Context-specific targeting of conservation agriculture reduces short-term yield penalty risk for maize and beans in Rwanda

Context and Objective Soil erosion is a major threat to smallholder farming systems in Rwanda. Terraces, while effective, are costly and reduce cultivable land area. Conservation agriculture (CA) offers a promising complementary practice if short-term yield penalties relative to current tillage (CT) remain acceptable to farmers. We assessed the seasonal and cumulative yield performance of a CA system relative to CT, farmers’ perceptions of soil erosion control under CA, and the site-level conditions under which CA yields remain within farmer-acceptable bounds. Methods A CA system combining reduced tillage, intercropped cover crops, and residue retention in a maize–bean rotation was evaluated over six cropping seasons (three years) across 473 farms in ten agroecological zones of Rwanda. A farmer-informed threshold for acceptable CA yield performance (CA:CT yield ratio ≥ 0.90) was derived from the empirical relationship between observed yield ratios and farmers' stated adoption intentions. Random forest models were used to identify site-level drivers of CA performance, and to inform a targeting approach based on observable field characteristics. Results and conclusions On average, CA yields were 19% lower than CT for maize and 24% lower for beans, yet farmers consistently perceived erosion reduction under CA, highlighting a trade-off between short-term yield penalties and longer-term soil protection. The amount of mulch cover at planting, accumulated rainfall, and weed pressure were the key drivers of CA performance. Site-level targeting increased the probability of achieving acceptable yield performance from 30% to 50%, though only ~20% of sites were initially identified as suitable for CA. Site-level targeting offers a realistic pathway to managing the trade-off between short-term yield penalties and longer-term soil protection benefits of CA. Significance This study demonstrates that scaling complex, system-based agronomic technologies in heterogeneous farming environments requires moving beyond average treatment effects. Co-defined performance thresholds and the identification of context-specific drivers can improve targeting but are insufficient on their own. We argue that combining these with adaptive dissemination strategies is necessary to offer a more realistic pathway to impact. The targeting framework developed here is directly applicable by extension programs in Rwanda, and provides a replicable model for other complex, system-based agricultural innovations.