Abstract
Forest Inventory and Analysis uses poststratification to calculate resource estimates. Each county has a different stratification, and the stratification may differ depending on the number of panels of data available. A ?5 by 5 sum? filter was passed over the reclassified forest/nonforest Multi-Resolution Landscape Characterization image used in Phase 1, generating an image in which each pixel represents the count of forested pixels inside a 5 by 5 window. The forested pixel count ranges from 0 to 25 or 26 classes. In the next step, the ground plots are overlaid on the class map generated by the 5 by 5 window. The objective is to find the break points in the 26 classes that minimize the difference in the number of acres/plot between strata while simultaneously maximizing the number of strata. These are conflicting goals. More strata imply larger deviances between the strata. Also, the stratum must have contiguous classes with at least four plots. This is a nonlinear integer programming problem. Because software is not readily available to solve a nonlinear integer programming problem, the problem was reformulated to finding the shortest path through the network. For each county, the optimal one, two, three, four, five, six, and seven strata are found, and various heuristics for determining the final solution are investigated and compared.
Parent Publication
Citation
King, Susan L. 2005. Stratum Weight Determination Using Shortest Path Algorithm. In: Proceedings of the fifth annual forest inventory and analysis symposium; 2003 November 18-20; New Orleans, LA. Gen. Tech. Rep. WO-69. Washington, DC: U.S. Department of Agriculture Forest Service. 222p.