Intensifying rice production to reduce imports and land conversion in Africa

Site selection

Our study focused on the 15 rice-producing countries in Africa, including Egypt in North Africa, Burkina Faso, Côte d’Ivoire, Ghana, Mali, Niger, Nigeria, and Senegal in West Africa, and Ethiopia, Kenya, Madagascar, Rwanda, Tanzania, Uganda, and Zambia in East Africa⁷¹ (Supplementary Information Text Section 1 and Supplementary Fig. 1 and Table 1). These countries account for 65% and 80%, respectively, of the total harvested rice area and production in Africa (average from 2018-2020)⁶. The 15 countries portray the diversity of rice cropping systems and agro-ecological zones where rice is grown across Africa. We focused on the three primary types of rice production environment in Africa: irrigated rice, rainfed lowland rice, and rainfed upland rice production²⁷ (Supplementary Tables 2 and 3). We noted that due to the small rainfed rice area, only irrigated rice was considered in Egypt, Kenya, Madagascar, Niger, and Rwanda. Similarly, only rainfed rice (lowland and/or upland) was considered in Côte d’Ivoire, Ethiopia, Uganda, and Zambia. Both irrigated and rainfed rice were considered in the remaining six countries: Burkina Faso, Ghana, Mali, Nigeria, Senegal, and Tanzania^4,24,57,72 (Supplementary Information Text Section 2 and Supplementary Table 1). Overall, we included a total of 20 country-water regime combinations in our study.

We selected a number of representative sites following the Global Yield Gap Atlas (GYGA) protocol (www.yieldgap.org)^28,73,74. Briefly, the Spatial Production Allocation Model map (SPAM 2010; www.mapspam.info), together expert knowledge from colleagues from Africa Rice Center (www.africarice.org) and national partners, were used to identify the distribution of the rice harvested area in each of the 20 different country-water regime combinations⁷⁵. For each of the country-water regimes, we selected one or more reference weather stations (RWS) based on the current distribution of weather stations, rice harvested area, and a climate zone (CZ) scheme that accounts for spatial variation in three key parameters affecting crop yield and its variability: annual growing-degree days, aridity index, and temperature seasonality^73,74.

Following this approach, we selected the CZs where rice is grown that accounts for more than 5% of the total harvested rice area for each water regime in each country^73,74. A buffer with a radius of 100 km was created around each RWS, and this circle was then clipped by the CZ where the RWS was located. We selected buffers for each country-water regime combination, beginning with the one that had the largest harvested rice area and following with the one with the second largest areas, after discarding buffers that overlapped with the selected buffers by more than 20%. This process was continued until the overall rice coverage across all the selected buffers reached at least 50% of the national total harvested rice area for each water regime. Additional RWS were created for those rice-producing areas where weather stations did not exist²¹. The final selected sites were checked by researchers (and corrected as needed) to ensure proper representation of rice production areas and revised as needed. At the end, we selected a total of 45, 45, and 26 RWS for irrigated rice, rainfed lowland rice, and rainfed upland rice, respectively (Supplementary Information Text Section 2 and Supplementary Fig. 2 and Table 2). Selected RWS buffers accounted for 38% of rice harvested area across the 15 countries. The limited coverage by RWS buffers was due to low coverage of rainfed rice compared with irrigated rice (28% and 55%, respectively). Achieving a higher coverage of the rainfed rice area was challenging given the spread of its harvested area and the lack of data (weather, soil, management, and yield) needed for the simulations. Nevertheless, the coverage can be considered acceptable considering that selected sites are located in CZs that account for 54% and 71% of rainfed and irrigated rice area, altogether covering two-thirds of the total rice harvested area across these countries (Supplementary Information Text Section 2).

Weather data

Our goal was to determine the average yield potential for each site via crop modeling. Hence, it is critical to have a reasonable number of years included in our simulations to account for the effect of year-to-year variation in weather on yield potential. Previous studies have shown that 10 years are sufficient for robust simulation of yield potential and its variations in crops grown in favorable environments, as it is the case of irrigated and rainfed lowland rice in our study^76,77. For crops grown in unfavorable environment, as it is the case of rainfed upland rice in our study, more years (15 to 20) are needed. In the present study, we simulated yield potential (or water-limited yield potential for rainfed rice) and its year-to-year variation based on 20 years of recent weather data (2000–2019) retrieved for the RWS in 10 of the 15 countries, including Burkina Faso, Ghana, Mali, Niger, Nigeria, Ethiopia, Kenya, Tanzania, Uganda, and Zambia. For the other five countries (Egypt, Côte d’Ivoire, Senegal, Madagascar, and Rwanda), our estimates of yield potential are based on 11 years of older weather data (1995–2005). Because we did not detect any trend in yield potential over time in the 10 countries for which we had weather data from 2000 to 2019, we included the estimates of yield potential from these other five countries in our analysis. Weather data used for the simulation of yield potential in all RWS in these 15 countries included daily solar radiation, maximum and minimum temperatures, precipitation, vapor pressure deficit, and wind speed. Weather data for selected weather stations were subjected to quality control measures to fill in missing data and identify and correct erroneous values by using linear interpolation to fill out missing data (https://www.yieldgap.org/methods-weather-data). Of the total of 82 selected RWS, measured daily weather data, propagated gridded weather data⁷⁸, and gridded weather data were available for 23%, 76%, and 1% of them (Supplementary Information Text Section 4), respectively.

