan Africa: (a) Ethiopia, (b) Niger, (c) Mozambique. Data sources: World Bank, data.worldbank.org/indicator and the IRI data library, iridl.ldeo.columbia.edu/.

Changes in extremes have been observed since 1950, and there is evidence that some of these changes are a result of anthropogenic influences, although attribution of single extreme events to these influences remains challenging ( IPCC, 2012 ). Global aridity has increased substantially since the 1970s due to recent drying over Africa, southern Europe, East and South Asia, and eastern Australia – the percentage of global land (between 60°S and 75°N) defined as dry areas has increased from 17% in the 1950s to about 27% in the 2000s ( Dai, 2011 ). There is considerable uncertainty regarding projected changes in extremes to the end of the current century, and confidence in projecting changes in the direction and magnitude of climate extremes is generally low, although as the IPCC (2012) points out, low confidence in projections of changes in extremes does not mean that such changes are necessarily unlikely. Similarly, given current limits of understanding of the underlying processes regarding climate in many regions, it may be that low-probability, high-impact changes in extremes will occur. A partial summary of observed changes in some extremes, their attribution and their future projection, is shown in Table ​ Table1, 1 , extracted from table 3.1 in IPCC (2012) .

Summary of observed and projected changes of five extremes at a global scale (taken from table 3.1 in IPCC, 2012 )

A summary analysis of the numbers of people affected by environmentally related disasters is given in Raleigh & Jordan (2010) based on data compiled by the Centre for Research on the Epidemiology of Disasters ( CRED, 2008 ). A disaster is entered into the CRED database if at least one of the following criteria is fulfilled: 10 or more people reported killed, 100 or more people reported affected, a declaration of a state of emergency and a call issued for international assistance. An aggregated summary of these data is shown in Table ​ Table2. 2 . Chronic environmental hazards such as drought are not the most common, but they do affect the most people, with impacts on an average across all years of 10 per cent of a country's population (in low-income states, this increases to 13 per cent of a country's population). Raleigh & Jordan (2010) note that only in the case of drought is a significant proportion of a state affected. Floods tend to be more localized (for obvious reasons), but may still affect millions of people. The total number of disaster events in each region since 1970 is particularly noteworthy; and since 2000, the average number of events per year is running at more than 380 ( Raleigh & Jordan, 2010 ).

Population affected by selected disasters (aggregated from Raleigh & Jordan, 2010 )

There is a considerable literature on the economic costs of climate variability and extremes. Globally, annual damage from large weather and climate events increased eightfold between the 1960s and the 1990s; between 1980 and 2004, the costs of extreme weather events amounted to US 1.4 trillion ( Mills, 2005 ). Since 1980, annual costs have ranged from a few US billion to above US $200 billion (in 2010 dollars) for 2005, the year of Hurricane Katrina ( IPCC, 2012 ). While there is considerable regional variation, the relative economic burden of climate extremes as a proportion of GDP is substantially higher in developing countries than it is in developed countries – up to 8% in the most extreme cases. A strong upward trend in overall losses due to climate extremes is indicated since 1980 ( Munich Re, 2011 ), although how these will play out during the course of the current century is highly uncertain; and as yet, there is no evidence to link this trend to anthropogenic climate change ( Bouwer, 2011 ). Extreme events may have considerable impacts on sectors that have close links with climate, such as water, agriculture and food security, forestry, health, and tourism, and concomitantly in countries whose economies depend more heavily on such sectors ( IPCC, 2012 ).

Impacts of climate variability and extremes

Biological systems.

Warmer climates will generally accelerate the growth and development of plants, but overly cool or hot weather will also affect productivity. Earlier flowering and maturity of several crops have been documented in recent decades, often associated with higher temperatures ( Craufurd & Wheeler, 2009 ). Increases in maximum temperatures (as climate or weather) can lead to severe yield reductions and reproductive failure in many crops. In maize, each degree day spent above 30 °C can reduce yield by 1.7% under drought conditions ( Lobell et al. , 2011 ). Impacts of temperature extremes may also be felt at night, with rice yields reduced by 90% with night temperatures of 32 compared with 27 °C ( Mohammed & Tarpley, 2009 ). In contrast to the effects of temperature and photoperiod at optimum and suboptimum temperatures, crop response to temperature and photoperiod at supraoptimal temperatures is not well understood ( Craufurd & Wheeler, 2009 ).

Climate variability and extreme events can also be important for yield quality. Protein content of wheat grain has been shown to respond to changes in the mean and variability of temperature and rainfall ( Porter & Semenov, 2005 ); specifically, high-temperature extremes during grain filling can affect the protein content of wheat grain ( Hurkman et al. , 2009 ).

At aggregated level as well as at the plot level, rainfall variability is a principal cause of interannual yield variability. For example, Hlavinka et al . (2009) found a statistically significant correlation between a monthly drought index and district-level yields in the Czech Republic for several winter- and spring-sown crops, each of which has a different sensitivity to drought. Both intra- and interseasonal changes in temperature and precipitation have been shown to influence cereal yields in Tanzania ( Rowhani et al. , 2011 ). The increases in rainfall variability expected in the future will have substantial impacts on primary productivity and on the ecosystem provisioning services provided by forests and agroforestry systems. Despite the uncertainty surrounding the precise changes, climate variability needs to be taken into account. For example, the impacts of climate change to the middle of this century on crop yields in parts of East Africa may be underestimated by between 4% and 27%, depending on the crop, if only changes in climatic means are taken into account and climate variability is ignored ( Rowhani et al. , 2011 ).

Changes in temperature and rainfall patterns and amounts will combine to bring about shifts in the onset and length of growing seasons in the future. Projected changes in length of the growing period for Africa to the 2090s were estimated by Thornton et al . (2011) for an ensemble of 14 GCMs. A large proportion of the cropping and rangeland area of sub-Saharan Africa is projected to see a decrease in growing season length, and most of Africa in the southern latitudes may see losses of at least 20 per cent. At the same time, the probability of season failure is projected to increase for all of sub-Saharan Africa, except for central Africa; in southern Africa, nearly all rain-fed agriculture below latitude 15°S is likely to fail one year out of two ( Thornton et al. , 2011 ). The robustness of these estimates, in terms of intramodel variability, is particularly low in the Sahel region and in parts of south-western Africa, however ( Thornton et al. , 2011 ). In terms of timing of growing season onset, Crespo et al . (2011) demonstrate that it may be possible to adapt to projected climate shifts to at least the 2050s in maize production systems in parts of southern Africa by changing planting dates.

In situations where changes in climate and climate variability may be larger, more fundamental changes may occur, particularly if critical thresholds in temperature and/or rainfall are reached ( Gornall et al. , 2010 ). Changes in the nature and timing of the growing season may induce smallholders to grow shorter duration and/or more heat- and drought-tolerant varieties and crops, for example.

Most domesticated livestock species have comfort zones between 10 and 30 °C; at temperatures below this, maintenance requirements for food may increase by up to 50%, and at temperatures above this, animals reduce their feed intake 3–5% per additional degree of temperature ( NRC, 1981 ). In many livestock systems, changes in temperature and rainfall and rainfall variability affect feed quantity most directly. Droughts and extreme rainfall variability can trigger periods of severe feed scarcity, especially in dryland areas, which can have devastating effects on livestock populations. In the recent past, the pastoral lands of East Africa have experienced droughts about one year in five, and even under these conditions it is generally possible to maintain relatively constant cattle herd sizes, but increases in drought frequency from one year in five to one year in three would set herd sizes on a rapid and unrecoverable decline ( Thornton & Herrero, 2009 ). In Kenya, some 1.8 million extra cattle could be lost by 2030 because of increased drought frequency, the value of the lost animals and production foregone amounting to US $630 million ( Ericksen et al. , 2012 ).

