Multiscale Validation of an Operational Model of ForestInventory Attributes Developed with ConstrainedRemote Sensing Data

摘要

This paper summarizes the output of an imputation model that simultaneously estimates multiple operational-scaleforest inventory attributes in the Laurentian mixed forest type of the United States. The model was constrained by nationalforest inventory privacy protocols and temporal uncertainties in feature and reference data. Estimates were most accurate at thecounty level and more variable across smaller spatial extents. Model development and validation highlighted that performanceand reliability were influenced by our approach of using publicly available remote sensing predictors and ground reference datain model building. Comprehensive validation included diagnostics of the chosen model and leveraged multiscale independent datafor analysis of lack-of-fit spatially and by individual feature variables. Relatively poor performance in some forest types pointedto an impact of temporal mismatch in the estimation of forest stocking in typically even-aged stands dominated by fast-growingspecies.