Mineral fertilizer input is indispensable to offset yield stagnation in rainfed upland rice production in sub-Saharan Africa (SSA). The present study is the first attempt to perform a meta-analysis based on a Bayesian approach with the objective of quantitatively assessing the impact of mineral fertilizer application on upland rice yield and quantifying the effects of soil type and precipitation on the yield response to mineral fertilizer application. The data were gathered from 13 field studies on the rice variety NERICA 4 in 8 SSA countries, which provided a total of 151 paired observations. The yield gain with fertilizer application (YG) varied considerably, ranging from –0.8 to 3.0 t ha⁻¹, with an average of 0.6 t ha⁻¹. Based on the empirical relationships among the datasets, the total precipitation during the cropping season, N fertilizer application rate, and binarized soil type (i.e., low clay [≤ 20 %] and high clay [> 20 %]) were selected as key factors for the determination of YG. High clay soils exhibited higher YG than low clay soils did (i.e., 0.87 vs. 0.37 t ha⁻¹, respectively). The relationships of YG with the N fertilizer application rate and precipitation were modeled for each soil type using a Bayesian approach. The results of the Markov chain Monte Carlo simulation indicated that greater precipitation improved YG with high credibility irrespective of soil type. Additionally, a greater rate of N fertilizer application in high clay soil also improved YG with high credibility, while its contribution to YG in low clay soil was inferior. These results highlight the need to develop a field-specific nutrient management strategy for rainfed upland rice with a focus on fine-tuning the N fertilizer input based on the soil texture and expected precipitation for improving upland rice yield and nutrient use efficiency in SSA. The Bayesian procedure offers a new approach for the meta-analysis of the yield response to mineral fertilizers as affected by biophysical factors. However, including more data points in the database and additional factors in the data analysis are warranted to improve the model predictability and reliability.

矿物肥料输入对于抵消撒哈拉以南非洲（SSA）雨养旱稻产量停滞的影响必不可少。本研究首次尝试基于贝叶斯方法进行荟萃分析，目的是定量评估施用矿物肥料对旱稻产量的影响，并量化土壤类型和降水对矿物肥料产量响应的影响。应用。数据来自 8 个 SSA 国家对水稻品种 NERICA 4 的 13 项田间研究，共提供了 151 个配对观察结果。施肥 (YG) 的增产变化很大，从 –0.8 到 3.0 t ha ^-1不等，平均为 0.6 t ha ^-1. 根据数据集之间的经验关系，选择耕作季节的总降水量、施氮肥量和二值化土壤类型（即低黏土[≤ 20 %]和高黏土[> 20 %]）作为关键因素用于测定YG。高黏土比低黏土表现出更高的 YG（即 0.87 对 0.37 t ha ^-1， 分别）。使用贝叶斯方法模拟了每种土壤类型的 YG 与氮肥施用率和降水的关系。马尔可夫链蒙特卡罗模拟的结果表明，无论土壤类型如何，更大的降水量都以高可信度改善了 YG。此外，在高黏土中，较大的氮肥施用量也可以提高 YG 的可信度，而在低黏土中对 YG 的贡献较差。这些结果强调了为雨养旱稻制定田间特定养分管理策略的必要性，重点是根据土壤质地和预期降水微调氮肥输入，以提高 SSA 旱稻产量和养分利用效率。贝叶斯程序提供了一种新方法，用于对受生物物理因素影响的矿物肥料的产量响应进行荟萃分析。但是，需要在数据库中包含更多数据点并在数据分析中加入更多因素，以提高模型的可预测性和可靠性。