Understanding the Role of Operations Research Models for Short and Long Term Systems Management in Natural Resource and Other Systems

The focus in this research is the interaction between a shorter-term problem as part of a longer term/larger system. This arises within the natural resource industries but is quite general. Hierarchical planning systems typically use simple long term models to define goals and constraints for shorter term optimization problems. The solution to the short-term problems becomes the control decisions for the upper level. Complex systems must often be decentralized spatially and temporally both to comprehend and to control. Decentralization poses significant questions how to combine the individual optimizations to produce system optimality. With different managerial views, it isn't clear that optimality can be easily defined. Decentralization is particularly problematic when there are short time horizons due to the evaluation framework or to discounting of future costs and benefits. In forestry, highgrading implies a short term optimization of the harvest which can maximize immediate revenue from the land but render future harvests uneconomic. Long-term forest management optimizes management schedules for classes of timber on classes of land. However, implementation is via short term problems such as road building, access plans (5-20 years), and shorter term harvest plans ( 1-5 years). There can be strong tendency for the short-term to "highgrade" the long-term harvest. It is possible to highgrade many different manufacturing settings other than forestry. For example, sawmills are typically controlled by a series of optimizers, each of which optimize the individual log breakdown with respect to a product price list. However a problem that occurs is the optimization produces a product mix that does not match demand We intend to develop a number of situations, primarily involving forestry and forest products manufacturing, where we can combine optimization models to try to understand the practical implications. We believe that some simplified, large-scale approaches using concepts of approximate dynamic programming can help. We hope then to test these insights on realistic scale problems.