Development of such crop models will involve basic research into the effect of subfield variation in availability of nutrients for crop growth, the effect of variation in soil quality on crop nutrient uptake and pest (insect, weed, and disease) prevalence, and crop-pest ecosystem interactions. It will also involve applied empirical research attempting to quantify crop yield responses to nutrients, soil quality, and pest prevalence across the growing season for important crops in different locations. It will require interdisciplinary study involving agronomists, entomologists, economists, plant pathologists, weed scientists, and ecologists. Developing such models will not be an easy undertaking, and efforts requiring coordination across many disciplines are essential.
The current research model for crop sciences employs carefully controlled research plot experiments with replicated block designs in which only the factor of interest (i.e., fertilizer rate) is allowed to vary. Because precision agriculture has already been identified as a technology of some promise, detailed small-plot studies and technology-adaptation experiments may not be necessary (Gomez and Gomez, 1984; Gotway Crawford et al., 1997). Further, the system may need to evolve so that innovation and learning can exploit both traditional research plot experiments and information captured from actual field operations through precision agriculture. Precision agriculture has the potential to collect many layers of data for entire fields and record detailed variation in many variables that affect crop growth. Thus, precision agriculture could change the research paradigm from station-based plot studies to farm-based studies by (a) using more complex experimental designs, such as incomplete block designs, row-column designs, nearest neighbor designs, and split plot designs; (b) specifically incorporating spatial variability in experimental designs; (c) supplementing mean-based analyses with comparisons of entire distributions; and (d) using statistical methods such as multiple regression (Gotway Crawford et al., 1997). Under this paradigm, groups of producers would collect precision data; agronomic researchers would analyze the data with statistical methods to estimate how small changes in manageable factors affect crop yield in various resource and weather situations. The incentives for and obstacles to producer data sharing need to be fully explored and carefully understood if such cooperative on-farm research is to succeed (see sections on data ownership and privacy).
Improved farm management methods are equally necessary. At present, farm management models are based on budgeting, which assumes fixed input-output ratios. Recommendations based on improved crop models will likely need to be derived from nonlinear optimization, such as profit maximization, because such crop models will likely be characterized by variable input-output ratios which may change throughout the growing season. Stochastic optimization frameworks may be needed to take into account the random occurrence of rainfall. Dynamic optimization may also be needed to take into account changes in recommendations as the growing season progresses.
As discussed previously, it is not clear a priori that precision agriculture