Virginia Tech
2 files

Virginia Pine Harvest Activity Classification, 2014-2017

posted on 2022-06-01, 11:25 authored by Valerie ThomasValerie Thomas, Randolph WynneRandolph Wynne, Jobriath Kauffman, Wyatt McCurdy, Evan BrooksEvan Brooks, Robert ThomasRobert Thomas, Jim Rakestraw


This is a 15 m classification of pine forest harvests in Virginia, U.S.A. from 2014-2017 (Thomas et al. 2021). It was generated from 10 predictor variables in a random forest machine learning model. The variables include the constant, sine and cosine from a single-harmonic Fourier fit to the Landsat panchromatic band (all images from 2014-2016), the R^2 and RMSE of this harmonic fit, the RMSE and R^2 of the single-harmonic Fourier fit to Landsat NDVI (all images from 2014-2016), and the Global Forest Change loss year, gain, and tree cover layers (Hansen et al. 2013). The overall classification accuracy is 86%. Class accuracies are 86% for non-harvested pines, 83% for thins, and 90% for clear cuts. The model was applied to areas in Virginia that were classified by the Cropland Data Layer Product (Johnson and Mueller, 2010; Lark et al. 2017) as evergreen (CDL class 142), mixed forest (CDL class 143), or woody wetlands (CDL class 190) for 2013 or 2017. Our results indicate that 12% of pine clear cuts and 17% of pine thins fall within areas classified as deciduous forests by the CDL for this time period.


Regionally specific drivers of land-use transitions and future scenarios: a synthesis considering the land management influence in the southeastern US, NASA LCLUC program, award number NNX17A109G.



University Libraries, Virginia Tech


  • English (US)


Southeastern United States