Improving sample-based National Forest Inventory estimates of tree cover using Landsat-derived land cover data as auxiliary information

In Canada, satellite-derived data are available as National Terrestrial Ecosystem Monitoring System (NTEMS) products, which can be used as auxiliary information to improve sample-based National Forest Inventory (NFI) estimates. This study explored statistical approaches of using these satellite-derived data to improve vegetated tree (VT) cover estimate in the Atlantic Maritime Ecozone. First, a model-assisted regression estimator with beta regression (MA beta ) was evaluated using simulated population datasets. Then, the efficiency of the MA beta estimator was compared to a design-based ratio estimator (DB) in two scenarios: one with temporally matched survey and auxiliary data, and another with temporally unmatched survey and auxiliary data. The assisting model was also used to generate model-based bootstrap aggregate annual estimates (Mb BAE ) of VT cover proportion to explore temporal trends and assess sensitivity to disturbances during 2007–2017. Results showed that the MA beta estimator provided nearly unbiased estimates with a coverage rate of about 95% (when n ≥ 100). The MA beta estimates were more precise than the DB estimates, especially when survey and auxiliary data were temporally matched (RE = 3.04 with g-weights, RE = 3.25 without g-weights). Moreover, the MB BAE -based annual estimates of VT cover proportion indicated that stand-replacing disturbances drive VT cover dynamics in the ecozone. The results suggest that incorporating remote sensing based, independently derived, wall-to-wall data products can improve the accuracy and precision of Canada's NFI, aiding in monitoring forest resources at various scales.