Crop management

Given that yield potential depends on climate, including solar radiation, temperature, and water supply in the case of rainfed crops, it is important to simulate yield potential in the context of the current crop systems (as determined by the date and method of crop establishment, crop cycle length, and crop sequence) and environment (irrigated, rainfed lowland, and rainfed upland)^28,29. Simulating yield potential of rice required a thorough understanding of rice-based cropping systems in Africa, including input from local experts to determine crop calendars and dominant rice environments in each country and extensive on-the-ground data collection, including weather, soil, and management information across 15 African countries that include a total of 10 M ha cultivated with rice^6,28. Data on crop management practices for each buffer were retrieved through agronomists from AfricaRice, which is the most important rice research organization in Africa, including a vast network of researchers linked with agronomists and extension specialists, and strongly connected with policymakers in the major rice-producing countries in Africa (www.africarice.org), and national agricultural research institutes and extension agents. The requested information included dominant crop sequences, ecosystems (upland/lowland), water regime (rainfed/irrigated) and proportion of each of them to the total harvested rice area, crop establishment method (transplanted/direct-seeded), average sowing dates for both transplanted and direct-seeded rice and transplanting date for transplanted rice, and dominant rice variety name and maturity. Reported dates of establishment (either transplanting or direct seeding) correspond to the dominant establishment date of each cropping system in each region reported by local agronomists and extension agents. Rice crop calendars for representative rice cropping systems in each country are shown in Supplementary Fig. 1. In each of the buffers, we identified dominant rice cropping systems, which are characterized by ecosystem, water regime, and rice cropping intensity.

Yield potential simulation

We used the well-validated crop simulation model ORYZA v3 and site-specific data on weather, soil, and crop management practices to estimate yield potential. The model was developed to simulate the growth and development of rice and has been validated and used across a wide range of rice cropping systems^20,79,80. Given the lack of experimental data from well-managed experiments needed to calibrate rice varieties, we used generic crop parameters derived for rice varieties in Africa in previously published studies^57,81,82,83. Briefly, in these previous studies, an ORYZA model version named as ORYZA2000v2n14, which builds upon the ORYZA2000v2n13s14 version, was used to derive the genetic parameters of these rice varieties through iterating calibration and validation processes with initial values of crop parameters from a widely cultivated variety, IR72. This updated iteration includes enhancements in modeling heat sterility, cold sterility, and phenology^57,83. Specifically, ORYZA2000v2n14 incorporates features such as explicit simulation of transpirational cooling and earlier flowering in hotter climates.

In each RWS buffer, we collected data on water regime, crop establishment, rice variety name, sowing or transplanting (for transplanted rice only) date, and maturity date from local agricultural specialists. Subsequently, we employed the DRATE v2 program, which was integrated into the ORYZA v3 model, to calibrate phenology development rate parameters, including development rates for juvenile (DVRJ), photoperiod sensitivity (DVRI), panicle development (DVRP), and reproductive phases (DVRR), based on data on phenological stages and growth duration⁸⁰, assuming that the 50% flowering date was fixed to occur 30 days before the maturity date⁵⁷. The data for other variables (e.g., assuming a base temperature for development of 14 °C, a maximum optimum photoperiod of 10 h, no photoperiod sensitivity, a lower air temperature threshold for growth of 12 °C, consecutive number of days below the lower air temperature threshold that crop dies of 3 d, a critical temperature of spikelet sterility of 35.6 °C, a fraction of sunlight energy that is photosynthetically active of 0.5, and a fraction of carbohydrates allocation to stems that is stored as reserves of 0.2) were obtained from previous studies^57,81,84.

