Identifying the Presence of Assessment Errors in Forest Inventory Data by Data Mining

Abstract

Abstract All forest inventory methods are susceptible to assessment errors, and although the majority of these errors are relatively minor, some can be exceptionally large. Errors reduce data reliability and increase the probability of nonoptimal decisions in forest planning. We propose that outlier detection techniques based on data mining could be used to detect some of the assessment errors in forestry databases. We tested four outlier detection algorithms presented in previous data mining studies for detecting the errors in compartment-wise field inventory data used in forest planning and examined the relations between the outliers and assessment errors. There was a clear relation between outliers and assessment errors, but this varied somewhat among the algorithms. Compartments with large assessment errors had a higher probability of being classified as outliers. The findings suggest that outlier detection techniques based on data mining could provide a cost-efficient tool for detecting some of the largest assessment errors in inventory data and thus improve the reliability of the whole forest planning process.