Accuracy of partially visually assessed stand characteristics: a case study of Finnish forest inventory by compartments

In Finland, forest management planning is based on standwise assessment of forest variables. The data is traditionally gathered partly by (subjective) sampling and partly by visual assessment, which makes the accuracy assessments difficult. This study consists of an experiment where the visual assessments of field technicians were compared with the accurately measured values. The data consists of assessments from 18 sample plots made by 19 technicians. Each technician assessed four forest characteristics from each stand, for each tree species and each tree class. Basal area was observed in all cases; the other three variables varied according to 18 different measurement strategies. From these observations, mixed models were estimated to analyze to what extent the assessment errors depend on forest characteristics. Variation among both sample plots and field technicians was also considered. The results show that some of the variables could be interpreted as Berkson cases. The assessment errors were also often highly hetero sce dastic. Therefore, variance was explicitly modeled, and the final error models were estimated with weighted mixed regression using the variance estimates as weights. The results show clear variation among technicians, especially in characteristics that include personal judgment. The effect of training could be detected from variation between the technician groups. Furthermore, the broad-leaved tree classes were generally more difficult to assess than conifers.