Integrated object-based spatiotemporal characterization of forest change from an annual time series of Landsat image composites

Identifying and mapping the location, extent, severity, and causal agent of forest change events is necessary for a wide range of information needs. Using six ecologically representative test sites (each ~800 km2) in the province of Saskatchewan, Canada, for the period 1998–2012, our objective was to prototype an object-based approach for identifying spectral change features, filling data gaps in Landsat reflectance annual Best-Available-Pixel (BAP) composites, and attributing change processes to the derived stand-like spatial objects. The Tasseled Cap Angle (TCA), which combines information from the visible, near-infrared, and mid-infrared, enabled the description of landscape condition, and its temporal derivative, the Process Indicator (PI), relates the rate and directionality of change at the landscape level. Data gaps, as well as anomalous values identified using time-series similarity analysis with Dynamic Time Warping and cross-correlation measures, were replaced by spatiotemporal interpolation, resulting in annual proxy composites with no missing values. An assessment of proxy values against surface reflectance values indicated high agreement for reflectance bands (R = 0.79–0.96, RMSE = 0.005–0.021) and for TCA (R = 0.93, RMSE = 0.005), and a decrease in reliability of the proxy value as the size of the spatiotemporal gap increased, with longer temporal gaps having a greater impact on infill reliability than larger spatial gaps. Distinctive change dynamics of the sample sites were captured, demonstrating a capacity to simultaneously identify low and high magnitude changes as well as positive (e.g., growth) and negative (e.g., wildfire) trajectories using the PI. The approach presented herein provides a robust option for monitoring forest change by simultaneously describing state and ongoing change processes.