Craine et al . (2012) found that in a temperate grassland, the effects of drought and high temperatures decline over the season, to the point where climate variability may have minimal impact later in the growing season. Key ecosystem processes are seasonally sensitive to climate variability, and increased understanding of plant productivity will need to recognize that the timing of climate variability may be just as important as its magnitude ( Craine et al. , 2012 ). In both temperate and tropical grasslands, species composition is a key determinant of livestock productivity. As temperature, rainfall patterns and CO 2 levels change, so will the composition of mixed grasslands change. Small climatic changes may affect the dynamics and balance of different grasslands species, and these may result in changes in livestock productivity ( IPCC, 2007 ). The overall effects of changes in temperature and rainfall and their variability on species composition and grassland quality, however, are still far from clear and remain to be elucidated ( IPCC, 2007 ).

Droughts in grasslands can also be a predisposing factor for fire occurrence in many regions ( IPCC, 2012 ), and intensified droughts could exacerbate the problem. There is some evidence that recent years have already seen an increase in grassland fire disasters in parts of China and tropical Asia. In the future, wildfires may be 60 per cent more frequent in much of South America for a temperature increase of 3 °C, and in parts of Australia, the frequency of very high and extreme fire danger days could rise by up to 70 per cent by 2050 ( IPCC, 2012 ).

Mixed crop–livestock systems are prevalent in much of the developing world ( Herrero et al. , 2010 ), and climate change and changing climate variability in the future may affect the relationship between crops and livestock in the landscape in many places. In places that will become increasingly marginal for crop production, livestock may provide an alternative to cropping. Such transitions could occur in up to 3% of the total area of Africa, largely as a result of increases in the probability of season failure in the drier mixed crop–livestock systems of the continent; these are projected to increase from the current rate of approximately one year in five to one year in four or three, depending on the combination of emissions scenario and climate model used ( Jones & Thornton, 2009 ).

Changes in climate variability and in the frequency of extreme events may have substantial impacts on the prevalence and distribution of pests, weeds, and crop and livestock diseases.

For example, in the past, combinations of drought followed by high rainfall have led to wide-spread outbreaks of diseases such as Rift Valley fever and bluetongue in East Africa and of African horse sickness in South Africa ( Baylis & Githeko, 2006 ). Future increases in the frequency of extreme weather events could allow the expansion of Rift Valley fever northwards into Europe, for example ( Martin et al. , 2008 ). In general, the effects of future changes in climate variability on pests, weeds and diseases are not well understood ( Gornall et al. , 2010 ).

Evidence of vegetation shifts resulting from increasingly frequent extreme climatic events is still comparatively rare, although what there is supports the existence of stabilizing processes which tend to minimize and counteract the effects of these events, reinforcing community resilience ( Lloret et al. , 2012 ). Better understanding of these stabilizing processes and the community inertia that is frequently observed in vegetation under extreme events, are crucial for the establishment of sound management strategies that can improve ecosystem resilience under climate change ( Lloret et al. , 2012 ).

Globally, the negative effects of climate change on freshwater systems, in terms of changes in quantity and distribution, are expected to outweigh the benefits of overall increases in global precipitation due to a warming planet; several parts of the tropics and subtropics, including parts of Central-West Asia, North Africa, Asia and North America, are likely to be particularly affected by reduced freshwater availability ( Rosegrant et al. , 2009 ). It is expected that more than half the world's population will live in countries with severe water constraints by 2050 ( Rockström et al ., 2009 ).

Climate models project increased aridity during the current century over most of Africa, southern Europe and the Middle East, most of the Americas, Australia and Southeast Asia. There is considerable uncertainty in such results, but the projections are alarming because a very large population may be severely affected in the coming decades. At the same time, precipitation may become more intense but less frequent (i.e. longer dry spells). This has the potential to increase flash floods and runoff, and as a result increase soil erosion, diminish soil moisture and increase the risk of agricultural drought ( Dai, 2011 ), as well as increasing the potential for crop losses due to flooding and affecting the dynamics of livestock diseases and their vectors, for example.

Food systems, health and nutrition

There is little literature on the effects of climate variability and extreme climatic events on food systems as opposed to food production. Of nearly 600 pages in the SREX report ( IPCC, 2012 ), for example, there is only one page on the impacts of climate extremes on food systems and food security.

At the local level, Codjoe & Owusu (2011) studied communities in Ghana and showed how extreme climatic events affect rural food production, transportation, processing and storage. Food security in this region could be enhanced by increasing farm-based storage facilities; improving the transportation system, especially feeder roads that link food production areas and major markets; providing farmers with early warning systems; extending credit to farmers; and the use of supplementary irrigation. Some cultural practices, particularly those that prohibit the consumption of certain foods, may reduce the resilience of some individuals and ethnic groups to food system disruptions.

Climate variability has both direct and indirect impacts on human health. Extreme heat affects health, especially among the elderly ( McMichael et al. , 2006 ). Other direct impacts are largely expressed through the interaction of infectious and vector-borne diseases with temperature and precipitation. Malaria, dengue and cholera, for example, are all highly affected by changes in seasonal distribution of precipitation, including changes in flood and drought patterns ( McMichael & Kovats, 2000 ; Costello et al. , 2009 ). Although changes in malaria vectors will occur due to the gradual increase in temperature, the incidence of disease is also quite sensitive to changes in precipitation. If changes in climate variability lead to changes in spatial and temporal variation in vegetation and water distribution, we could see more epidemics as the vector moves to new areas ( McMichael et al. , 2006 ). Both malaria and dengue fever have associations with La Niña and El Niño cycles ( McMichael et al. , 2006 ). Human displacement from extreme events, especially floods, could become more frequent with an increase in climate variability. This also often has negative consequences for human health, not least because of crowded conditions with poor sanitation. Diarrhoeal disease is regularly a problem in such situations ( Haines et al. , 2006 ). In addition, as inadequate access to health services is already a leading cause of poor health in developing countries, displacement and infrastructure damage from extreme events, especially floods, can exacerbate this (although people also often move in response to prolonged drought). If water scarcity increases, this also has an impact on sanitation and health outcomes if clean water is less available ( Few, 2007 ).

Nutrition is correlated with positive health outcomes, and both adequate amount of calories as well as sufficient nutritional diversity and proteins are important. As outlined above, overall availability of food shows some correlation with climate variability. The study by Lloyd et al . (2011) builds upon previous work of Nelson (2009) to show clearly that climate change and increased climate variability, through their impact on food production, will have a negative impact on the prevalence of undernutrition, increasing severe stunting by 62% in South Asia and 55% in East and southern Africa by the 2050s. Although nutrition is determined not only by food availability, but also access to food as well as nutritional and child care practices, there are almost no studies on these other aspects of nutrition determinants ( Tirado et al. , 2010 ).

