Digital Decision Support for Smallholder Farmers using Explainable Multi Criteria Agricultural Intelligence

Under uncertain weather, fluctuating prices, constrained liquidity and limited access to advice, smallholder farmers have to make decisions on what to grow, but many digital advisory services provide a single, black-box answer that can't be questioned by farmers and extension agents. This article introduces an explainable multi-criteria decision support system which ranks candidate crops for each plot, explains in simple language the reasons for choosing a particular crop and provides confidence in the rank. A fuzzy agro-ecological suitability is combined with a water-stress yield model which incorporates Monte Carlo rainfall and price shocks, together with seven benefit and cost criteria, Analytic Hierarchy Process (AHP) weights and TOPSIS ranking. The closeness margin between the recommended crop and the runner-up crop is distributed in a pie chart across individual criteria and Dirichlet perturbation of the weights measures recommendation stability. All data are simulated: 20 synthetic landscapes of 40×40 plots (2.5 km cells), six crops with indicative parameters, and a hypothetical expert panel (AHP consistency ratio 0.013). The recommended crop for each landscape was groundnut on 37.5%, sorghum on 31.0% and pearl millet on 17.4%. It produced a mean net return of 210 USD/ha and has a loss probability of 18.7% under out-of-sample seasons with 25% fatter rainfall and price shocks than assumed during the design process, while a traditional rule of thumb has a mean net return of 67 USD/ha and a loss probability of 38.9%. The Shapley attributions summed exactly to the closeness margin (maximum error 5.6e-16), and the mean stability was 0.84, with 53.3% of plots above 0.9. These results demonstrate the mechanics and behaviour of the framework in a controlled setting; they are not field evidence.