The National Ecological Observatory Network’s (NEON) terrestrial organism and soil stratified-random sample design allocates samples proportional to land cover type. The National Land Cover Dataset (NLCD) provides consistent and comparable suitable for the network for distributed NEON sites, but observations in the field suggest the continental characterization of land cover is not always accurate at local scales. Misclassification of land cover and the associated allocation of NEON sample plots has the potential to result in erroneous estimates of site level parameters, such as vegetation structure and plant diversity. This study describes the process to create corrected land cover classification maps at a subset of NEON sites. The effort used spectrometer and LiDAR data collected from the NEON airborne observation platform to perform a random forest classification, creating a land cover classification map based on the NLCD ruleset and hierarchy. The original NLCD classification and the classification generated by this work were evaluated by 1) confusion matrices that compare maps to ground-based observations and 2) the capacity of stratified-random samples from each land cover classification to describe site-level mean and variance of tree canopy height as described the LiDAR.
Results/Conclusions:
Initial results from the confusion matrices suggest that the airborne-derived NLCD map better describes conditions found on the map. Results from the comparison of sample designs from each classification map to enable samples that recreate canopy height are pending. This study focuses on the importance of correct land cover proportion and how misclassification can influence uncertainty of mean values of parameter estimates from the aforementioned data products.