iFEED is a modelling framework used to calculate changes to crop yields, land use and food production and ultimately to assess nutrition security in 2050. In this analysis, we define nutrition security as adequate calorie and nutrient supplies to meet dietary requirements at a population level. For most nutrients, this is based on the Reference Nutrient Intake (RNI), which is the amount that is sufficient to ensure that the needs of 97.5% of individuals in the population are met. While RNI is based on individual requirements, a weighted average requirement was calculated based on population demographics and was used as this provides a benchmark for whether the national supply is theoretically sufficient. We acknowledge that this does not guarantee nutrition security at the individual level, as actual adequacy is contingent upon equitable food distribution, affordability and access, which are not measured in this supply-side analysis.
The iFEED modelling framework assesses nutrition security given changes to land use, food production and international trade. iFEED has been previously described17, and similar data and methods have been applied for Zambia2. Here, we describe our approach to assess future changes to land use, yields, food production and nutrition security for both average and extreme future years. A comparison of iFEED with alternative integrated assessment models for assessing future food systems is provided in Supplementary Section 3.
We first simulate climate change impacts on crop yields using the General Large Area Model for annual crops (GLAM34,35,36). GLAM is run on a 0.5° grid using a subset of bias-corrected climate data for Africa19 (Supplementary Informaton section 2) and soil data from the Harmonized World Soil Database (HWSD) version 1.2 (ref. 37). The soil input parameters required for crop modelling were the drained lower limit, drained upper limit and saturation limit. Percentage values of sand and clay, aggregated to the 0.5° grid, were used to calculate these hydrological parameters38. We compared HWSD versions 1.2 and 2 and found minimal difference in crop model results when driven by the two datasets (Supplementary Fig. 3). The crops modelled were maize, soybean, potato and groundnut. Simulations for these crops underpinned projections for the full range of crops grown in Zambia. Historical yield data39 were altered in future to represent climate change impacts using average projected changes from soybean, groundnut and potato for C3 crops, and using maize climate change projections for C4 crops. Simulations were conducted using a historical period around the year 2000 (1990–2010), and a future period centred on 2050 (2040–2060). Simulations include autonomous adaptation to climate change by allowing planting dates to change in response to climate and altering crop varieties to mitigate growing season reduction due to warming. Planting windows and crop varieties were selected based on the combination that returned the highest yield in each year (using a rolling 21-year average), with crop varieties defined using different maturity classes. Both scenarios applied the same technology trend to increase crop yields by 2050; for each crop, continuation of historical trends as seen in Food and Agriculture Organization of the United Nations Statistics Division (FAOSTAT39) yield data from 1960–2010 was applied, as in previous analyses2,17.
Baseline and future simulations assumed irrigated conditions according to Monthly Irrigated and Rainfed Crop Areas (MIRCA) data40. Irrigation grid cells were determined using a majority grid cell approach using the MIRCA data. If more than 50% of growing area in a grid cell was irrigated, fully irrigated (that is, no water stress) simulations were used in that grid cell.
We then calculate changes to agricultural land using iFEED land use allocation routines that maximize production on available land as described in the scenario land use allocation section below. Crop production changes to 2050 are then calculated using land use and yield projections. Livestock production changes are calculated using projected changes to livestock pasture, crop residues and crop production used as livestock feed, and assume historical relationships between livestock feed and livestock meat and dairy production remain the same by 205041. This assumes that future livestock feed ratios and livestock productivity remain the same to 2050, agnostic to any changes to herd structures and production systems. For more details and implications of the livestock modelling, see Supplementary Section 1.
