POTENTIAL FOR USING LONG-TERM FIELD-RESEARCH DATA TO DEVELOP AND VALIDATE CROP SIMULATORS

At the time when many long-term field experiments were initiated, the principal method of analyzing crop yield data was to use multiple regression analysis. This technique is based on the assumption that the interacting effects of environmental factors on plants can be described with empirical additive or multiplicative models. We now know that such models do not adequately describe the interactions between factors. In the early days of agricultural experimentation, when gross nutrient deficiencies in soils were being corrected, the exact form of the model did not matter much. Now, with soil nutrient levels near optimum, additional (plant and weather) factors have to be considered in our analysis and our models must deal with the interaction between all factors realistically. The most realistic way of considering how a large number of factors affect plants is to use a limiting factor model. This enables us to develop mechanistic crop models (simulators) that simulate many of the processes going on in the plant, soil, and atmosphere. However, the validation of crop simulators requires more types of data than does the development of empirical models. Many of the data used to develop crop simulators come from studies in controlled-environment plant growth chambers, where environmental factors can be manipulated independently. However, it is widely recognized that plants grown in chambers differ from those in the field, and model parameters often have to be adjusted to fit field data. Thus, field data are essential to the final stages of model development and validation. Long-term field research is especially useful because: (i) it provides data in which some environmental factors are constant over time, and observed effects can be attributed to uncontrolled factors, and (ii) it provides data in which the unknown effects of previous treatments have been minimized. Despite these advantages, long-term field data have rarely, if ever, been used in the development and validation of crop simulators. Modelers would benefit from working more closely with scientists running long-term field experiments, showing them what additional types of data are needed and how existing crop simulators can be used to interpret results.