The global extent and multi-decadal timespan of Landsat satellite data presents an unprecedented data source for understanding ecological dynamics. Tropical forest succession represents an ecological dynamic that is highly variable across landscapes and regions. Nevertheless, forecasting rates of second-growth forest succession is crucial for restoration planning, including to determine which areas will recover under natural regeneration. The Landsat satellite record represents a potential dataset that can inform historic rates of forest recovery, however, distinguishing between biological variability and measurement error related to remote sensing challenges interpretation of Landsat-derived successional trajectories. We present a Bayesian state-space modeling framework for disentangling biological process from measurement error to model canopy height dynamics in second-growth forests. Our approach enables model-based estimates of canopy height from Landsat imagery via fusion with aerial lidar. We demonstrate our framework in Southwestern Panama, a heterogeneous landscape undergoing variable rates of secondary succession. We found that data fusion using our state space model improved accuracy of hindcasts of forest succession and decreased uncertainty in which sites were undergoing fast vs. slow succession. As a multitude of remote sensing data sources come online, each with their own strengths and limitations, data fusion represents a powerful tool to combine information for inference on ecological dynamics. We conclude with a discussion of how our approach could be upscaled to model forest recovery trajectories across large spatial extents.
tropical dry forest, lidar, natural regeneration, Azuero Peninsula