AUTHOR=Jemo Martin , Devkota Krishna Prasad , Epule Terence Epule , Chfadi Tarik , Moutiq Rkia , Hafidi Mohamed , Silatsa Francis B. T. , Jibrin Jibrin Mohamed 

TITLE=Exploring the potential of mapped soil properties, rhizobium inoculation, and phosphorus supplementation for predicting soybean yield in the savanna areas of Nigeria

JOURNAL=Frontiers in Plant Science

VOLUME=Volume 14 - 2023

YEAR=2023

URL=https://www.frontiersin.org/journals/plant-science/articles/10.3389/fpls.2023.1120826

DOI=10.3389/fpls.2023.1120826

ISSN=1664-462X

ABSTRACT=Rapid and accurate soybean yield prediction at an on-farm scale is important to ensuring sustainable yield increase and food security maintenance in Nigeria. We conducted large-scale demonstration experiments to assess yield responses to the Rhizobium (Rhz) +phosphorus (P) fertilizers combination in the Savanna areas of Nigeria (i.e., the Sudan Savanna (SS), Northern Guinea Savanna (NGS), and Southern Guinea Savanna (SGS)). The experimental design had four treatments, a control without Rhz inoculant and P fertilizer, the Rhz inoculated, P fertilizer application, and the Rhz+P combination at each farmer’s site. We further deployed ensemble machine-learning technics to predict yields using mapped soil properties, weather variables, rhizobium inoculation, and P supplementation. Long-term adoption scenarios and impacts on import reductions simulation models were implemented on yield increase from the Rhz +P combination. Conditional inference regression random forest (RF) for predicting the soil, weather, and factorial variables constructed on the soybean yield responses was conducted. Scenarios analysis to simulate long-term adoption impacts on national supply demand, food security, and currency saving were implemented using the IMPACT model. Our study found that yields of the Rhz +P combination were consistently higher than the control in the three agroecologies. Average yield increases were 128%, 111%, and 162% higher in the Rhz +P combination compared to the control non-treated in the SS, NGS, and SGS areas. The highest training coefficient correlation (R2=0.75) for yield prediction was from the NGS dataset, and the lowest training R2=0.46 was from the SS samples. The results from the IMPACT model showed a reduction of 10% and 22% for the low (35%) and high (75%) adoption scenarios of soybean imports from 2029 in Nigeria. A significant reduction in soybean imports is realizable if the Rhz +P inputs are implemented rapidly at the on-farm field and massively adopted by farmers in Nigeria.