Spectral-temporal features over South Africa (2013-2015) derived from multi-sensor Landsat imagery

The three data sets are spectral-temporal features derived from multi-sensor Landsat imagery over South Africa. They were derived as follows:

All Landsat-5, -7 and -8 data from 2013 to 2015 covering South Africa (70 tiles) were acquired and pre-processed following the procedure implemented in Hansen et al. (2013), Potapov et al. (2014), Potapov et al. 2012. Note that four tiles (path/row: 176/77, 175/78, 174/78, 174/79) were discarded because no crop is grown there. Four spectral bands were kept: the red, the near-infrared (NIR), and the two short-wave infrared (SWIR) bands. The blue and green bands were discarded due to their sensitivity to atmospheric effects. We applied a three-step procedure to normalize the radiometry. First, Landsat data were converted to top-of-atmosphere reflectance and then normalized by taking the corresponding MODIS top-of-canopy reflectance data as target. Third, we adjusted cross-track surface anisotropy effects by modeling the Landsat reflectance per spectral band as a function of the viewing angle. The above-mentioned processing steps incrementally improved the appearance of the data, providing more spatial coherence and increasing the generalization and internal consistency of the multi-spectral feature space.

Hansen MC, Potapov PV, Moore R, Hancher M, Turubanova S, Tyukavina A, et al. High-resolution global maps of 21st-century forest cover change. science. 2013;342(6160):850{853.

Potapov P, Dempewolf J, Talero Y, Hansen M, Stehman S, Vargas C, et al. National satellite-based humid tropical forest change assessment in Peru in support of REDD+ implementation. Environmental Research Letters. 2014;9(12):124012.

Potapov PV, Turubanova S, Tyukavina A, Krylov A, McCarty J, Radeloff V, et al. Eastern Europe's forest cover dynamics from 1985 to 2012 quantified from the full Landsat archive. Remote Sensing of Environment. 2015;159:28 43.