| Automated crown detection and delineation using high spatial resolution imagery has been heavily investigated and proven feasible for extracting information at the individual tree level. However, the accuracy of delineation compared with field-measurements is insufficient to predict the value of tree parameters (e.g., tree crown) from delineated crowns. In this study, two alternatives are investigated to address this issue. One is to investigate indicators for identifying poor delineation from remotely sensed imagery. The other is to examine error-in-variable regression for modeling relationships between error-included variables. The results indicate that the indicators proposed are suitable for identifying poor delineation, which can be important for deriving other parameters from delineated crowns with simple regression. The second component of the results shows that error-in-variable models present advantages over simple regression in estimating regression coefficients and predicting dependent variables when variance of measurement error stays constant in both estimation and prediction data.;Key words: high spatial resolution imagery, crown detection and delineation, identifying poor delineation, error-in-variable regression... |
