A comparison of dynamic programming and A* in optimal forest stand management

Abstract

A model that adds a forward value estimator to traditional forward recursion dynamic programming (DP) is developed. State classes are used to store continuous values for stand basal area and stem density. A simple forward heuristic that estimates future value from deviations within a node is used in the model in combination with accumulated value to determine path choice (A*). A loblolly pine growth and yield model with both pulpwood and sawtimber product outputs is used in the model. The model is tested on combinations of initial planting density, site index, and interest rates to compare A* to DP in optimization of thinning age, thinning intensity, and rotation age. Results from tests using single-year stage intervals show increases of 0.0% to 1.2% in net present value for A* compared to DP. An A* approach increases computer program size very little but requires additional time for parameter searching. However, A* always generated solutions at least as high as DP.