Some more detailed work has been done at national level. For example, a dynamic economy-wide model of Bangladesh has been used to estimate economic damages from historical climate variability and future anthropogenic climate change. Using a combination of historical yield variability and 10 climate projections, future anthropogenic climate change damages are estimated to reduce national rice production in Bangladesh by about 9 per cent to midcentury, and most of these losses are attributed in the analysis to flooding damage and climate variability ( Thurlow et al. , 2011 ). Another example is the work of Ahmed et al . (2011) , who used a modelling approach to estimate how changes in climate variability might affect crop yields and thence poverty rates in Tanzania to the early 2030s. They found that future climate scenarios with the largest increases in climate volatility rendered Tanzanians increasingly vulnerable to poverty through its impact on the production of staple grains.

At the global level, one of the few studies so far to model climate shocks and their impacts on commodity prices in different regions is Willenbockel (2012) . Results are indicative only but interesting nevertheless. For example, a drought in North America in 2030 of a similar scale to the historical drought of 1988 would have a dramatic temporary impact on world market export prices for maize and a strong impact on world market price for wheat. These impacts would feed through to domestic consumer prices, with particularly profound effects in parts of sub-Saharan Africa. For instance, Nigeria depends almost entirely on imports of wheat, and under such a scenario the average domestic price for wheat in the country would spike by 50% above the baseline 2030 price, with potentially substantial impacts on households. The treatment of the impacts of climate variability as opposed to the impacts of slow-onset climate change in global economic models is a heavily underresearched area, particularly how harvest failures in one continent may influence food security outcomes in others.

How may changes in climate variability and extremes affect food security in the future?

Human populations are vulnerable to the impacts of climate change largely because of the socioeconomic and political context in which they live. Thus, vulnerability to climate change is highly differentiated (O' Brien et al. , 2007 ) across geography, income levels, type of livelihood and governance arrangements, among other things. Human vulnerability can be evaluated in terms of a range of different outcomes such as food security or household income. Thus, areas vulnerable to disasters are not necessarily the same as those whose food availability is likely to be negatively affected by changes in climate variability. A major challenge in viewing human vulnerability as the result of multiple and dynamic factors is the need to take a synthetic approach to translate the sectoral impacts of changes in climate and climate variability into consequences for people. Food security is a particularly important developmental outcome that is highly vulnerable to climate change. This vulnerability is a product of climate change impacts on biological systems, affecting food availability, as well as economic and social impacts that affect food utilization, access to food and the stability of food security ( Ericksen, 2008 ).

As noted above, there is only limited information on the potential impacts of climate variability on food availability at broad scales such as national and regional. For economies that are agriculturally based, Fig. ​ Fig.2 2 suggests that rainfall variability can have substantial effects on agricultural growth at the national level, although that relationship will be modified by many other factors. Links from climate variability to poverty indicators are also not that straightforward to demonstrate. We undertook some new analysis using recent global data sets to try to throw some light on the possible links between climate variability and food security. Herrero et al . (2013) recently generated maps showing global kilocalorie production per capita from edible animal products, including milk and meat from ruminant species (bovines, sheep and goats) and meat and eggs from monogastric species (pigs and poultry). To estimate total kilocalorie production from crops, we used data on crop yields and harvested areas from the Spatial Production Allocation Model (SPAM) of You et al . (2012) . SPAM contains data for the year 2000 and includes 14 food crops or crop groups: banana and plantain, barley, beans, cassava, groundnut, maize, millet, other pulses such as chickpea, cowpeas, pigeon peas, and lentils), potato, rice, sorghum, soybean, sweet potato and yam, and wheat. We calculated the total food production from these 14 crops and crop groups using calorie contents as given in FAO (2001) . The SPAM data set matches FAOSTAT country totals for 2000 and details crops grown in three types of system (irrigated, rain-fed commercial and rain-fed subsistence). Multiple cropping is also taken into account. We then calculated total kilocalorie production from both livestock and the 14 crops at a resolution of 5 arcmin (gridcells of side about 9 km at the equator). Each grid cell was then stratified on the basis of rainfall variability. To do this, we utilized a weather generator, MarkSim, and methods outlined in Jones & Thornton (2013) to estimate the coefficient of variation (CV) of annual rainfall for the globe, from 100 years of generated daily rainfall data. We estimated the human population in each stratum (CIESIN Center for International Earth Science Information Network Columbia University & Centro Internacional de Agricultura Tropical (CIAT), 2005a ). To relate climate variability to some proxy of food insecurity, we used the subnational data set of CIESIN (2005b) on the proportion of children under five who are underweight for their age, and again estimated the average proportion for each stratum. The human population and children underweight data sets are both for the year 2000. Results are shown in Table ​ Table3, 3 , split between developing and developed countries. Here, we defined the developing countries as those in the Americas between Mexico in the north and Brazil, Paraguay, Bolivia and Peru in the south, all of Africa, and in Asia up to 45°N excluding Japan. The remainder, we classified as developed countries.

Proportion of total calorie availability per person per day from livestock products and from 14 food crops in developing and developed countries, by rainfall variability class

Several points can be made about Table ​ Table3. 3 . First, some 5.4 billion people, or just under 90 per cent of the global population in 2000, live in places that produce at least some crop and livestock calories. On the basis of this analysis, the 14 crops or crop groups account for 70 per cent of all calories produced and livestock 30 per cent (note that several important crops that provide calories for human nutrition are not included here, including sugar and oil crops). Second, it is noteworthy that developing countries (as defined above) account for 78 per cent of the people, but only 40 per cent of the calories available; conversely, the temperate regions account for 22 per cent of the people and 60 per cent of the calories produced. Third, the relationship between rainfall variability and the average prevalence of underweight children seems not to be straightforward: in the developed regions, the value of the food insecurity proxy increases as rainfall variability increases, whereas in developing countries, it increases up to a rainfall CV of 30 per cent and then falls slightly for further increases in rainfall CV. A possible explanation for this is that in the higher CV regions, most food is brought in via imports or food aid, for example. Fourth, nearly eight times as many people live in areas of high rainfall variability (with a CV of 30 per cent or more) in the developing countries as they do in the developed countries (407 million compared with 54 million); yet, these areas of high rainfall variability in developing countries account for only 3 per cent of all the calories produced, and they also tend to be areas with relatively high child malnutrition. Clearly, many such areas may be targets for the provision of food aid and social safety nets.

We can show that increased rainfall variability will affect agricultural growth and economic development in certain types of countries (Fig. ​ (Fig.2). 2 ). The analysis presented above is highly simplified, as there are many other factors and drivers that will interact in complex ways, but there may also be impacts of increased rainfall variability on food security as shown by a proxy such as the prevalence of child malnutrition (Table ​ (Table3). 3 ). In the absence of information concerning the nature of increases in rainfall variability in the coming decades, one question that might be asked is, how sensitive are the data in Table ​ Table3 3 to shifts in rainfall variability? To test this, we made several changes across the board to rainfall CV and then restratified the data. Results are shown in Fig. ​ Fig.3, 3 , in terms of population by rainfall CV, for the developing world and the developed world, for ‘current’ conditions and for situations with decreased (−1 per cent) and increased (+1 per cent and +2 per cent) rainfall variability. While the likelihood of such changes is essentially unknown, a + 2 percentage point increase in annual rainfall CV leads to increases in the population living in areas of high rainfall variability (CV> 30%) in developing countries of more than 230 million to 643 million people (58 per cent), while in the developed countries the number more than doubles from 54 to 112 million.

