A sensitivity analysis of the prediction of the nitrogen fertilizer requirement of cauliflower crops using the HRI WELL_N computer model

HRI WELL_N is an easy to use computer model, which has been used by farmers and growers since 1994 to predict crop nitrogen (N) requirements for a wid e range of agricultural and horticultural crops. A sensitivity analysis was carried out to investigate the model predictions of the N fertilizer requirement of cauliflower crops, and, at that rate, t he yield achieved, yield response to the fertilizer applied, N uptake, NO3- N leaching below 30 and 90 cm and mineral N at harvest. The sensitivity to four input factors - soil mineral N before planting, mineralization rate of soil organic matter, expected yield and duration of growth - was assessed. Values of these were chosen to cover ranges between 40% and 160% of values typical for field crops of cauliflowers grown in East Anglia. The assessme nts were made for three soils - sand, sandy loam and silt - and three rainf all scenarios - an average year and years with 144% or 56% of average rainf all during the growing season. The sensitivity of each output variable to e ach of the input factors (and interactions between them) was assessed using a unique 'sequential' analysis of variance approach developed as part of t his research project. The most significant factors affecting N fertilizer requirement across all soil types/rainfall amounts were soil mineral N before planting and expecte d yield. N requirement increased with increasing yield expectation, and dec reased with increasing amounts of soil mineral N before planting. The respo nses to soil mineral N were much greater when higher yields were expected. Retention of N in the rooting zone was predicted to be poor on light soils in the wettest conditions suggesting that to maximize N use, plants needed to grow rapidly and have reasonable yield potential. Assessment of the potential impacts of errors in the values of the input fa ctors indicated that poor estimation of, in particular, yield expectation a nd soil mineral N before planting could lead to either yield loss or an inc reased level of potentially leachable soil mineral N at harvest. The research demonstrates the benefits of using computer simulation models to quantify the main factors for which information is needed in order to pr ovide robust N fertilizer recommendations.