A Bayesian modeling approach to site suitability under conditions of uncertainty.

Precision agriculture in the developed world tends to rely on the collection and analysis of significant amounts of data using sophisticated equipment. For smallholder farmers in the developing world this data is simply not available. However, data does exist that can be used to perform analysis at sub-farm level. This data includes spatial information in the form of geographical information databases on climate and soils, crop trial data, expert knowledge and farmer knowledge. Often this data and knowledge is sparse and only reliable to a certain extent. Most statistical techniques are unworkable in this situation, with too many assumptions needing to be made in order to perform any analysis. Bayesian modeling techniques provide a simple yet robust way of combining existing data and knowledge in the form of probabilities, whilst keeping uncertainties in the data and knowledge explicit. A tool called CaNaSTA (Crop Niche Selection in Tropical Agriculture) has been developed demonstrating these techniques for recommending forage crops to farmers and their advisors in Central America. This paper discusses the development of this tool and compares Bayesian modeling with conventional methods, and shows that Bayesian techniques provide the best model in the context of selecting suitable crops using sparse and uncertain data.