Continuous change detection and classification of land cover using all available Landsat data

Timely and detailed inventories of forest resources are of critical importance to guiding sustainable forest management decisions. As forests occur across large spatial extents, remotely sensed data are often used to augment conventional forest inventory measurements. When combined with field plot measurements, airborne laser scanning (ALS) data can be used to derive detailed enhanced forest inventories (EFIs), which provide spatially explicit and wall-to-wall characterizations of forest attributes. However, these EFIs represent a static point in time, and the dynamic nature of forests, coupled with increasing disturbance and uncertain future conditions, generates a need for the continuous updating of forest inventories. This study used a time series of optical satellite data to update an EFI generated for a large (similar to 690 000 ha) forest management unit in Ontario, Canada, at a two-week interval. The two-phase approach involved first building a relationship between single-year EFI attributes (2018) and spectral variables representing within-year slope, amplitude, and trend of a time series (2000-21) of 14 spectral bands and indices. For each of the 20 strata representing different species groups and site productivity classes, a k-nearest neighbor (kNN) model was developed to impute seven common EFI attributes: aboveground biomass, basal area, stem density, Lorey's height, quadratic mean diameter, and stem volume. Across all strata, models were generally accurate, with relative root mean square error ranging from 11.47% (canopy cover) to 31.82% (stem volume). In the second phase of the approach, models were applied across the entire study area at two-week intervals in order to assess the capacity of the methodology for characterizing change in EFI attributes over a three-year period. Outputs from this second phase demonstrated the potential of the approach for characterizing changes in EFI values in areas experiencing no change or non-stand replacing disturbances. The methods developed herein can be used for EFI update for any temporal interval, thereby enabling more informed decisions by forest managers to prescribe treatments or understand the current state of forest resources.