Improving the precision of estimates of forest carbon emissions and removals using national forest inventory data and dense time series of Landsat imagery.

This study examines the utility of dense time series of Landsat imagery for improving the precision of estimates of change in forest stocks. Monthly composites of Landsat 7 ETM+ imagery from a 2.25 million hectare study area in the state of Wisconsin, USA for the decade of 2003-2012 were transformed to brightness, greenness, and wetness values using the Tasseled Cap (TC) transformation. Harmonic regression was used to fit a Fourier series to each set of TC component values for each pixel for each of two 5-year periods: 2003-2007 and 2008-2012. These estimated Fourier coefficients were used in conjunction with 1,446 re-measured national forest inventory (NFI) plot data from the Forest Inventory and Analysis program over the same decade as the imagery to estimate changes in live tree basal area via the k-nearest neighbor (kNN) estimator. The model-assisted regression estimator was used to incorporate the kNN estimates with the NFI plot information to improve the precision of estimates based on the plots alone. The results indicated a relative efficiency of 17%, suggesting that the sample size would have to be increased by 17% in order to achieve a comparable precision.