The differential impacts of across-the-board changes in rainfall CV of −1%, +1% and +2% on population distribution by rainfall variability in developing (a) and developed (b) countries.

It is not just rainfall that increases variability in yield. The temperature-related processes reviewed above also contribute to this. Few climate change impact studies report changes in CV; analysis of those that do shows that increases in CV of more than 50% may not be uncommon from the 2040s onwards ( Challinor et al. , 2014a ).

Even though simplified and with a high degree of uncertainty, our analysis helps to substantiate the hypothesis of an increase in child malnutrition rates in both developed and developing countries in the future as a result of variability changes, all other things being equal. These increases could be particularly large in sub-Saharan Africa as a result of high population growth rates and relatively large areas with high rainfall variability. Sub-Saharan Africa is already by far the largest recipient of food aid: average annual shipments amount to about 2 per cent of all food consumed. Under many scenarios, the number of food-insecure people in sub-Saharan Africa by 2020 is still likely to be at least 500 million ( USDA, 2010 ), and this is a challenge that will clearly not be made any easier by increases in rainfall and temperature variability.

Responses of vulnerable people

Most of the literature and analysis discussed above relate to how climate variability will affect exposure or sensitivity of biological and food systems, and hence food security outcomes. However, the most important element of reducing vulnerability is to enhance the adaptive capacity of people, at various levels of decision-making from the individual up to the national and regional. Institutions play a key role in enabling such adaptation. Increases in variability, which are largely unpredictable in the short and long term, will force institutions (defined loosely as social patterns including organizations) to be more proactive and flexible, qualities that are difficult to foster ( Gupta et al. , 2010 ). This applies not only to the national level; for example Ncube et al . (2012) discuss the example of local government infrastructure in South Africa to provide basic commodities such as water and energy. They found a strong relationship between water and energy demand and rainfall variability, and concluded that local governments will increasingly need to be proactive in planning for adaptation to climate change, because of its influence on their operations and budgets ( Ncube et al. , 2012 ).

By focusing on how climate variability might change, we are trying to better characterize what climate change means for vulnerability – a better answer to the question ‘Vulnerability to what’ ( Misselhorn et al. , 2010 ). Uncertainty or lack of predictability is considered a real hindrance to planning for adaptation. However, if for example we can explore how sensitive food availability in a given location is to a range of increases in precipitation variation, or what the limits of current institutional arrangements are for dealing with the consequences of increased frequency of extreme events, we can get a clearer definition of the development problem that climate change might exacerbate. This better prepares communities and governments to develop robust adaptation strategies in spite of uncertainty about the precise impacts of climate change.

What might vulnerable people who are partially or wholly dependent on natural resources for their livelihoods do in response to substantially increased climate variability? There is already a considerable literature on the ways in which resilience of agricultural production systems may be increased in the face of climate change, particular under the ‘climate-smart agriculture’ rubric (e.g. FAO, 2010 ; Thornton et al. , 2013 ). Options range from increasing the efficiency of crop and livestock systems via various components such as soil and nutrient management, water harvesting and retention, improving ecosystem management and biodiversity, diversification of on-farm activities, use of weather forecasts and early warning systems, and methods for managing risk such as index-based insurance and risk transfer products ( Barnett et al. , 2008 ). In relation to options for the drylands, the literature is not particularly sanguine. As many have pointed out, particularly in more marginal areas, farmers have already been substantially changing their practices. For example, farmers in northern Burkina Faso have adopted many techniques intended to increase crop yield and reduce yield variability ( Barbier et al. , 2009 ). The drivers of these shifts are not climate variability but growing land scarcity and new market opportunities. While improved water harvesting and storage techniques may be able to reduce farmers’ dependence on rainfall, they are not likely to be sufficient to significantly reduce vulnerability to drought ( Barbier et al. , 2009 ). Institutional change may be critical in enhancing resilience in dryland pastoral systems. In the Kalahari, land privatization policies have increased the vulnerability of poorer communal pastoralists, but increasing access to markets and improving the ability of these farmers to operate in a market economy could reduce their vulnerability ( Dougill et al. , 2010 ). At the same time, alternatives that make sense from the perspective of current economic risk or land scarcity, such as the use of higher yielding crop varieties or improved animal breeds, may not be robust choices for dealing with climate change if they do not outperform local varieties under highly variable conditions (see, for example, Rodríguez et al ., 2011 ). This underpins the importance of crop varieties with increased tolerance to heat and drought stress for managing future climatic variability ( Hellin et al. , 2012 ).

There are several ways in which the stability of food systems can be strengthened. These include governments investing in smallholder agricultural production, particularly in downstream activities such as storage, trace, processing and retailing; implementing and scaling up options that help producers to be more resilient to climate volatility, such as the now wide-spread use of smallholder crop insurance schemes in India and certain other countries; and establishing safety net programmes for the most vulnerable households, such as has been implemented successfully in Ethiopia ( Lipper, 2011 ). Insurance may be an increasingly important way to help smallholders become more resilient, in view of the impacts of climate change on yield variance and the resulting demand for effective risk-reducing measures. Using a microeconomic farm model, Antón et al . (2012) found that area yield and weather index insurance are robust policy options across different scenarios and are generally cheaper than individual yield insurance. They also found that ex post indemnity payments can be effective in dealing with extreme systemic risk situations and are similarly robust across different scenarios, even with frequent occurrence of extreme events, although they can be costly to implement ( Antón et al ., 2012 ).

One recurring thread in recent discussions concerning responses to increasing climate variability is the role of indigenous knowledge. Agro-pastoralists in dryland Kenya (and probably in many other places too) rely on indigenous indicators of rainfall variability and use them as a framework within which to position and interpret meteorological forecasts ( Rao et al. , 2011 ); at the same time, few are able to adapt their practices because of a general lack of adaptive capacity ( Speranza et al. , 2010 ). Integrating different types of knowledge and bringing different stakeholder groups together pose significant challenges, however, and considerable innovation in participatory action research will be needed ( Ziervogel & Opere, 2010 ). But there would seem to be a rich area of research in investigating the reliability and validity of indigenous knowledge concerning climate variability, and seeing how it can be better integrated into formal monitoring systems to enhance its acceptability, thereby increasing smallholders’ resilience to climate variability.

For some communities in marginal areas, climate may decreasingly be the primary concern. Nielsen & Reenberg (2010) present results from northern Burkina Faso that indicate that villagers there are ‘beyond climate’: current livelihood strategies are increasingly independent of climate. There as elsewhere, people have engaged in livelihood diversification in attempts to ameliorate the negative impacts of climate variability on agriculture. At some stage, tipping points are reached such that transformative adaptation alternatives may be the only viable options that remain. There are many examples of such changes to livelihood systems, such as substitution of one crop or livestock species for another. In many parts of sub-Saharan Africa, a highly spatially distributed mode of living is prevalent, and clearly it can be a highly effective way of dealing with change and variability. This is intriguingly mirrored in developed-country situations also, in Australian farming households over the last few years that have seen crippling, multi-year drought followed by record flooding, for example. Many such households are developing more spatially distributed modes of farming and living, whereby multiple priorities and pressures can be accommodated by moving between widely distributed farm businesses, employment and children's activities ( Rickards, 2012 ). Endurance and accommodating change may be widely valued, but others would challenge this world view and emphasize innovation and the conscious creation of innovative alternatives (O' Brien, 2012 ; Rickards, 2012 ). Many people may have no choice, and chronic or sudden-onset environmental disasters related to climate change may force large-scale migration; however, this is not expected to be common in the next two decades ( Raleigh & Jordan, 2010 ).