We use FAOSTAT39 Food Balance Sheet (FBS) data, along with crop nutrient content data33 and United Nations medium variant population projections, to assess changes to population-level nutrition security. In this analysis, we assume that, for each food commodity, international imports and exports remain in the same proportion to domestic production as in the baseline. The FBS data provide an estimate of the supply of food commodities based on domestic production, imports and exports, including stock variation of each commodity. In iFEED, the supply of energy and micronutrients are calculated as follows. FBS food commodities are converted to food as eaten, adjusting for unavoidable waste (such as inedible peel and bones) and household waste (uneaten edible food), as well as adjusting for losses from storage and transportation accounted for in the FBS, and food used for livestock feed, seed and processing. These food uses are assumed to continue in the same proportion to baseline domestic supply quantity in the future, that is, if 10% of food is lost in storage in the baseline, 10% is assumed lost in 2050. The food commodities are disaggregated into food items and matched to foods in a region-specific food composition table, which provides an estimate of calories, protein, fat, carbohydrate, saturated fat, fibre, calcium, zinc, iron, vitamin C, thiamin, riboflavin, niacin, folate and vitamin B6 per 100 g for each commodity group. Each food item is then weighted to represent the quantity of each food eaten at a country level, before being aggregated back to food commodity groups. We assume no changes to the weightings of foods within each food item between baseline and future for this calculation. Adequate nutrition requirements are set according to World Health Organization recommendations. The population-level nutrient requirements are adjusted for projected demographic changes (population size, age, sex and fertility rates) based on medium-variant United Nations projections to 2050.
In this analysis, we use iFEED to explore two contrasting scenarios that offer potential pathways towards nutrition security for Zambia given climate extremes: nutrition focus and production focus. The next two sections describe the methods used in the scenarios, and our methods to assess climate extremes, respectively.
Scenario land use allocation
Both scenarios assume crop area expansion from 2000 to 2050 to different extents. This study prioritizes a yield-based approach to land use allocation that is focused on biophysical potential rather than complex economic modelling. We avoid uncertainties associated with markets, income, food prices and food demand changes to 2050 to determine agricultural land use. Optimization of land use is based on placing crops where they have the highest yield until area requirements are met, with the crops grown specified by total areas nationally and varying in each scenario as detailed below. Crop diversity at the level of each grid cell is determined by a parameter that specifies the maximum fraction of each crop that can be placed in each grid cell. This parameter was specified to be the same for both scenarios, and to be as low as possible while still allowing national crop targets to be met for all climate models.
The production focus scenario maximizes the crop production on available land by optimizing the placement of crops based on simulated yields in the future period, without considering nutritional requirements. The scenario assumes a maximum possible increase in agricultural land, which was determined using Land-Use Harmonization II data42, and assumes all land was available for agricultural expansion if not forested, urban or protected according to The World Database on Protected Areas43. The total areas of each crop nationally are kept in the same proportion in the future as in the baseline; using the total arable area, and these proportions, future crop area targets are calculated. To maximize production, crops were allocated to grid cells using a ‘highest-yield-first’ priority ranking. In this approach, land was allocated to the highest-yielding crop–cell combination first, continuing iteratively until either the physical land within a grid cell was exhausted or the precalculated national area target for a specific crop was reached.
The nutrition focus scenario is designed to improve nutrition security using changes to crop and livestock production to meet nutrient requirements by mid-century, minimizing cropland expansion, and without expanding pasture for livestock, to minimize the negative environmental impacts associated with land use expansion. We first calculate the quantities of crop and livestock production required to ensure nutrition security per person, minimizing changes to current diets. This step uses linear programming to calculate per capita food commodity (FAOSTAT FBS data) requirements that achieve nutrition security. We allow changes to food commodities derived from only the crop and non-ruminant livestock (primarily pig and poultry) commodities produced in Zambia. The crop commodities included in the linear programming are ‘Wheat and products’, ‘Maize and products’, ‘Millet and products’, ‘Sorghum and products’, ‘Sugar cane’, ‘Barley and products’, ‘Cassava and products’, ‘Groundnuts (Shelled Eq)’, ‘Onions’, ‘Potatoes and products’, ‘Rice (Milled Equivalent)’, ‘Cottonseed’, ‘Sunflower seed’, ‘Sweet potatoes’, ‘Tomatoes and products’, ‘Vegetables, Other’, ‘Bananas’, ‘Fruits, Other’, ‘Oranges, Mandarines’, ‘Pulses, Other and products’ and ‘Soyabeans’. In most SSA production systems, ruminants rely on natural pastures and crop residues, which account for the majority of feed despite recent increases in fodder crop intake44,45. Therefore, no change to ruminants per person was permitted in the linear programming, as pasture does not expand in this scenario. The linear programming minimized the total amount of change from current food supplies, with nutrient requirements as constraints. A maximum possible decrease of 50% was allowed for each commodity, with this constraint increased in increments of 5% until a solution was reached; this ensured that the solution to meet all nutrient requirements minimized changes from current diets. Using the per capita, per commodity amounts from the linear programming, we then calculate the required changes in domestic crop and livestock production needed to support these dietary requirements by 2050 given population growth. We lastly take these crop production amounts and allocate land use to hit these targets. We minimize the cropland needed to support this food production by placing crops in grid cells that have the highest yields until production targets are achieved. This optimization does not consider food prices given future uncertainties in how these could change; instead, we minimize changes to current diets to meet requirements while minimizing potentially culturally insensitive changes, and minimizing any changes to affordability. Explicitly accounting for affordability in this optimization could be explored in future analysis, although optimizing to minimize costs would probably result in larger changes from current diets to meet requirements.
