Regression tree analysis of soil fertility and agro-economic practices and the effects on yield in Tanzania

Food security and yield production have been extensively studied in regard to fertiliser response in the context of Sub-Saharan Africa. Soil fertility gradients within farms interact with other complex factors such as plot management, alongside bio-physical as well as socio-economic constraints, to create extensive heterogeneity in yield production. Hence, blanket recommendations regarding increasing inputs may not just lead to limited or no change in productivity but may affect long term sustainability. To try and explore the relationships between these factors, Classification and Regression Tree (CART) analysis was used to simplify the effect of plot management decisions. Data was from the focal plots of the 2016 and 2017 Tanzanian Agronomy Panel Survey (TAPS 2017; n = 580) with a range of yields and socio-economic contexts (2.10tha-1 – 3.68tha-1). Results suggested that in low resource fields, management factors are subservient to extreme soil degradation while in high resource fields good management such as optimal planting are needed for maximum predicted yield (4.48tha-1; n = 51). Boundary line analysis was conducted, and maximum yield-nutrient response values calculated. The yield gap obtained suggested only up to 60% of locally obtainable yield is reached, highlighting the necessary balance between intensity of resource use and good management for sustainable and sufficient crop production, especially in the case of extreme soil degradation.