Effects of Choosing Different Parameterization Data in Two-Phase Forest Inventories for Standing Stock Estimation

Basel

e-journal

17508

200212680

The demands on national forest inventories to provide detailed information for small geographical regions are rising. Two-phase estimators are often employed to obtain forest resource estimates, yet there is little information on optimal training data selection. This study evaluates the impact of different training data on two-phase estimators, with a focus on small area estimators for standing stock and aims to develop guidelines on selecting appropriate training datasets. Linear regression models were parameterized using multiple datasets and subsets based on ecological and administrative boundaries. The models were then applied on varying scales, and their estimates and their confidence intervals were compared to each other as well as to the single-phase, purely terrestrial forest inventory. Results suggest that the different two-phase models generally yield comparable estimates but differ notably from single-phase estimates. Specifically, differences increase in smaller areas and with correspondingly smaller training datasets, suggesting a minimum of 100 data points. To ensure robust estimates, we recommend adapting training sets to local conditions and exercising caution with small training datasets and areas because implausible results may occur. Pooling appropriate datasets is the preferable solution.Keywords: national forest inventory; model parameterization; uncertainty; small area estimation; forest stock volume