Development of web-based software for identifying adaptability of cereal varieties to local conditions

Abstract

Data from variety trials provide national or regional estimates of variety performance and these averages represent information that is widely applicable. However the averages can conceal differences in variety responses, either to permanent features of individual fields, e.g. their location, soil type, PK status; or to management factors which may be under farmers' control, e.g. sowing date, manuring rate.

Estimation for local environments is not attempted for several reasons, but one of the main reasons is analytical complexity. Problems arise from:

• Trials data that are heavily influenced by seasonal and location effects, and which also involve the introduction of new varieties and the discarding of old varieties over time;
• The strong association that often occurs in trials between environmental factors, such as between soil N status and the rate of applied N, whose confounding effects are difficult to disentangle;
• The uncertainty about conditions in the local environment where variety recommendations are needed, e.g. the information about the soil N status of a field is likely to be imprecise;
• The absence of effective methods for putting all of the information together as useful decision tools for farmers
• Developments in statistics and computing e.g. Bayesian inference and Markov chain Monte Carlo computational methods, suggest that some of these problems can now be overcome.