Management-oriented modeling: optimizing nitrogen management with arti®cial intelligence$

Increasing nitrate levels in groundwater have been attributed to inappropriate nitrogen (N) management. The application rates, timing, and methods of both N fertilization and irrigation are important management tools that determine and control the fate and behavior of N in soilÿplant systems. For example, multiple applications with small amounts of fertilizer (e.g. split application) usually enhance plant uptake and reduce potential nitrate leaching, although increasing costs. Precision N management requires that the models evaluate so many alternatives that traditional N models are challenged beyond their intended use. The objective of this study was to construct and test a model that searches for optimal N management to: (1) minimize nitrate leaching; (2) maximize production; and (3) maximize pro®ts. Managementoriented modeling (MOM), a dynamic simulation modeling with arti®cial intelligence optimization techniques, was developed for these purposes. MOM consists of a generator that generates a set of plausible management alternatives, a simulator that evaluates each alternative, and an evaluator that determines which alternative meets the user-weighted multiple criteria. MOM uses `hill-climbing' as a strategic search method that uses `best-®rst' as a tactical search method to ®nd the shortest path from start nodes to goals. In a maize production scenario, MOM found an optimal management solution that would have increased the pro®t from $570 to $935 haÿ1 and reduced the nitrate leaching from 36 to 7 kg N haÿ1. Goal-driven simulation of MOM o€ers new opportunities to balance N and irrigation water management while meeting multiple objectives. # 2000 Published by Elsevier Science Ltd.