## Introduction

Agriculture is the primary source of income and provides the main livelihood for the world’s poorest people in developing countries1,2,3. The nature of climate change (CC) impacts is variable4 and will drive varying responses in crops and regions around the world5. CC is expected to have positive and negative impacts on crop yields1,6, that will have important consequences for food prices and food security7. The risks to adverse effects of CC are expected to be greatest in Least Developed Countries (LDCs), particularly in sub-Saharan Africa (SSA) and South East Asia8 with the most vulnerable communities3,9,10. Projections from the 2013 Intergovernmental Panel on Climate Change (IPCC) report indicate that in Southern Africa (SAF), the most profound effects of CC will be temperature increases over 4 °C by the end of the century and less rainfall11. The agriculture sector in SSA is highly vulnerable to CC because of the high dependence on rainfed agriculture, which makes the region susceptible to adverse weather events such as flooding and drought12,13. This is exacerbated by the region’s immense reliance on natural resources, high population growth rates, acute poverty levels, and poor infrastructure3,8,9,10.

Located in the eastern part of SAF (Fig. 1), Malawi is an LDC14 with a huge poverty challenge13. The agriculture sector plays a pivotal role in Malawi’s socioeconomic wellbeing and poverty alleviation. It contributes to as much as 27% of the nation’s gross domestic product (GDP)15 and employs 85% of the population16. Over 90% of the agriculture sector consists of smallholder farmers17. Rainfed agriculture is widespread in Malawi, with less than 5% of famers using irrigation13. There is some evidence to suggest that Malawi’s fluctuating historical trend in GDP growth is a consequence of changes in seasonal rainfall (Supplementary Fig. S1); this emphasises the vulnerability of the country to CC. Over 95% of farmers in Malawi cultivate to meet their subsistence needs, highlighting the importance of farming for the nation’s food security18.

Maize (zea mays L) is the nation’s staple and most important food crop19. Maize is by far the most widely grown crop in the country and accounts for as much as 80% of the cultivated land20. It is thought that the country’s food security is defined by maize harvests and access to maize21,22. Up to 59% of Malawi’s land is used for cultivation23. In addition to maize, other crops that are important for food security in Malawi are cassava, groundnuts, peas, potatoes, pulses, and sorghum19,20. In Supplementary Table S1, we present a summary of recent (2013) production and international trade value information about these crops in comparison to maize.

High population growth in Malawi is a key driver behind the increasing national maize food requirement over the last few decades (Fig. 2). Fluctuating trends in Malawi’s GDP closely follow those of agricultural productivity and maize production (Fig. 3). These agricultural and economic fluctuations are reflected in food prices. Malawi’s domestic maize prices are more volatile than international prices18,19 which can have devastating social outcomes during food shortages when prices rise sharply24.