For each of the 20 country-water regime combinations, we simulated yield potential of irrigated rice (or water-limited yield potential of rainfed lowland and upland rice) for each rice cycle within the dominant cropping system (Fig. 2). We assumed no water limitation for irrigated rice, whereas simulations of rainfed lowland and upland rice accounted for the amount and distribution of precipitation and soil properties influencing the soil water balance. Water-limited yield potential simulations of rainfed rice were conducted using the assumption of a non-puddled clayey loam soil with a bund height of 25 cm for rainfed lowland rice. The effect of groundwater depth on rainfed rice yield is highly contextual, varying greatly among locations, seasons, and landscapes^57,85. We simulated water-limited yield potential for rainfed lowland rice under two scenarios of groundwater depth to account for the variety of scenarios and associated uncertainty during the entire crop cycle (shallow [40 cm] and deep [100 cm]), as upstream and downstream valley bottom groundwater depths were in this range⁸⁶. We also assumed that the area of rainfed lowland rice in each buffer is split evenly (50:50) between the two groundwater scenarios (40 & 100 cm deep lowland groundwater depths). For rainfed lowland rice, the two scenarios basically portray rainfed favorable (shallow water table) and drought-prone (deep water table) environments. Initial volumetric water content, saturated volumetric water content of ripened, saturated hydraulic conductivity of soil was assumed to be 0.57 m³ m^-3, 0.56 m³ m^-3, 10.79 cm d^-1 for rainfed lowland rice, respectively, and 0.39 m³ m^-3, 0.38 m³ m^-3, 99.77 cm d^-1 for rainfed upland rice. In the case of rainfed upland rice, water-limited yield potential was simulated with a groundwater depth during the entire crop cycle of 1000 cm in a nonpuddled sandy loam soil without a bund (Supplementary Information Text Section 3). Following previous studies, sensitivity analysis was performed to assess the sensitivity of simulated yields to assumptions on parameters related to soil water holding capacity, presence of hardpan, bunding height, and groundwater table depth⁸¹.

Validating our estimates of yield potential in Africa is challenging due to the limited availability of experimental data collected from well-managed crops that grow without nutrient limitations and kept free of weeds, diseases, and insect pests⁵⁷. We performed a cross-validation of simulated yield potential for a subset of sites for which measured yield in well-managed rice crops were available (Supplementary Table 4). In areas lacking experimental yield data, we extended the cross-validation by comparing yield potential in a particular climate zone against that reported by the Global Yield Gap Atlas for the same climate zone in other regions of the world, such as Southeast Asia, where the yield potential has been well validated. The year-to-year variation in yield potential of irrigated rice (or water-limited yield potential of rainfed rice) was assessed by determining the inter-annual coefficient of variation and semi-deviation for irrigated versus rainfed rice and rainfed lowland versus rainfed upland rice (Supplementary Information Text Section 3 and Supplementary Table 3). The computation of semi-deviation was performed with a downside risk approach using the “PerformanceAnalytics” package in R software version 4.1.2⁸⁷.

Yield gap estimation

For each of the 20 country-water regime combinations, the yield gap was determined for each cycle of rice production by difference between the yield potential (irrigated rice) or water-limited yield potential (rainfed lowland and upland rice) and the average yield of the farmers²⁹. We note that while the yield potential for five countries, including Egypt, Côte d’Ivoire, Senegal, Madagascar, and Rwanda, were based on the average between 1995 and 2005, we used the most up-to-date data to estimate the actual farmer yields for these five countries, as well as for the other 10 countries. Data on average farmers’ yields for irrigated and rainfed rice were collected separately from national statistics, previous publications and databases, and local agronomists (Supplementary Information Text Section 5 and Supplementary Table 6). All yield data reported in our study is reported as paddy rice at a standard moisture content of 140 g H₂O kg^-1 grain. The yield gap for irrigated and rainfed rice in each RWS buffer was estimated separately for each country. Average yield gap in each RWS buffer was estimated by weighting yield potential and actual farmers’ yield based on the proportion of harvested rice area of each cycle in each cropping system (Supplementary Fig. 1).

Current (2018-2020) and future (2050) rice demand and self-sufficiency

We considered the average annual domestic rice demand during the 2018–2020 period as the baseline for our study (Fig. 1). Current national domestic rice demand for each country was estimated based on the average annual national rice production, imports, exports, and stock change during the period from 2018 to 2020⁵ (Supplementary Table 5). We estimated future demand for rice by 2050 in each country by multiplying their projected populations, based on the medium fertility variant of the UN population prospects³, and the per-capita rice demand in 2050. The latter was calculated based on the relative change in average per-capita rice demand between the baseline period (2018–2020) and the year 2050, which was derived for each country from the International Model for Policy Analysis of Agricultural Commodities and Trade (IMPACT) database⁸ (Supplementary Table 5). The IMPACT projections account for various socio-economic factors, including effective consumer prices, region-specific population dynamics, income growth rates, government blending mandates, energy prices, producer subsidy equivalents encompassing subsidies and trade measures, commodity-specific indices for all commodities, as well as price and income elasticity considerations⁸. For all the 15 countries, the total rice demand that is projected for the year 2050 is predicted to be higher than the current demand (2018–2020), with the size of increase ranging from 79% to 469% (Supplementary Table 5). The increasing demand for rice is driven by both projected increase in population and per-capita rice consumption in Kenya, Madagascar, Rwanda, and Zambia, whereas in the rest of the countries, demand increase is being driven primarily by the projected increase in population. In our study, all rice yield, production, per-capita rice demand, and total rice demand were reported as paddy rice at a standard moisture content of 140 g H₂O kg⁻¹ rice grain. The per-capita rice demand was converted to paddy rice by dividing initially reported milled rice from the USDA database by the respective country’s rice milling rate^5,8, ranging from 0.63 to 0.69 across countries (Supplementary Table 5).

