Application of a Bayesian approach to quantify the impact of nitrogen fertilizer on upland rice yield in sub-Saharan Africa

Mineral fertilizer input is indispensable to offset yield stagnation in rainfed upland rice production in subSaharan 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 metaanalysis 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.