In the 2080s, the projected lower maize yields, coupled with increasing population, could lead to maize shortages and translate to between 0.1% and 12.2% of the population in Lilongwe District being vulnerable to food insecurity. A recent study by the Overseas Development Institute (ODI)48 in Malawi, communicated that CC is expected to adversely affect all four facets of food security namely availability, access, utilisation and stability. Taking into account 2013 values reported by the FAO for maize imports47, we estimated that the deficit in maize production could equate to as much as US$355 thousand being invested annually in international markets to feed Lilongwe District’s growing population. ## Discussion ### Impacts of climate change on Malawi’s agricultural system and food security Compared to other reported findings of massive reductions in maize yields in SSA and Malawi10,30,45,49, our study findings anticipate the decline in future maize yields in Lilongwe District to be moderate. This is because the projected changes for both the amount of rainfall and temperature increase, in the study area, are modest. In the short-term future, farmers could benefit from CC due to the expected increase in maize yields under favourable climate conditions. It has been reported that despite the devastating projected impacts of CC, some areas in SSA could fare relatively well1. This could have important research and policy implications for Malawi’s agricultural sector. Interestingly, higher maize production under favourable CC conditions could intensify the existing problem of poor soil fertility in Malawi. Accelerated crop growth can lead to high nutrient requirement and result in soil nitrogen deficiency6. We, therefore, envision that in the short-term future, the increasing demand for food and the positive impacts of CC on maize growth could adversely impact soil fertility. Given the moderate increases in temperature and relatively low decline in rainfall amount and variability, in the short-term future, impacts of soil fertility on maize yields far outweigh any predicted impacts of CC. In the medium to long-term future increasing maize demand caused by a growing population, coupled with decreasing maize production, could exacerbate vulnerability to food insecurity. Lower maize production is expected to be driven by (i) higher rainfall variability leading to more droughts and floods, (ii) increasing temperatures, (iii) a higher frequency of years when farmers fail to sow crops due to late onset of rainfall40, and (iv) a shorter maize growing season45. By the end of the century, maize requirements are projected to be greater than production and would therefore lead to food shortages (Fig. 8) and higher maize prices. ### Policy implications for addressing projected impacts of climate change In responding to the impacts of CC, the United Nations Framework Convention on CC (UNFCCC) advocates for the response of LDCs to be adaptation rather than mitigation. The two main schools of thought in CC adaptation strategy planning are sustainable resource management and research and development (R&D)12. Sustainable resource management includes consideration of soil, water, and crops; whilst under R&D, approaches such as the use of drought resistant and heat tolerant maize varieties could be considered50. Scholars hold the view that through its National Adaptation Programme of Action (NAPA), Malawi has the potential and opportunity to manage the threats and impacts of CC13. Through good field management practices such as irrigation and improved soil fertility, maize yields in developing countries can be increased from a range of 1 to 2 t/ha to as much as 11 to 14 t/ha46. The GoM and a number of non-governmental organisations (NGOs), academics and authors recognise the important role that irrigation has to play in improving food security in Malawi51. With the Ministry of Agriculture and Food Security (MoAFS) reporting that only 14% of Malawi’s irrigation potential is currently being utilised40, there is a great opportunity to improve maize yields through adopting this approach. Many farmers in Africa use crop diversification to build resilience into the agriculture sector49. The feasibility of such an approach is questionable especially in a nation such as Malawi where the prevalence of maize in the diet is very high19. Therefore crop diversification could raise big cultural questions, that is forsaking a traditional staple food - maize - for an alternative CC tolerant one9. Drought tolerant crops that have been suggested for Malawi include cassava, sorghum and millet13,20. Crop diversification has additional benefits for food security; it reduces malnutrition and places households in good stead to overcome volatile food prices52. Given the projected impacts of CC on soil fertility in the short-term future, a priority for the GoM would be to improve soil fertility. It is a widely held view that by improving affordability of farm inputs including fertiliser, through Malawi’s FISP, there have been tangible benefits for food security50. Soil composting22 and conservation agriculture (CA), through rotation of maize and cowpea crops45, have been recommended for Malawi. The application and effectiveness of CA methods has been exemplified in a study where maize yields in a region of Malawi were increased by 40%42. In their effort to promote the year 2015 as International Year of Soils, the FAO produced a series of radio programmes including one titled Chuma Chiri M’nthaka (“wealth is in the soil”) which was broadcasted in Malawi to promote integrated soil fertility management practices amongst smallholder farmers53. In the possible absence of adverse impacts of CC, in the short-term future as demonstrated in this study, maize yields could be greatly improved. Better maize yields could translate to increased food security in Malawi where most of the maize farming is for subsistence use. ### Prospects for improved income Being the main cereal-producing