Adopting yield-improving practices to meet maize demand in Sub-Saharan Africa without cropland expansion

The study addresses how Sub-Saharan Africa (SSA) can meet rapidly rising maize demand—projected to increase by 233% by 2050—without further cropland expansion or costly imports. Despite significant growth in maize area over the past two decades, yield increases have been minimal, with average farm yields (~2 t ha−1) far below climatic yield potential across SSA. Previous regional yield gains in SSA lag far behind those in Southeast Asia and South America. The authors posit that narrowing the large yield gap on existing cropland through improved agronomic management is a viable strategy to boost production, alleviate pressure on natural ecosystems, and reduce dependence on imports. The study’s purpose is to identify, across diverse environments, the suites of management practices that consistently raise on-farm maize yields and to quantify their potential contribution to regional self-sufficiency by 2050.

The paper notes extensive evidence that SSA maize yields are far below potential and that past analyses were constrained by limited sample sizes, narrow environmental coverage, and focus on individual practices. Prior work often used farmer-reported yields, which can be biased upward, especially on small plots. Studies have examined yield variability and constraints in Ethiopia, Ghana, and Kenya, emphasizing nutrient management and variable responses to fertilizers and amendments. The authors contrast broad, ideology-labeled approaches (e.g., conservation agriculture, agroecology, climate-smart/regenerative/nature-based/digital agriculture) with empirically validated, context-tuned agronomic practices that directly raise yields. They argue for evidence-driven prioritization of management practices—nutrient management, cultivar choice, establishment timing and density, and pest/weed control—supported by large, measured-yield datasets and spatially explicit analysis.

Data comprised 14,773 measured-yield observations from smallholder maize fields (pure stands) collected by One Acre Fund from 2016–2022 across five SSA regions: north-central Nigeria (NG), Rwanda–Burundi (RW–BI), central Zambia (ZM), southwest Tanzania (TZ), and eastern Uganda–western Kenya (UG–KE). About half of the fields were from One Acre Fund participants, increasing management variability. Yield, plant density, and row spacing were measured from two 36 m² boxes per field; geolocation was recorded for 70%, with town/district used otherwise. Farmers reported sowing/harvest dates, cultivar, fertilizer types and quantities (inorganic/organic), fertilization method, liming, weeding, pesticide use, and incidences of pests/diseases/Striga and other adversities. Quality control removed implausible values and missing key variables, yielding 14,773 records. Inorganic fertilizers were converted to elemental nutrient rates; organic inputs encoded as binary/continuous variables. Cultivars were classified as hybrids or open-pollinated varieties (OPVs). Hybrid metadata (maturity, disease tolerance, release year) were compiled from seed catalogs. Fertilizer placement was categorized as in-hole versus surface. Weeding frequency was binned (0, 1, ≥2). Sowing date was expressed as deviation from the climate zone-season average. To control confounding due to environmental heterogeneity, fields were stratified into 25 production environments using a high-resolution (≈1 km) version of the Global Yield Gap Atlas (GYGA) climate zone framework based on growing degree days, aridity index, and temperature seasonality; zones with two seasons were split by season. Additional covariates included elevation, seasonal and phase-specific precipitation (CHIRPS), soil water-holding capacity (World Soil Information), soil properties (iSDA: clay, pH, organic C, effective CEC), and topographic wetness index. Within each climate zone, conditional inference tree (CIT) models identified management practices significantly associated with yield, using stringent significance thresholds (typically α=0.01) and controls to avoid overfitting (limits on depth and node sizes). The frequency with which practices appeared across zones quantified their relative importance. To estimate yield gains from combined practices, farmers were grouped by technological level based on terciles of key continuous variables: baseline (OPV, low N, low plant density, average/late sowing), hybrid+high N, hybrid+high N+high density, and hybrid+high N+high density+early sowing. A linear mixed-effects model (square-root-transformed yield) estimated marginal means by technology level with climate zone and year as random effects. A Gradient Boosting Machine (GBM) was trained on all fields with management and environmental predictors to validate CIT findings, generate SHAP values for individual practice effects, and support extrapolation. Model tuning used spatial cross-validation (k-fold nearest-neighbor distance matching). Two technology scenarios (baseline vs intensified) were predicted across SSA maize areas with similar environmental conditions (area of applicability constrained by a dissimilarity threshold), incorporating seasonal climate variables. Scenario assessment for SSA maize production and self-sufficiency to 2050 used FAOSTAT (2018–2020) for current production/demand, UN population projections (medium variant), and IMPACT projections for per-capita demand changes. Maize area was held constant at 40 million ha. Two 2050 scenarios were evaluated: continuation of historical yield gain rates (27 kg ha−1 yr−1) and accelerated gains achieving the yield gap closure seen among high-technology farmers. Yield potential used GYGA estimates. Costs of adopting the highest technology level were estimated from 2023 One Acre Fund prices (Kenya, Rwanda, Tanzania, Burundi) plus program operational delivery costs, yielding an approximate per-hectare investment requirement.

