3. Application of a machine learning tool to cocoa in Ghana
Random Forest classification
4. Random Forests for classification
• A forest is an ensemble of
trees. The trees are all
slightly different from one
another.
• The output is the mean
classification
• Very robust against
overfitting
Is the soil good?
Is the dry
season long?
Is the heat
strong?
One decision tree
Source: Criminisi et al 2013
10. Current distribution of suitability classes
for cocoa
AEZ Bioclim A Bioclim B Soils
Type 1 Low annual precipitation Strong dry season Below average soils
Type 2 Low temperatures Long dry season Average soils
Type 3 Elevated temperatures Reliable precipitation Average soils
Type 4 High temp Low seasonal variation Above average soils
11. Current distribution of suitability classes
for cocoa
• MSNW
Moist semi-decidious North
West
• MSSE
Moist semi-decidious South-
East
• ME
Moist evergreen
23. Conclusion
• Cocoa production is shaped by climate and soils
• Cocoa soil characteristics are different from other soils in the
country
• Results show four distinct production zones that align with
ecological zones in Ghana:
Moist semi-deciduous, subtypes NW, (central),SE
Moist evergreen
• The moist evergreen climate (type 4) will be the dominant climate in
the future
• The moist semi-deciduous (type 1) region in the North West will
become marginal
24. Conclusion
• Yield and production will be affected by climate
change
• Adaptation has to be a priority for the cocoa
sector