district in Malawi, and with the projected short-term positive impacts of CC, Lilongwe District could produce food surpluses. If these could reach the international food markets, under the expected global price increase for cereals9,10,12, the Malawian economy stands to possibly gain from the short-term impacts of CC. There are other reports of Malawi’s agriculture sector benefitting from CC; for example, a study in Mzimba District in the northern region of Malawi concluded that in the period between 2040 and 2070, 56% of farmers would gain from CC because of higher maize yields29. However, despite the projected attractive global prices and increasing global demand for maize, the ability of Malawi’s agriculture sector to become an established exporter is questionable given the need to feed a growing population. A major hindrance to the implementation of CC adaptation measures is poverty2. As mentioned previously, poverty increases the vulnerability of Malawi to CC impacts. A recurring challenge for the GoM is in mobilizing resources to respond to immediate food security crises whilst concurrently investing for the nation’s long-term future in agriculture18. What is more, the sustainability of agriculture as a source of income for Malawian households in the future has been questioned20. However, we believe that by investing in R&D and increasing agricultural productivity and yields, smallholder farmers could produce crop surpluses which could be sold to bring in extra household income. Such economic gains could increase per capita income, drive poverty levels down and empower smallholder farmers. Furthermore, it has been demonstrated empirically that higher income can produce tangible benefits for nutrition in Malawi52. Economic empowerment could allow smallholder farmers, who make up 90% of Malawi’s agriculture sector, to lift themselves out of poverty by increasing their access to education, health, and financial services28. By empowering Malawi’s poorest people, there could be tangible benefits for the whole nation. Given the pivotal role that the agriculture sector holds in the Malawian economy (Fig. 3a), any CC adaptation measures that can mitigate the projected adverse CC impacts are likely to benefit a significant number of farmers and the population. ### Study limitations It is important to bear in mind the possible bias in the results from this study. Although we used state-of-the-art GCMs to predict future CC in Lilongwe District, climate projections are subjective. The acknowledged sources of uncertainty in GCMs are: uncertainty in future anthropogenic greenhouse gas (GHG) emissions and natural forcings; limited knowledge of current climate conditions; challenges of representing variability in future projections; and imperfections in the GCMs54. Moreover, with a small number of GCMs, caution must be applied, in interpreting our study results. A major drawback of using a small number of GCMs is that the CC projections could fail to account for seasonal and regional biases of climate model simulations, which could be overcome by using more GCMs. Additionally, we did not make adjustments for how our CC projections would affect distributions of dry and wet spells in the future. A more in-depth assessment of rainfall variability and temperature changes would highlight probable future duration of wet and dry spells. This could be particularly useful in Malawi where it is a widely held view that socioeconomic wellbeing is linked with seasonal rainfall (Supplementary Fig. S1). Other sources of bias in our results could come from having limited information about the farming conditions in Lilongwe District for the AquaCrop model calibration, and using a restricted set of assumptions for the food security vulnerability assessment. Furthermore, we did not consider other environmental parameters which could have an effect and influence the results presented here. Finally, we did not take into account pre-existing socioeconomic factors such as access to local and international markets; market conditions under the predicted increasing global demand for maize; changes in consumer preference; and potential interventions such as subsidies. ## Conclusion We undertook a study to evaluate the impacts of CC on maize production and food security in Malawi’s Lilongwe District in three future time periods (2020s, 2050s and 2080s) compared to a baseline period (1971 to 2000), under RCP4.5 and RCP8.5 CC emission scenarios. We downscaled the outputs from five GCMs using a commonly used stochastic WG, the LARS-WG5. We then used a three-step statistical approach to account for the uncertainty of GCM outputs. Through this process, a range of probability percentiles of projected rainfall and temperature changes were developed (25th, 50th, and 75th). We used the FAO crop model, AquaCrop, to simulate maize yields in the future under CC. Our climate modelling results suggest that maximum temperature could increase by 0.7 °C to 0.8 °C, 1.6 °C to 2.3 °C, and 2.1 °C to 3.3 °C in the 2020s, 2050s and 2080s, respectively. Although the annual rainfall results did not reveal a strong increasing or decreasing trend, we expect more rainfall variability in the future. Under CC, maize yields are expected to both increase and decrease in the future. We predict that maize yield gains and losses would range from 4.6% to 5.4%, −1.2% to 1.0% and −3.0% to 0.2% in the 2020s, 2050s and 2080s correspondingly. The results reflect the opposing apparent effects of CC on crop production. Despite the projected high population growth rates for Malawi, maize requirements could be met in the short to medium-term future. In the 2080s, under the assumptions used in the study, lower maize production could result in between 0.1% and 12.2% of the population being at risk of food insecurity. We estimate that as much as US$ 355 thousand would need to be allocated annually for purchasing maize from international markets to feed the growing population of Lilongwe District alone.

