Abstract:
Soybean is one of the crops grown in Ghana that generate income and serve as
a source of protein, animal feed, and food security. However, yields are low,
averaging 1.3 Mt/h compared to a potential yield of 3.0 Mt/h, despite an 8 kg/h
increase in fertilizer application. The study aimed to analyse how soybean yields
in Ghana respond to fertilizer application through spatial methods. We
employed a yield modeling approach utilizing data from agricultural trials. To
evaluate the variability in observed yields, we used the Multiple Linear
Regression-Akaike Information Criterion (MLR-AIC). Additionally, we
applied a Random Forest spatial prediction framework to analyze and map the
predicted yields. The final and best MLR model achieved one (R=51%),
indicating that the model explains about 51% of the variation in the dependent
variable. A detailed regression analysis revealed that calcium (Ca), sodium (Na)
and minimum temperature (Tmin) were the variables that had a significant
negative (<1000 kg/ha) impact on yield. pH, carbon and potassium were the
variables with the greatest positive impact on yield (>1000 kg/ha). The
predicted soybean yield based on the trained random forest model ranged from
1.0 to 2.2 t/ha. The forecast remained at 1 to 1.8 t/ha in the northern parts and
2.0 to 2.2 t/ha for the southwest. Policy makers in Ghana need to consider highpotassium
fertilizers and maintain sound agronomic practices to increase
soybean yields.