The linear programming fixes as constant the per person amounts of FAOSTAT FBS commodities that are not grown domestically. It also fixes as constant commodities that we do not simulate owing to a lack of historical evaluation data, primarily fish products and the ‘Meat, Other’ commodity. In total, 55 commodities are eaten, processed into food or used as livestock feed in the FBS. Thirteen of these commodities are not included in the linear programming owing to lack of data or because they are not growing domestically. Absolute amounts of these commodities therefore remain the same in the nutrition focus 2050 scenario, resulting in per person reductions in these commodities. With business-as-usual imports accounted for, this results in deficient calcium and inadequate fat at the population level; an insufficient calorie, zinc and fibre supply can also occur if assuming no access to food imports in the future. An alternative would have been to compensate in the linear programming for the per capita reductions of these commodities; this would involve introducing further changes to current food supplies, in that per capita amounts of these commodities would reduce and be compensated by others. We favoured the simpler assumption of not changing the non-simulated commodities.
Climate extremes
We drive the iFEED modelling framework with Coupled Model Intercomparison Project Phase 5 (CMIP5) climate projections19. These data were bias-corrected to more accurately represent the African climate using WFDEI data (WATCH Forcing Data methodology applied to ERA-Interim data) and the cumulative distribution function transform (CDF-t) method.
A subset of the bias-corrected CMIP5 climate data was used to assess nutrition security in an average future year and an extreme future year that represents the simultaneous largest possible shocks to the production of each crop from drought and high temperatures. We focus on RCP8.5, and this subset, to explore the plausible worst-case production shocks in this crop-climate ensemble by mid-century and how food systems might adapt to these shocks to ensure food system resilience. We assume the expanded growing areas as used in the production focus scenario to calculate crop production to determine the subset. The crop projections used to select the subset are maize, soybean, groundnut and potato—the four crops that are simulated using GLAM in this analysis that form the basis of wider crop–climate projections as in previous iFEED analysis2,17. The crop yield simulations in both production focus and nutrition focus scenarios represent adaptation to climate change through changes to planting dates, and changes to crop varieties to compensate for warming-induced reduction in the crop growing season2,17. GLAM does not simulate flood impacts; therefore, production shocks are driven by low rainfall and high temperatures. Crop yield shocks are projected to approximately double in frequency without adaptation to climate change with RCP8.52,46.
A subset of five climate models was selected for RCP8.5 by calculating the largest food production shocks (the largest difference in production between the average and worst future year, normalized by future production standard deviation; other ranking methods were tested, such as inclucing coefficient of variation, and found to result in similar climate model rankings), ranking the worst models for each crop and selecting the five models that gave the aggregate lowest ranking of production years across all modelled crops. We then calculated food production in an average future year for this subset, and an extreme future year. The extreme future year results were obtained by aggregating the simultaneous largest possible shocks to the production of each crop.
Compared with the full bias-corrected CMIP5 ensemble and the representative subset of five models used in previous iFEED analysis2, the subset used in this analysis has comparable mean changes in production:
Maize: whole ensemble −15.8%, previous subset −17.7%, current subset −16.4%.
Soybean: whole ensemble −22.4%, previous subset −24.4%, current subset −21.5%.
Groundnut: whole ensemble −11.8%, previous subset −13.2%, current subset −11.4%.
Potato: whole ensemble −16.3%, previous subset −16.9%, current subset −17.2%.
The temperature and precipitation changes in the iFEED CMIP5 subset sit within the broader range across the ensembles of CMIP5 and CMIP6 models (Supplementary Fig. 2).
Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
