Effects of positional errors in model-assisted and model-based estimation of growing stock volume

Positional errors may cause problems when field and remotely sensed data are combined in connection with forest surveys. In this study we evaluated the effects of such errors on statistical estimates of growing stock volume using model-assisted and model-based estimation. With model-assisted estimation, positional errors affect the model parameter estimates for the models that are used as part of the estimation framework. In addition, positional errors affect the estimators, since the deviations between model predictions and field measurements are often larger than they would have been without positional errors. Using model-based estimation positional errors affect the model parameter estimates and thus the estimators. We compared the effects of positional errors in model-assisted and model-based estimation through Monte Carlo sampling simulation in a simulated study area resembling the forest conditions in Kuortane, western Finland. The forest population was created using a copula modelling approach based on field, Landsat and LiDAR data. We found that positional errors led to slightly biased estimators, and estimators with larger variances compared to the cases where data were perfectly geo-located. The relative increase of the variances of the estimators was of equal magnitude for model-assisted and model-based estimation, when models were developed and applied to data with geopositional errors. Further, the variance estimators were always more precise for the model-based estimators compared to the model-assisted estimators. When the models were developed based on perfectly geo-located data but applied to data with positional errors, model-based estimation was superior to model-assisted estimation.