FIELD - A summary simulation model of the soil-crop system to analyse long-term resource interactions and use efficiencies at farm scale

Resources for crop production are often scarce in smallholder farming systems in the tropics, particularly in sub-Saharan Africa (SSA). Decisions on the allocation of such resources are often made at farm rather than at field plot scale. To handle the uncertainty caused by both lack of data and imperfect knowledge inherent to these agricultural systems, we developed a dynamic summary model of the soil–crop system that captures essential interactions determining the short- and long-term crop productivity, while keeping a degree of simplicity that allows its parameterisation, use and dissemination in the tropics. Generic, summary functions describing crop productivity may suffice for addressing questions concerning trade-offs on resource allocation at farm scale. Such functions can be derived from empirical (historical) data or, when they involve potential or water-limited crop yields, can be generated using process-based, detailed crop simulation models. This paper describes the approach to simulating crop productivity implemented in the model FIELD (Field-scale Interactions, use Efficiencies and Long-Term soil fertility Development), based on the availability of light, water, nitrogen, phosphorus and potassium, and the interactions between these factors. We describe how these interactions are simulated and use examples from case studies in African farming systems to illustrate the use of detailed crop models to generate summary functions and the ability of FIELD to capture long-term trends in soil C and crop yields, crop responses to applied nutrients across heterogeneous smallholder farms and the implications of overlooking the effects of intra-seasonal rainfall variability in the model. An example is presented that evaluates the sensitivity of the model to resource allocation decisions when operating (linked to livestock and household models) at farm scale. Further, we discuss the assessment of model performance, going beyond the calculation of simple statistics to compare simulated and observed results to include broader criteria such as model applicability. In data-scarce environments such as SSA, uncertainty in parameter values constrains the performance of detailed process-based models, often forcing model users to ‘guess’ (or set to default values) parameters that are seldom measured in practice. The choice of model depends on its suitability and appropriateness to analyse the relevant scale for the question addressed. Simpler yet dynamic models of the various subsystems (crop, soil, livestock, manure) may prove more robust than detailed, process-based models when analysing farm scale questions on system design and resource allocation in SSA