LandTrendr PySTEM Impervious Surface CONUS (1990-2017)

This annual landcover dataset contains model-predicted impervious surface cover (ISC) across the Contiguous United States from 1990-2017. The model is the product of locally trained estimators. A common set of spectral predictors is used across the modeling region, with classification tuned at the'neighborhood' scale, allowing the PySTEM ensemble model to capture spatially variable predictor-response relationships. LandTrendr supplied annual, temporally segmented spectral data that allowed per-pixel landscape history to be incorporated into classification decisions.

The ensemble model approach of PySTEM benefits from the localized prediction scheme of regionally discrete models, with the added value of expressing gradients of model prediction change across the landscape. The model’s ability to auto-recalibrate its estimators for a given landcover type based on regional characteristics allows for efficient multi-biome model fits.

National Land Cover Database (NLCD) landcover maps from 2001 were used as training data. Three different products from the NLCD were modeled, each separately: landcover, impervious surface cover, and forest canopy cover.

LandTrendr supplied annual data in two indices: fitted tasseled-cap brightness, greenness, and wetness; and normalized burn ratio (NBR). These indices were supplemented with temporal segmentation data, including year-on-year change in index values and the number of years since the last prior disturbance. Mean squared error for each temporal segment is also included at the pixel scale, allowing for analysis of the spatial variability of model performance.

For additional information, please see:https://www.sciencedirect.com/science/article/pii/S0034425718301330

Please see the implementation of the LandTrendr algorithm on Google Earth Engine at:https://emapr.github.io/LT-GEE/

Citation:Hooper, S., & Kennedy, R. E. (2018). A spatial ensemble approach for broad-area mapping of land surface properties. Remote Sensing of Environment, 210, 473-489.