Application of Most Similar Neighbor Inference for Estimating Marked Stand Characteristics Using Harvester and Inventory Generated Stem Databases

Abstract

The purpose of this study was to develop and test the application of non-parametric Most Similar Neighbor Inference (MSN) for wood procurement planning. An application developed using this method would be a part of a stem database in Finnish forest enterprises and could predict characteristics of a marked stand with accuracy demanded by bucking simulation. A stem database includes representative samples of stands and stems, applications to control and update data and applications to utilize the database. The study materials used consist of two different kinds of data: data collected by harvesters and historical forest inventory data. The harvester collected stem data came from stands in central Finland, whereas forest inventory data was obtained from all over Finland. The accuracy of the MSN method was analyzed by estimating characteristics of tree stocks and by comparing simulated spruce, pine and birch log length-diameter distributions with the information from actual stands.

The application presented was found to be a useful and flexible tool for predicting characteristics of marked stands based on the stem data collected by a harvester. The forest inventory data was found less suitable for reference data. The most efficient way to create a length-diameter distribution was to calculate length-diameter class estimates from reference stands as weighted averages of the corresponding length-diameter class. The proposed method appears robust against measurement errors of search variables.

Keywords:

• non-parametric estimation
• planning inventory
• stand characteristics
• stem database
• wood procurement
• Betula
• Picea Abies
• Pinus Sylvestris
• forest inventory
• Finland
• harvester
• log-length-diameter distribution