Conclusions: refining the research agenda

Most of the climate change impacts work carried out to date either ignores or downplays variability. On the one hand, this is somewhat understandable. Regarding expected changes in rainfall and temperature variability in the future, there is high uncertainty: IPCC (2012) provides no assessment of projected changes in extremes at spatial scales smaller than for large regions. Indeed, the prognosis for robust quantification in the foreseeable future of changes in weather and climate variability over short temporal and high spatial scales is rather gloomy (Ramirez- Villegas et al. , 2013 ). But on the other hand, we already know a reasonable amount about how current levels of climate variability have considerable impacts on biological systems and health. While we cannot let limited predictive capability constrain adaptive responses, it does suggest that we will need to become increasingly creative to arrive at actionable answers in response to questions from a wide range of decision makers concerning the appropriate adaptation of biological and food systems. One example of a suitable framework is the combination of impact and capacity approaches (broadly, top down and bottom up respectively) to adaptation planning; there is considerable potential in this and other problem-orientated approaches for producing robust knowledge and actions in the face of uncertainty ( Vermeulen et al. , 2013 ). This is not without its challenges, however: recent assessments indicate an increased probability of future tipping events, in part because of positive feedbacks in the climate system (e.g. Cory et al. , 2013 ), and the corresponding impacts are estimated to be large, making them significant risks ( Lenton, 2011 ). Below, we briefly discuss five areas that warrant considerable attention if we are to address these challenges.

First, there are still important knowledge and data gaps in our understanding of the effects of climate variability and extreme events on biological systems. With regard to crops, Craufurd & Wheeler (2009) identified several areas, including the need for more information on crop development and temperature by photoperiod interactions at the higher end of the temperature scale. There are key knowledge gaps with regard to the ways in which climate variability and extreme events may exacerbate multiple stresses for animals and plants, and how these stresses may interact and combine. There are also important knowledge gaps regarding the impacts of climate variability and extreme events on the prevalence, incidence and severity of crop and livestock diseases, and on key agricultural pests and weeds and how their prevalence may change.

Second, there are substantial limitations in our impact models, at all scales. This certainly applies to models of crops and livestock and on the effects of variability on the quantity and quality of crop and livestock production. Identification of synergies between global, regional and local studies is a promising avenue for improvement ( Challinor et al. , 2014b ). Much work is needed on extending the applicability of current crop and livestock models to the higher temperature and more variable climates projected as increasingly likely under higher greenhouse gas emission scenarios. Equally importantly, such gaps exist in relation to models of farming systems and the ways in which biophysical and socioeconomic drivers of change combine in particular situations ( Challinor et al. , 2009 ), and information concerning the way in which climate variability and climate extremes may affect thresholds and tipping points among different farm enterprises in relation to different household objectives is largely missing. Gaps also exist concerning the appropriate incorporation of risk and dynamics in farming system models. For smallholders, higher risks usually imply more costs, directly or indirectly, and so there is a need to link risk to decision-making profiles of farmers and their attitudes to investments and technology adoption. Some work has been done on this (see, for example, Willock et al. , 1999 ; Solano et al. , 2000 ), but more in-depth studies on this topic are needed, because increasing adoption rates of key practices under risk is a significant challenge, and targeting options to risk management profiles is essential. At the national and global scales, more sophisticated output is needed from global and regional economic models concerning welfare gains and losses arising from different policy action, and how changes in welfare from gradual climate change and climate shocks are differentially distributed among different groups in society, such as producers and urban poor, and men and women ( Skoufias et al. , 2011 ).

Third, there is a great need to improve the monitoring of local conditions, not only to provide data and information for improving our understanding and our models, but also to guide effective adaptation (for example, through downscaling climate model output to local situations) and to provide information for yield early warning systems and locally appropriate indices for weather-based crop and livestock insurance schemes. The situation for climate and weather data monitoring in many developing countries is poor and deteriorating. There is considerable research activity in combining satellite and land-based information to produce long-term, high-resolution weather data sets (for example, Maidment et al. , 2013 ). Such hybrid data sets have considerable potential to ease the weather data problem in some countries, but they are not a replacement for land-based weather measurement, however, and considerable investment will be needed to improve climate and weather monitoring. Improved monitoring of local food systems (in relation to food production and accessibility, for example) and of the environment (in relation to local crop and rangeland conditions, for example) is also needed to provide readily actionable information. The tradition of monitoring and surveillance for disease outbreaks within the health community, to allow for better early warning and anticipatory response in relation to food systems, is a promising model, although it can be costly.

Fourth, enhancing food security for the 9.5 billion people projected by 2050, more than 86% of whom will be living in the less developed countries ( UNDESA, 2013 ), will mean adapting biological and food systems to the increasingly variable climate and to increasingly frequent extreme events, which in turn will entail considerably enhanced understanding of the complex system of production, logistics, utilization of the produce and the socioeconomic structure of communities ( IPCC, 2012 ). This strongly supports the notion of viewing adaptation and vulnerability reduction not as discrete events but as processes through time, from the shorter term to the longer term. The impacts of climate variability and extreme events are often most acutely experienced at the local level ( IPCC, 2012 ), and they also usually occur over short time scales. At local and short temporal scales, the uncertainties associated with their prediction may be at their largest. Food security, health and nutritional outcomes are all the product of multiple interacting stressors, not just climate patterns. This could be one of the reasons the disaster relief community and the agricultural research for development community have not talked much together – the former has a ‘variability’ orientation, the latter a ‘changing means’ orientation. There are exceptions – for example, the story of weather forecasts for emergency aid provision in West Africa in 2008 ( Tall et al. , 2012 ) – but there do not seem to be many to date. There are surely synergies to be explored between these two communities of practice, particularly given rapid developments in the field of seamless prediction of weather and climate ( Brown et al. , 2012 ; Meehl et al. , 2013 ). In time, seamless prediction may provide a bridge between the shorter term (days, weeks, season) and the longer term (years, decades) and between risk management and adaptation planning. Using models to express uncertainty as the time intervals in which key changes are expected, rather than focussing on a particular time and expressing uncertainty in other ways, may help forge stronger links between prediction and adaptation ( Vermeulen et al. , 2013 ). The effectiveness of the links between different spatial and temporal scales will depend on enhanced understanding, models and monitoring of the impacts of climate change and climate variability on both biological and socioeconomic systems and the ways in which they interact within and across scales. Enhancing food security in the less developed countries in the coming decades will need balanced, integrated approaches that encompass changes in variability and extreme events as well as changes in means in quantifying impacts on, and identifying appropriate adaptation of, biological and human systems.