Our findings highlight the importance and implications of future policy formulation and additional research with regards to CC adaptation measures. A reasonable path to achieve food security, under inevitable CC, could be through a holistic approach that considers sustainable resource management as well as economic and market factors. To meet the probable growing soil nutrient deficiency, caused by faster growth crop under CC in the short-term future, the GoM would need to consider measures that improve soil quality and fertility. To mitigate against the projected adverse effects of CC on maize yields and food security in the medium to long-term future, we consider irrigation and access to local and international markets to be key factors. Given the pivotal role that agriculture holds in Malawi’s economy and the imminent growing population, such practices are likely to have benefits that will drive income and social empowerment for many smallholder farmers in Lilongwe District and Malawi.

## Methods

### Climate change modelling

We used the outputs from five GCMs (Supplementary Table S6), under two emission scenarios (RCP4.5 and RCP8.5), described in the Fifth Assessment Report (AR5) of the IPCC, to make projections for rainfall and temperature in 2011 to 2040 (2020s), 2041 to 2070 (2050s) and 2071 to 2100 (2080s). For each GCM, we extracted daily time series of rainfall, temperature and short wave radiation for RCP4.5 and RCP8.5 for 2011 to 2100 as well as for the baseline period (1971 to 2000). Founded on the Coupled Model Intercomparison Project Phase 5 (CMIP5) multi-model experiment of the World Climate Research Programme, the RCPs are the 2013 CC emission scenarios published by the IPCC. The RCP4.5 scenario is characterised by CC policy and increased environmental sustainability resulting in stable GHG11,55. On the other hand the RCP8.5 scenario assumes high GHG emissions caused by high population growth and energy usage, low rates of development in developing countries and a heavy reliance on fossil fuels11,55. We compared the short (2020s), medium (2050s) and long-term (2080s) future changes in climate to the baseline period. We obtained historical observed weather data, collected by Chitedze Research Station in Lilongwe District, from the Department of Climate Change and Meteorological Services (DCCMS) in the Ministry of Natural Resources, Energy and Environment in Malawi.

Understanding performance of GCMs is an important aspect of CC science54. Therefore we assessed the uncertainty in the GCMs by applying a probability analysis method with bounded distribution functions in a three step approach5. In the first step, we assessed the ability of each GCM to reproduce baseline rainfall and temperature using the Mean Observed Temperature-Precipitation (MOTP) method5. We weighted each model by calculating, for both rainfall and temperature, the difference between the monthly mean observed values and the simulated values for the baseline period as follows5,37:

where Wij is the weight of GCM j in month i, Δdij is the absolute difference in rainfall or temperature between the monthly mean simulated by GCM j in month i of the baseline period and the corresponding observed value, n equals five (the number of GCMs). It is worth highlighting some factors which we identify as limitations in the research methods in this study. Although our methodology tries to address the uncertainty of the GCMs, using a limited number of GCMs could introduce error into the climate modelling procedure5.

In the subsequent step, we generated discrete PDFs showing the relationship between the difference in monthly rainfall and temperature and the calculated weight of the corresponding GCM (Supplementary Figs S2a and S3a). This involved calculating the ratio of rainfall and absolute difference for temperature in each future time period, under each RCP, and comparing it to the baseline data for each month as follows5:

where ΔRij and ΔTij are CC scenarios of rainfall and temperature in month i and year j, RGCM, fut, ij and TGCM, fut, ij are the rainfall and temperature for each GCM in month i and year j in the future, RGCM, base, i and TGCM, base, i are the average simulated rainfall and temperature for each GCM in month i for the baseline period. At each future time period, at each emission scenario, we plotted 24 PDFs (12 for temperature and 12 for rainfall for each month); therefore we developed a total of 144 PDFs for the three future time periods under RCP4.5 and RCP8.5. To make CC projections about the future, we developed time series of rainfall and temperature from continuous PDFs. We fitted the Beta distribution function (equation (4)) onto each discrete PDF to convert it to a continuous PDF (Supplementary Figs S2b and S3b)37:

where x is rainfall or temperature, p and q are the shape parameters for the Beta distribution function, a and b are the minimum and maximum rainfall or temperature changes, and B(p,q) is the Beta function given by equation (5) and equation (6).

We used the maximum likelihood estimation method to identify the shape parameters (p and q) for the Beta distribution function. We calculated these by minimising the SSE as follows5,37:

where yi is the calculated weight for each GCM, Yi is the estimation of the Beta function, and n equals five (the number of GCMs).

In the final step, we converted the PDFs to CDFs so that future rainfall and temperature changes could be calculated at different probability percentiles (Supplementary Figs S2c and S3c). In our risk assessment interpretation we considered that (i) a low CC risk level scenario represents low temperature (25% probability percentile) and high rainfall (75% probability percentile)5,30, (ii) a high CC risk level scenario represents high temperature (75% probability percentile) and low rainfall (25% probability percentile)5,30, and (iii) the 50% probability percentile represents a medium risk level scenario of CC37.

