Estimating downscaled eMODIS NDVI using Landsat 8 in the central Great Basin shrub steppe

The study's goal was to downscale 2013 250-m expedited Moderate Resolution Imaging Spectroradiometer (eMODIS) Normalized Difference Vegetation Index (NDVI) to 30 m (Gu, Y. and Wylie, B.K., 2015, Developing a 30-m grassland productivity estimation map for central Nebraska using 250-m MODIS and 30-m Landsat-8 observations, Remote Sensing of Environment, v. 171, p. 291-298)using 2013 Landsat 8 data. The eMODIS NDVI was downscaled for four periods: mid spring, early summer, late summer and mid fall. The objective was to capture phenologies during periods that correspond to 1) annual grass growth, 2) annual grass senescence, 3) the optimal NDVI profile separation between sagebrush and other shrubs in the region, and 4) sagebrush leaf off. The study area is defined as the central Great Basin in the western United States. Two tiles (approximating Days 175 and 250 in 2013) of best pixel data from bands 2 through 7 of Landsat 8 were acquired (Nelson, K.J., and Steinwand, D., 2015, A Landsat Data Tiling and Compositing Approach Optimized for Change Detection in the Conterminous United States: Photogrammetric Engineering & Remote Sensing, v. 81, no. 7, p. 573-586.), spatially averaged to 250 m, and used as independent variables to develop regression-tree models that estimated eMODIS NDVI weeks at 250 m. Training points for the mid-spring time period were limited to elevations at or below 7,000 feet (2133 m). Training points for all other time periods were not limited by an elevation threshold. Points above 7, 000 feet made up approximately 2.5% of the training and test datasets. Otherwise, the models were trained on pixels where the 2011 National Land Cover Database classified the pixel as ≥ 70% shrub/scrub, herbaceous, or bare ground; these are likely rangelands. Also, if both Landsat images had no data in the same pixel, that pixel was excluded as a training point. Regression-tree models were built with weekly 250-m eMODIS NDVI and spatially averaged 250-m Landsat data. The model rules and algorithms, along with the original 30-m Landsat data, were applied via a mapping application to render 30-m mapped eMODIS NDVI for the four time periods defined above.