Finally, greater and more effective communication is needed between scientists and decision makers, and between natural and social scientists. Currently, climate information is severely underutilized in supporting decision-making, which Weaver et al . (2013) partially attribute to a failure to incorporate learning from the decision and social sciences into climate-related decision support. There is a great deal that can be done on the cogeneration of information and its communication in appropriate ways, and in engaging meaningfully with decision makers at local and national policy levels, for example. Participatory scenario development may be one useful tool for facilitating some of these processes ( Vervoort et al. , 2014 ), in addition to much stronger links between biological and communications scientists. In general, the top-down and bottom-up approaches identified above rarely meet in the form of integrated analyses. Given what is known about vulnerability to climate, what foci should environmental scientists have? Changes in variability are often more important for communities than changes in mean quantities; yet, the focus of modelling studies is often on the latter. The ongoing focus on quantifying uncertainty in impacts studies is important if we are to avoid errors; however, these analyses can be targeted more clearly at adaptation ( Challinor et al. , 2013 ; Vermeulen et al. , 2013 ). Systematic intercomparison of impacts studies, with coordinated cycles of model improvement and projection, is useful in reducing uncertainty and synthesizing knowledge ( Challinor et al. , 2014b ). Observational data to constrain models at a range of scales are central to these endeavours.

Acknowledgments

We thank two anonymous reviewers for helpful comments on an earlier draft. PKT thanks Michael Bell for help with WASP data. The CGIAR Research Program on Climate Change, Agriculture and Food Security (CCAFS) is funded by the CGIAR Fund, AusAid, Danish International Development Agency, Environment Canada, Instituto de Investigação Científica Tropical, Irish Aid, Netherlands Ministry of Foreign Affairs, Swiss Agency for Development and Cooperation, Government of Russia, UK Aid and the European Union, with technical support from the International Fund for Agricultural Development.

