Canadian Forest Service Publications
Area-level analysis of forest inventory variables. 2017. Magnussen, S.; Mauro, F.; Breidenbach, J.; Lanz, A.; Kändler, G. Eur J Forest Res 136:839–855.
Issued by: Pacific Forestry Centre
Catalog ID: 38948
Availability: PDF (request by e-mail)
Available from the Journal's Web site. †
† This site may require a fee
Small-area estimation is a subject area of growing importance in forest inventories. Modelling the link between a study variable Y and auxiliary variables X—in pursuit of an improved accuracy in estimators—is typically done at the level of a sampling unit. However, for various reasons, it may only be possible to formulate a linking model at the level of an area of interest (AOI). Area-level models and their potential have rarely been explored in forestry. This study demonstrates, with data (Y = stem volume per ha) from four actual inventories aided by aerial laser scanner data (3 cases) or photogrammetric point clouds (1 case), application of three distinct models representing the currency of area-level modelling. The studied AOIs varied in size from forest management units to forest districts, and municipalities. The variance explained by X declined sharply with the average size of an AOI. In comparison with a direct estimate mean of Y in an AOI, all three models achieved practically important reduction in the relative root-mean-squared error of an AOI mean. In terms of the reduction in mean-squared errors, a model with a spatial location effect was overall most attractive. We recommend the pursuit of a spatial model component in area-level modelling as promising within the context of a forest inventory.
Plain Language Summary
Forest inventories have successfully exploited remotely sensed data - in particular from airborne laser scanners – to improve precision of key forest inventory attributes in areas of interest (e.g. a forest stand). The success has been with unit level models where field plot data is linked directly to the remotely sensed data. However, this direct link is not always possible. As an alternative this study demonstrates with actual data from four inventories an area-level analysis whereby summaries of field data for an area of interest is linked to summaries from remotely sensed data. A model that includes spatial location effect in one or more of the remotely sensed auxiliary variables was successful and lowered the uncertainty in an area specific estimate by approximately 50%. We believe the method will have wide application in practice to improve precision of estimates from, for example, cut-blocks where an accurate geo-location of field data collected with variable radius plots prevents applications of models with a direct link between field data and the remotely sensed data.