Improving the precision of dynamic forest parameter estimates using Landsat

•We present a method for improving precision of dynamic forest parameter estimates.•We used a Landsat image stack covering the entire Thematic Mapper history.•NDVI trajectories were built with harmonic regression due to aperiodic image data.•Three methods were tested: post-minimum NDVI, recent NDVI, and recent squared NDVI.•The methods gave estimated relative efficiencies of up to 1.2 vs. an SRS benchmark.

The use of satellite-derived classification maps to improve post-stratified forest parameter estimates is well established. When reducing the variance of post-stratification estimates for forest change parameters such as forest growth, it is logical to use a change-related strata map. At the stand level, a time series of Landsat images is ideally suited for producing such a map. In this study, we generate strata maps based on trajectories of Landsat Thematic Mapper-based normalized difference vegetation index values, with a focus on post-disturbance recovery and recent measurements. These trajectories, from 1985 to 2010, are converted to harmonic regression coefficient estimates and classified according to a hierarchical clustering algorithm from a training sample. The resulting strata maps are then used in conjunction with measured plots to estimate forest status and change parameters in an Alabama, USA study area. These estimates and the variance of the estimates are then used to calculate the estimated relative efficiencies of the post-stratified estimates. Estimated relative efficiencies around or above 1.2 were observed for total growth, total mortality, and total removals, with different strata maps being more effective for each. Possible avenues for improvement of the approach include the following: (1) enlarging the study area and (2) using the Landsat images closest to the time of measurement for each plot. Multitemporal satellite-derived strata maps show promise for improving the precision of change parameter estimates.