- Management practices with consistent positive effects on yield included higher nitrogen (N) and phosphorus (P) fertilizer rates, in-hole fertilizer placement, hybrid seed use over OPVs, earlier sowing dates, higher plant densities, effective pest control, and weeding. Synergistic interactions were observed (e.g., hybrid benefits greatest with early sowing and higher N). - No detectable effects of specific hybrid traits (cycle duration, release year, disease tolerance) on on-farm yield, likely due to overriding effects of management quality. - SHAP analyses (GBM) corroborated CIT results, showing strong positive contributions from N and P rates, proper placement, hybrid use, higher plant density, earlier sowing, weeding, and pesticide use. - Yield benchmarks: regional rainfed yield potential ≈10.6 t ha−1; average actual yield in study regions ≈1.7–2.0 t ha−1; well-managed on-station trials achieved ≈7.5 t ha−1 (~70% of potential). - Technology package impacts (marginal mean yields): baseline ≈1.8 t ha−1; hybrid+high N (~50 kg N ha−1) yielded 61% higher than baseline; adding high density and early sowing achieved ≈4.3 t ha−1 (2.4× baseline), narrowing the yield gap by ~30% (~2.5 t ha−1). - Regional extrapolation suggests adoption of improved practices could raise SSA maize production from ~80 to ~168 million tons on existing area by 2050, approaching self-sufficiency (SSR ~0.9) and reducing import or land expansion needs. - Without accelerated yield gains, meeting 2050 demand would require ~28 million additional hectares or ~76 million tons of imports (assuming fixed current yields and areas). Historical yield growth rates: SSA ~27 kg ha−1 yr−1 vs SEA 113 and SAM 142; maize area expanded by ~935,528 ha yr−1 in SSA (2000–2020).

Findings affirm that straightforward, empirically validated agronomic improvements—nutrient management (higher N and P, proper placement), hybrid seed adoption, timely planting, optimal plant density, and effective pest/weed control—can double maize yields in SSA, offering a near-term pathway to enhance food security and farmer incomes without cropland expansion. While diverse approaches (e.g., conservation, agroecology, climate-smart/regenerative/nature-based, digital agriculture) are promoted for multiple societal goals, the authors caution that broad labeling may distract from immediate, context-specific practices with proven yield impacts. Despite reliance on observational data, the study’s strengths include measured yields, large sample size, extensive environmental coverage (25 zones, seven countries), robust stratification, and agreement between CIT and GBM analyses. Scaling improved practices could lift average yields to ~4.2 t ha−1 by 2050—tripling current annual yield growth rates and aligning with gains achieved in South America and Southeast Asia—substantially reducing the need for imports or expansion into marginal lands. Realizing these gains requires AR&D prioritization, extension, and enabling policies, as adoption entails costs (~US$200–250 ha−1) and depends on improved access to inputs, markets, finance, insurance, and infrastructure. Climate change and expansion into lower-productivity lands could further increase the urgency to intensify on existing cropland.

Improved agronomic management in SSA maize—hybrid seed use, higher and well-placed nutrient inputs, earlier sowing, optimal plant density, and pest/weed control—can feasibly double on-farm yields and deliver roughly an additional 82 million tons of maize on current cropland, moving the region close to self-sufficiency by 2050 without expanding maize area. The paper demonstrates a data-driven, scalable framework to identify yield-improving practices across diverse environments and provides clear targets for AR&D investments and policy support. Future work should focus on fine-tuning management for local soil fertility gradients and farming systems, addressing remaining yield gap components (beyond the ~30% closure achieved here), and integrating risk management and climate resilience. Sustained investment, input access, and supportive institutions are essential to accelerate yield gains to levels observed in other tropical/subtropical regions.

- Observational design limits causal inference; however, large measured-yield dataset, stratification, and model convergence (CIT and GBM) mitigate concerns. - Extrapolation to all SSA assumes environmental similarity and consistent responses; validated via spatial ML and prior evidence but still subject to uncertainty. - Potential confounding due to correlated inputs (notably N and P rates, r≈0.72) limits separation of nutrient-specific effects. - Soil databases may lack sufficient granularity; soil effects may be underrepresented. - Scenario assumptions: fixed maize area (40 M ha), no explicit negative climate change impacts, and equal productivity for any newly expanded cropland—all likely optimistic; real-world outcomes could be more challenging. - Farmer-reported management data may include errors despite quality control; nonetheless, sample size and consistency across zones increase robustness.