• Ahmed SA, Diffenbaugh NS, Hertel TW, Lobell DB, Ramankutty N, Rios AR, Rowhani P. Climate volatility and poverty vulnerability in Tanzania. Global Environmental Change. 2011; 21 :46–55. [ Google Scholar ]
• Antón J, Kimura S, Lankoski J, Cattaneo A. A comparative study of risk management in agriculture under climate change; Agriculture and Fisheries Working Papers, No. 58; OECD Food; 2012. OECD Publishing. Available at: http://dx.doi.org/10.1787/5k94d6fx5bd8-en (accessed 4 February 2013) [ Google Scholar ]
• Barbier B, Yacouba H, Karambiri H, Zorome M, Some B. Human vulnerability to climate variability in the Sahel: farmers’ adaptation strategies in northern Burkina Faso. Environmental Management. 2009; 43 :790–803. [ PubMed ] [ Google Scholar ]
• Barnett BJ, Barrett CB, Skees JR. Poverty traps and index-based risk transfer products. World Development. 2008; 36 :1766–1785. [ Google Scholar ]
• Baylis M, Githeko AK. Report for the foresight project on detection of infectious diseases. Department of trade and industry UK government; 2006. The effects of climate change on infectious diseases of animals. [ Google Scholar ]
• Bouwer LM. Have disaster losses increased due to anthropogenic climate change? Bulletin of the American Meteorological Society. 2011; 92 :39–46. January 2011, [ Google Scholar ]
• Brown A, Milton S, Cullen M, Golding B, Mitchell J, Shelly A. Unified modeling and prediction of weather and climate: a 25-year journey. Bulletin of the American Meteorological Society. 2012; 93 :1865–1877. [ Google Scholar ]
• Challinor AJ, Ewert F, Arnold S, Simelton E, Fraser E. Crops and climate change: progress, trends, and challenges in simulating impacts and informing adaptation. Journal of Experimental Botany. 2009; 60 :2775–2789. [ PubMed ] [ Google Scholar ]
• Challinor AJ, Stafford Smith M, Thornton PK. Agro-climate ensembles: emerging tools for quantifying uncertainty and informing adaptation. Agricultural and Forest Meteorology. 2013; 170 :2–7. [ Google Scholar ]
• Challinor AJ, Watson J, Lobell DB, Howden SM, Smith DR, Chhetri N. A meta-analysis of crop yield under climate change and adaptation. Nature Climate Change. 2014a; 4 :287–291. [ Google Scholar ]
• Challinor AJ, Martre P, Asseng S, Thornton PK, Ewert F. Making the most of climate impact ensembles. Nature Climate Change. 2014b; 4 :77–80. [ Google Scholar ]
• CIESIN. Center for International Earth Science Information Network. Global Subnational Prevalence of Child Malnutrition v1. 2005b. Available at: beta.sedac.ciesin.columbia.edu/data/set/povmap-global-subnational-prevalence-child-malnutrition (accessed 7 November 2012)
• CIESIN. Gridded Population of the World Version 3 (GPWv3): Population Grids. Palisades, NY: Socioeconomic Data and Applications Center (SEDAC), Columbia University; 2005a. Center for International Earth Science Information Network Columbia University and Centro Internacional de Agricultura Tropical (CIAT) Available at: http://sedac.ciesin.columbia.edu/gpw . (accessed 7 November 2012) [ Google Scholar ]
• Codjoe S, Owusu G. Climate change/variability and food systems: evidence from the Afram Plains, Ghana. Regional Environmental Change. 2011; 11 :753–765. [ Google Scholar ]
• Cory RM, Crump BC, Dobkowski JA, Kling GW. Surface exposure to sunlight stimulates CO 2 release from permafrost soil carbon in the Arctic. Proceedings of the National Academy of Science of the United states of America. 2013; 110 :3429–3434. [ PMC free article ] [ PubMed ] [ Google Scholar ]
• Costello A, Abbas M, Allen A, et al. Managing the health effects of climate change. Lancet. 2009; 373 :1693–1733. [ PubMed ] [ Google Scholar ]
• Craine JM, Nippert JB, Elmore AJ, Skibbe AM, Hutchinson SL, Brunsell NA. Timing of climate variability and grassland productivity. Proceedings of the National Academy of Sciences of the United states of America. 2012; 109 :3401–3405. [ PMC free article ] [ PubMed ] [ Google Scholar ]
• Craufurd PQ, Wheeler TR. Climate change and the flowering time of annual crops. Journal of Experimental Botany. 2009; 60 :2529–2539. [ PubMed ] [ Google Scholar ]
• CRED. EM-DAT: Emergency Events Database. Belgium: Université Catholique de Louvain; 2008. Centre for Research on the Epidemiology of Disasters. Available at: http://www.emdat.be/ [ Google Scholar ]
• Crespo O, Hachigonta S, Tadross M. Sensitivity of southern African maize yields to the definition of sowing dekad in a changing climate. Climatic Change. 2011; 106 :267–283. [ Google Scholar ]
• Dai A. Drought under global warming: a review. WIREs Climate Change. 2011; 2011 :45–65. [ Google Scholar ]
• Dougill AJ, Fraser EDG, Reed MS. Centre for Climate Change Economics and Policy Working Paper No. 32. UK: University of Leeds; 2010. Anticipating vulnerability to climate change in dryland pastoral systems using dynamic systems models for the Kalahari. [ Google Scholar ]
• Ericksen PJ. Conceptualizing food systems for global environmental change research. Global Environmental Change. 2008; 18 :234–245. [ Google Scholar ]
• Ericksen PJ, de Leeuw J, Thornton PK, Notenbaert A, Cramer L, Jones PG, Herrero M. Climate change in sub-Saharan Africa: what consequences for pastoralism? In: Catley A, Lind J, Scoones I, editors. Pastoralism and Development in Africa: Dynamic Change at the Margins. London: Earthscan; 2012. pp. 71–81. [ Google Scholar ]
• FAO. Food Balance Sheets: A Handbook. Rome, Italy: FAO; 2001. (Food and Agriculture Organization) of the United Nations. [ Google Scholar ]
• FAO. Climate-Smart Agriculture: Policies, Practices and Financing for Food Security, Adaptation and Mitigation. Rome, Italy: FAO; 2010. (Food and Agriculture Organization) of the United Nations. [ Google Scholar ]
• Few R. Health and climatic hazards: framing social research on vulnerability, response and adaptation. Global Environmental Change. 2007; 17 :281–295. [ Google Scholar ]
• Gornall J, Betts R, Burke E, Clark R, Camp J, Willett K, Wiltshire A. Implications of climate change for agricultural productivity in the early twenty-first century. Philosophical Transactionof the Royal Society B. 2010; 365 :2973–2989. [ PMC free article ] [ PubMed ] [ Google Scholar ]
• Gupta J, Termeer C, Klostermann J, et al. The adaptive capacity wheel: a method to assess the inherent characteristics of institutions to enable the adaptive capacity of society. Environmental Science and Policy. 2010; 13 :459–471. [ Google Scholar ]
• Haines A, Kovats RS, Campbell-Lendrum D, Corvalan C. Climate change and human health: impacts, vulnerability and public health. Public Health. 2006; 120 :585–596. [ PubMed ] [ Google Scholar ]
• Hellin J, Shiferaw B, Cairns JE, et al. Climate change and food security in the developing world: potential of maize and wheat research to expand options for adaptation and mitigation. Journal of Development and Agricultural Economics. 2012; 4 :311–321. [ Google Scholar ]
• Herrero M, Thornton PK, Notenbaert A, et al. Smart investments in sustainable food production: revisiting mixed crop-livestock systems. Science. 2010; 327 :822–825. [ PubMed ] [ Google Scholar ]
• Herrero M, Havlík P, Valin H, et al. Global livestock systems: biomass use, production, feed efficiencies and greenhouse gas emissions. Proceedings of the National Academy of Science of the United states of America. 2013; 110 :20888–20893. [ PMC free article ] [ PubMed ] [ Google Scholar ]
• Hlavinka P, Trnka M, Semeradova D, Dubrovsky M, Zalud Z, Mozny M. Effect of drought on yield variability of key crops in Czech Republic. Agricultural and Forest Meteorology. 2009; 149 :431–442. [ Google Scholar ]
• Hurkman WJ, Vensel WH, Tanaka CK, Whitehead L, Altenbach SB. Effect of high temperature on albumin and globulin accumulation in the endosperm proteome of the developing wheat grain. Journal of Cereal Science. 2009; 49 :12–23. [ Google Scholar ]
• IPCC. Intergovernmental Panel on Climate Change. Climate Change 2007: impacts, adaptation and vulnerability; summary for policy makers; 2007. Available at: www.ipcc.cg/SPM13apr07.pdf (accessed 3 November 2012) [ Google Scholar ]
• IPCC. Managing the risks of extreme events and disasters to advance climate change adaptation. In: CB Field, V Barros, TF Stocker, et al., editors. A Special Report of Working Groups I and II of the Intergovernmental Panel on Climate Change. Cambridge, NY, USA: Cambridge University Press; 2012. [ Google Scholar ]
• Jones PG, Thornton PK. Croppers to livestock keepers: livelihood transitions to 2050 in Africa due to climate change. Environmental Science and Policy. 2009; 12 :427–437. [ Google Scholar ]
• Jones PG, Thornton PK. Generating downscaled weather data from a suite of climate models for agricultural modelling applications. Agricultural Systems. 2013; 114 :1–5. [ Google Scholar ]
• Lenton TM. Early warning of climate tipping points. Nature Climate Change. 2011; 1 :201–209. [ Google Scholar ]
• Lipper L. FAO/OECD. Paris: Expert Meeting on Greening the Economy with Agriculture; 2011. Stability of food security in a green economy environment; pp. 5–7. September 2011. Working Document 3. [ Google Scholar ]
• Lloret F, Escudero A, Iriondo JM, Martinez-Vilalta J, Valladares F. Extreme climatic events and vegetation: the role of stabilizing processes. Global Change Biology. 2012; 18 :797–805. [ Google Scholar ]
• Lloyd SJ, Kovats RS, Chalabi Z. Climate change, crop yields and undernutrition: development of a model to quantify the impacts of climate scenarios on child undernutrition. Environmental Health Perspectives. 2011; 119 :1817–1823. [ PMC free article ] [ PubMed ] [ Google Scholar ]
