Fig. 4: Effect of individual crop management practices on maize yields in Sub-Saharan Africa based on a machine-learning model. | Nature Communications

Fig. 4: Effect of individual crop management practices on maize yields in Sub-Saharan Africa based on a machine-learning model.

From: Adopting yield-improving practices to meet maize demand in Sub-Saharan Africa without cropland expansion

Fig. 4

Shapley additive explanation (SHAP) values for the nine agronomic practices with the greatest impact on maize yields. SHAP values were computed based on a Gradient Boosting Machine model trained with the 14,773 maize fields included in the field-level farmer database and biophysical covariables listed in Table S2. Each SHAP value shows how much each agronomic practice contributes to the maize yield variation across smallholder fields as predicted by the machine-learning model. In (a), (b), (d), and (e), blue lines were fitted with LOESS regressions. In (c), (f–h), the central line of each boxplot indicates the median, the box encompasses the interquartile range, and the whiskers extend up to 1.5 times the interquartile range while more extreme values are depicted individually as outliers. All maize fields (n = 14,773) are included in (a–c) whereas (d) includes 10,166 fields, (e) 11,179, (f ) 13,543, (g) 10,967, and (h) 9,088 due to missing information for some maize fields. N refers to nitrogen and P to phosphorus. Source data are provided as a Source Data file.

Back to article page