We used the LARS-WG5 stochastic WG38 to downscale the GCM outputs in three distinct steps. In the model calibration step, we analysed the observed baseline (1971 to 2000) data (rainfall, maximum and minimum temperature, and hours of sunshine) to determine their statistical characteristics. We simulated weather data at the same grid point (−14.13 South and 33.38 East) as was used for the extraction of GCM data. To validate the LARS-WG5 model, we assessed its performance by comparing frequency distributions, mean values and standard deviations of the observed and simulated data of the baseline period using statistical tests (Kolmogorov-Smirnov test, T-test and F-test)56. We used these statistical tests to check the significance and the reliability of LARS-WG5 to predict future climate data, with reference to the baseline period, at the 0.01 significance level. Once we had deemed the LARS-WG5 performance to be acceptable, we used it to simulate future weather data. To manage the uncertainty of natural variability in the LARS-WG5, we generated ten 30-year time series for each future time period at the different probability percentiles. We used the average of these ten time series for the subsequent crop modelling step.

### Maize yield modelling

We used the FAO crop model, AquaCrop41, to simulate future maize yields under CC in three distinct steps: (i) calibration, (ii) validation, and (iii) simulation of future maize yields. A full list of parameters used to calibrate the model are summarised in Supplementary Table S7. In all three steps, evapotranspiration data (ETo) for the evaporation of water from the maize crop surfaces was required. We calculated this from weather data (maximum and minimum temperature, mean relative humidity, wind speed and hours of sunshine) using the ETo calculator which uses the FAO Penman-Monteith method described by Allen et al.57. We used a number of sources to obtain data for the model calibration to the farming conditions and practices in Lilongwe District including literature34,40,43,58,59,60,61 and a consultation with a research scientist at Chitedze Agricultural Research Station. Based on the maize yield reported by the MoAFS in the year 2000, we calibrated the model. We validated the model using maize yields from the years 2001 to 2005 and five statistical indicators (Pearson coefficient of determination, relative root mean square error, normalized root mean square error, Nash-Sutcliffe and Willmott’s index of agreement)41. We found the model’s performance satisfactory and used it to make projections for future maize yields using baseline period weather data, the outputs from the LARS-WG for future weather data, and future CO2 concentrations62. We used the Pearson coefficient of determination to independently assess the relationships between maize yields and rainfall, temperature and CO2 concentrations.

The crop model calibrated in this study can be improved further by obtaining more user-specific information63 through controlled field experiments to mimic the farming practices in Lilongwe District. Additionally, we did not consider the effects of pests and diseases, and response of different maize cultivars. With respect to growing seasons, we only accounted for the rainy season maize cultivation and did not consider the widespread cultivation of maize in the dry season in Malawi.

### Food security modelling

We modelled impending food security by assessing maize deficits and surpluses from estimates of maize production and maize requirement. We calculated maize production as follows:

where 6159 square kilometres (Km2) is the area of Lilongwe District31, 59% is the proportion of cultivated land in Malawi23, 80% is the proportion of cultivated land that is used for maize farming in Lilongwe District20 and Y is the simulated maize yield (in t/ha) from AquaCrop for that year.

We calculated maize requirement as follows:

where 120 Kilograms (Kgs) is the MVAC estimate for annual maize requirement per person64, p is the population estimate for that year based on UN population growth estimates35 and the reported population for Lilongwe District36. By using the three different UN future population estimates for Malawi, we included a sensitivity analysis in the future projections of food security vulnerability in Lilongwe District.

We calculated maize deficit and surplus as the difference between maize requirement and maize production:

We calculated the proportion of the population in Lilongwe District that could be vulnerable to food insecurity in any given year as follows:

where p is the population estimate for that year based on UN estimates35.

Our food security vulnerability assessment can be improved in future studies by including other factors in the Household Economy Approach (HEA) framework, such as effects of maize yields on GDP and food price65.

### Estimating financial value of projected maize production and requirements

We estimated the monetary (import and export) value of the projected maize surpluses and deficits using values reported by the FAO (2013)47 for the 2013 international trade value of maize in Malawi. We did this as follows:

where US$20,806,000 was the import value of 3,639,866 tonnes of maize47. where US$ 1,713,000 was the export value of 3,639,866 tonnes of maize47.