MAIZE YIELD MAPPING IN GHANA USING MACHINE LEARNING

This is a predictive analysis on maize grain yield in Ghana to establish the baseline profile of maize yields in the country and to inform further research actions. Variables used as predictors include soil (physical and chemical properties), and weather (precipitation).

  • Completed: January 2021
  • Context: IFDC-Fertilizer Research and Responsible Implementation
  • Tools: R Programming

VARIABLE IMPORTANCE PLOT AFTER MODEL FITTING

A Random Forest regression model was fitted on a dataset containing soil and weather variables as explanatory variables with Maize Yield as the dependent variable. It was found that soil properties play a major role in maize yield as shown in the variable importance plot.

MODEL EXPLAINED VARIANCE

The set of explanatory variables could explain about 40% of the variability in the maize yield.

MAIZE YIELD PREDICTION USING RANDOM FOREST

The fitted Random Forest model was used to make predictions Ghana to derive a map of maize yields based on explanatory variables at grid locations.