• Lobell DB, Bänziger M, Magorokosho C, Vivek B. Nonlinear heat effects on African maize as evidenced by historical yield trials. Nature Climate Change. 2011; 1 :42–45. [ Google Scholar ]
• Maidment RI, Grimes DIF, Allan RP, Greatrex H, Rojas O, Leo O. Evaluation of satellite-based and model re-analysis rainfall estimates for Uganda. Metrological Applications. 2013; 20 :308–317. [ Google Scholar ]
• Martin V, Chevalier V, Ceccato P, et al. The impact of climate change on the epidemiology and control of Rift valley fever. Revue Scientifique et Technique, Office International des Epizooties. 2008; 27 :413–426. [ PubMed ] [ Google Scholar ]
• McMichael AJ, Kovats RS. Climate change and climate variability: adaptations to reduce adverse health impacts. Environmental Monitoring and Assessment. 2000; 61 :49–64. [ Google Scholar ]
• McMichael AJ, Woodruff RE, Hales S. Climate change and human health: present and future risks. Lancet. 2006; 367 :859–869. [ PubMed ] [ Google Scholar ]
• Meehl GA, Goddard L, Boer G, et al. Decadal climate prediction: an update from the trenches. Bulletin of the American Meteorological Society. 2013 e-View. doi: http://dx.doi.org/10.1175/BAMS-D-12-00241.1 . [ Google Scholar ]
• Mills E. Insurance in a climate of change. Science. 2005; 309 :1040–1044. [ PubMed ] [ Google Scholar ]
• Misselhorn A, Eakin H, Devereux S, Drimie S, Msangi S, Simelton E, Stafford Smith M. Vulnerability to what? chapter seven. In: Ingram JSI, Ericksen PJ, Liverman DM, editors. Food Security and Global Environmental Change. London: Earthscan; 2010. [ Google Scholar ]
• Mohammed AR, Tarpley L. High nighttime temperatures affect rice productivity through altered pollen germination and spikelet fertility. Agricultural and Forest Meteorology. 2009; 149 :999–1008. [ Google Scholar ]
• Munich Re. TOPICS GEO, Natural Catastrophes 2010, Analyses, Assessments, Positions. Munich, Germany: Munich Re; 2011. [ Google Scholar ]
• Ncube M, Zikhali P, Musango JK. The impact of climate variability on water and energy demand: the case of South African local governments. Water and Environment Journal. 2012; 27 :29–41. [ Google Scholar ]
• Nelson GC. Climate Change: Impact on Agriculture and Costs of Adaptation. Washington DC, USA: International Food Policy Research Institute; 2009. [ Google Scholar ]
• Nielsen JO, Reenberg A. Temporality and the problem with singling out climate as a current driver of change in a small West African village. Journal of Arid Environments. 2010; 74 :464–474. [ Google Scholar ]
• NRC. Effect of Environment on Nutrient Requirements of Domestic Animals. Washington DC: Subcommittee on Environmental Stress, National Academy Press; 1981. [ Google Scholar ]
• O'Brien K. Global environmental change II: from adaptation to deliberate transformation. Progress in Human Geography. 2012; 36 :667–676. [ Google Scholar ]
• O'Brien K, Eriksen S, Nygaard LP, Schjolden A. Why different interpretations of vulnerability matter in climate change discourses. Climate Policy. 2007; 7 :73–88. [ Google Scholar ]
• Porter JR, Semenov MA. Crop responses to climatic variation. Philosophical Transactions of the Royal Society B. 2005; 360 :2021–2035. [ PMC free article ] [ PubMed ] [ Google Scholar ]
• Raleigh C, Jordan L. Climate change and migration: emerging patterns in the developing world. In: Mearns R, Norton A, editors. Chapter 4 in “Social Dimensions of Climate Change: Equity and Vulnerability in a Warming World”. Washington DC, USA: World Bank; 2010. pp. 103–131. [ Google Scholar ]
• Ramirez-Villegas J, Challinor AC, Thornton PK, Jarvis A. Implications of regional improvement in global climate models for agricultural impacts research. Environmental Research Letters. 2013; 8 :024018. [ Google Scholar ]
• Rao KPC, Ndegwa WG, Kizito K, Oyoo A. Climate variability and change: farmer perceptions and understanding of intra-seasonal variability in rainfall and associated risk in semi-arid Kenya. Experimental Agriculture. 2011; 47 :267–291. [ Google Scholar ]
• Rickards L. Critical Breaking Point: the effects of climate variability, change and other pressures on farm households. Report for the Birchip Cropping Group and the Sustainable Agriculture Initiative Platform Australia. 2012. Available at: http://www.bcg.org.au/cb_pages/publications.php (accessed 2 February 2013)
• Rockström J, Falkenmark M, Karlberg L, Hoff H, Rost S, Gerten D. Future water availability for global food production: the potential of green water for increasing resilience to global change. Water Resources Research. 2009; 45 :1–16. [ Google Scholar ]
• Rodríguez LC, Herrero M, Baltenweck I. Community-based interventions for the use and conservation of animal genetic resources: the case of indigenous scavenger chicken production in Benin. Tropical Animal Health Production. 2011; 43 :961–966. [ PubMed ] [ Google Scholar ]
• Rosegrant MW, Fernandez M, Sinha A. Looking into the future for agriculture and AKST (Agricultural Knowledge Science and Technology) In: McIntyre BD, Herren HR, Wakhungu J, Watson RT, et al., editors. Agriculture at a Crossroads. Washington DC, USA: Island Press; 2009. pp. 307–376. [ Google Scholar ]
• Rowhani P, Lobell DB, Linderman M, Ramankutty N. Climate variability and crop production in Tanzania. Agricultural and Forest Meteorology. 2011; 151 :449–460. [ Google Scholar ]
• Skoufias E, Vinha K, Conroy HV. The impacts of climate variability on welfare in rural Mexico; Policy Research Working Paper 5555; Washington DC, USA: World Bank; 2011. [ Google Scholar ]
• Solano C, Bernués A, Rojas F, Joaquıń N, Fernandez W, Herrero M. Relationships between management intensity and structural and social variables in dairy and dual-purpose systems in Santa Cruz, Bolivia. Agricultural Systems. 2000; 65 :159–177. [ Google Scholar ]
• Speranza CI, Kiteme B, Ambenje P, Wiesmann U, Makali S. Indigenous knowledge related to climate variability and change: insights from droughts in semi-arid areas of former Makueni District, Kenya. Climatic Change. 2010; 100 :295–315. [ Google Scholar ]
• Tall A, Mason SJ, van Aalst M, Suarez P, Ait-Chellouche Y, Diallo AD, Braman L. Using seasonal climate forecasts to guide disaster management: the Red Cross experience during the 2008 West Africa floods. International Journal of Geophysics. 2012; 2012 :1–12. [ Google Scholar ]
• Thornton PK, Herrero M. The inter-linkages between rapid growth in livestock production, climate change, and the impacts on water resources, land use, and deforestation; World Bank Policy Research Working Paper, WPS 5178; Washington, DC, USA: World Bank; 2009. [ Google Scholar ]
• Thornton PK, Jones PG, Ericksen PJ, Challinor AJ. Agriculture and food systems in sub-Saharan Africa in a four-plus degree world. Philosophical Transactions of the Royal Society Series A. 2011; 369 :117–136. [ PubMed ] [ Google Scholar ]
• Thornton PK, Lipper L, Baas S, et al. How does climate change alter agricultural strategies to support food security? 2013. Background paper for the conference “Food Security Futures: Research Priorities for the 21st Century”Dublin, 11-12 April 2013.
• Thurlow J, Dorosh P, Yu W. A stochastic simulation approach to estimating the economic impacts of climate change in Bangladesh. 2011. Working Paper No. 2011/86, United Nations University.
• Tirado MC, Crahay P, Hunnes D, Cohen M. Climate change and nutrition in Africa with a focus on sub-Saharan Africa. 2010. Avilable at: www.sunrayafrica.co.za (accessed 4 February 2013)
• UNDESA. United Nations Department of Economic and Social Affairs. World population prospects: the 2012 revision, medium variant. 2013. Online Available at: http://esa.un.org/unpd/wpp/unpp/panel_population.htm (accessed 20 April 2013)
• USDA. United States Department of Agriculture. Food Security Assessment, 2010-20; Report GFA-21; Washington DC, USA: Economic Research Service; 2010. [ Google Scholar ]
• Vermeulen SJ, Challinor AJ, Thornton PK, et al. Addressing uncertainty in adaptation planning for agriculture. Proceedings of the National Academy of Science of the United states of America. 2013; 110 :8357–8362. [ PMC free article ] [ PubMed ] [ Google Scholar ]
• Vervoort J, Thornton PK, Kristjanson P, et al. Challenges to scenario-guided adaptive action on food security under climate change. Global Environmental Change. 2014 (in press) [ Google Scholar ]
• Weaver CP, Lempert RJ, Brown C, Hall JA, Revell D, Sarewitz D. Improving the contribution of climate model information to decision making: the value and demands of robust decision frameworks. WIREs Clim Change. 2013; 4 :39–60. [ Google Scholar ]
• Willenbockel D. Extreme weather events and crop price spikes in a changing climate. Oxfam Research Reports. 2012. September 2012.
• Willock J, Deary IJ, Edwards-Jones G, et al. The role of attitudes and objectives in farmer decision making: business and environmentally-oriented behaviour in Scotland. Journal of Agricultural Economics. 1999; 50 :286–303. [ Google Scholar ]
• You L, Guo Z, Koo J, et al. Spatial Production Allocation Model (SPAM) 2000 Version 3 Release 1. 2012. Available at: http://MapSPAM.info (accessed 10 February 2013)
• Ziervogel G, Opere A. Integrating meteorological and indigenous knowledge-based seasonal climate forecasts in the agricultural sector. Ottawa, Canada: International Development Research Centre; 2010. Climate Change Adaptation in Africa learning paper series. [ Google Scholar ]