Boundary line analysis and machine learning models to identify critical soil values for major crops in Guatemala

Accurate and economically sound soil fertility recommendations are critical for ensuring profitable food production for smallholder farmers. However, such recommendations are lacking in many areas due to insufficient soil and crop response data. This study was conducted between 2016 and 2022 using boundary line analysis to evaluate crop response to soil fertility on 644 farms in Guatemala. We identify optimal soil property conditions for maize (Zea mays L.), common bean (Phaseolus vulgaris L.), and coffee (Coffea arabica L.) production in Guatemala and use crop price information to develop economically sound fertilization recommendations. We also analyzed the drivers of yield outcomes and assessed their relative importance with machine learning (ML) models. Results demonstrate that a majority of country-level data currently have sub-optimal soil nutrient levels and by optimizing nutrients, yields and profits could be improved in 64%, 51%, and 69% of maize, common bean, and coffee crops, respectively. In addition, ML underscored the central role of climate in shaping yield outcomes. Variable importance rankings from random forest indicated climate and water balance variables increased predictive accuracy more than soil parameters, highlighting the critical need for climate-smart adaptations in the region. The pairing of boundary line with ML analysis provided unique and complementary information about factors that drive yield of major crops in Guatemala. This is an effective approach for developing fertility recommendations in Guatemala and could be replicated in other countries where critical nutrient recommendations are currently lacking for closing yield gaps and identifying yield response to climatic stochasticity.