To help identify hotspots for yield intensification and/or area expansion on a country basis, we calculated current and future (year 2050) SSR and rice deficit in each of the 15 selected countries (Fig. 3). The SSR was calculated for each country by dividing the annual rice production by the annual rice demand, while net import or exports was determined by subtracting the annual rice production from the annual rice demand. The rice deficit was calculated as the difference between projected rice demand and extrapolated rice production, which was estimated as the product of extrapolated rice yield by 2050 and current rice area. To extrapolate rice yield by 2050, we assumed a continuation of the historical yield increase rate observed during the past three decades in each country, until the rice yield reaches the attainable yield, estimated as 70-80% of the simulated yield potential (see next section), which was the case of Côte d’Ivoire and Uganda (Supplementary Table 7).

Scenario assessment

In our study, we determined rice production and imports requirement in Africa for different scenarios of yield intensification and rice area expansion (Fig. 4; Supplementary Figs. 5 and 6). Following previous studies, the exploitable yield gap was defined as the difference between 80% of yield potential (irrigated rice) or 70% of water-limited yield potential (rainfed rice) and the current average farmer yield^19,88,89. Following prior assessments on food supply-demand scenarios^10,43,90, we selected 2050 as the target year for our evaluation. This 30-year timeframe strikes a balance between minimizing the long-term impact of climate change on rice yields and cropping systems^63,66, while providing enough time to plan for structural changes, implement technologies, mitigate the risk of unpredictable events (such as economic downturns in any given year), and formulate short- and long-term policies and orient agricultural R&D programs to eliminate the exploitable yield gap.

In the case of yield intensification, we estimated rice yield under different levels of exploitable yield gap, ranging from 52% (current) to full closure of exploitable yield gap. For the purpose of area expansion assessment, we considered three scenarios for rice area expansion by 2050: (i) expanding harvested rice area by 0.4 M ha annually, which is the historical rice area expansion rate in Africa over the past three decades, (ii) expanding harvested rice area at an annual rate of 0.2 M ha, representing a 50% reduction in the historical expansion rate, and (iii) expanding harvested rice area at an annual rate of 0.6 M ha, equivalent to a 50% increase in the historical expansion rate. Note that in our analysis, we assumed that there would be no change in the fraction of irrigated rice area or changes on crop intensity.

Our scenario assessment focused on estimating the aggregated rice SSR, net rice import, and financial expenditure associated with importing rice in Africa for the current baseline and different scenarios of yield intensification and area expansion by 2050. In our scenario assessment, we considered all 58 countries and disputed territories in Africa⁷¹. Unfortunately, the data needed to estimate rice demand by 2050 was not available for 21 countries and disputed territories. Since these countries and disputed territories account for 5% of the current rice demand estimated for the other 37 countries, we simply multiplied the predicted annual rice demand from the 37 countries by 1.05 to determine the total annual rice demand in Africa by 2050 (Supplementary Table 5). In the case of estimation of total rice production for Africa, our study included 15 major rice producing countries, which altogether account for 80% of total rice production in Africa. For the remaining 43 countries and disputed territories that were not included in our analysis, we assumed the average yield gap derived from the 15 countries, only considering those countries that produce rice (Supplementary Fig. 5). To determine the total cost associated with rice import, we multiplied the volume of rice imported by Africa with the rice market price. Similarly, we calculated the money earned from export (surplus) by multiplying the annual rice export by the rice market price. The rice market price (US$289 Mg⁻¹ paddy rice) was derived from the World Bank annual price of rice (Thai 5%) between 2018 and 2020⁷ (Supplementary Fig. 6).

Finally, a sensitivity analysis was conducted to understand the potential impact of climate change on the outcomes from our scenario assessment. Previous studies have reported 5−10% reduction in yield and land suitability for rice in Africa due to climate change^57,66,67,68. Hence, we re-estimated rice production and SSR by year 2050 for each of our yield intensification scenarios under current area expansion rate for different combinations of yield potential reduction (-5% and −10%) and limited cropland expansion (-5% and −10%) (Supplementary Information Text Section 6 and Supplementary Fig. 7).