Changes in tall shrub abundance on the North Slope of Alaska, 2000–2010

• Published in: Remote sensing of environment Vol. 219; pp. 221 - 232
• Main Authors: Duchesne, Rocio R., Chopping, Mark J., Tape, Ken D., Wang, Zhuosen, Schaaf, Crystal L.B.
• Format: Journal Article
• Language: English
• Published: New York Elsevier Inc 15.12.2018; Elsevier BV
• Subjects: Abundance; Arctic; Bidirectional reflectance; Boosted Regression Tree (BRT); CANAPI; Creeks; Distribution functions; Flood predictions; Floodplains; Fractional cover; Mapping; Multi-angle Imaging SpectroRadiometer (MISR); Reflectance; Regression analysis; Regression models; Remote sensing; Satellite imagery; Satellites; Scattering functions; Scientific imaging; Shrub expansion; Shrubs; Slopes; Spatial resolution; Spectroradiometers; Taiga & tundra; Tall shrubs; Tundra; Vegetation mapping; Weight; Alaska; CANAPI; Fractional cover; Shrub expansion; Arctic; Tall shrubs; Boosted Regression Tree (BRT); Tundra; Multi-angle Imaging SpectroRadiometer (MISR)
• ISSN: 0034-4257; 1879-0704
• DOI: 10.1016/j.rse.2018.10.009

Mapping canopy-forming shrubs in tundra is challenging because approaches that rely on spectral information alone struggle to differentiate low and tall shrubs. We present an evaluation of changes in tall shrub (>0.5 m) fractional cover on the North Slope of Alaska between 2000 and 2010, using a new mapping approach that uses satellite imagery from the NASA Multi-angle Imaging SpectroRadiometer (MISR) sensor, ancillary topographic and geographic variables, derived Ross-Li Bidirectional Reflectance Distribution Function (BRDF) kernel (scattering function) weights, and a Boosted Regression Tree (BRT) model. After training and validating the BRT model, 10 predictor variables explained 53% of the variation in tall shrub fractional cover. The isotropic kernel weight from the BRDF model, topographic slope, nadir camera red reflectance, and latitude were the variables that contributed to the model the most as they were used more frequently to generate the regression trees. At a spatial resolution of 250 m, predicted tall shrub fractional cover on the North Slope ranged from 0.00 to 0.21 (RMSE = 0.03), with the vast majority of sites having fractional cover <0.02. High shrub fractional cover values were predicted along floodplains, creeks, and sloped terrain. Temporal comparisons of estimated tall shrub abundance indicated that shrubs expanded during the period 2000–2010. The extent of the area that experienced a change in tall shrub fractional cover >0.06 was <1% (808 km2) of the total area of the North Slope of Alaska (176,524 km2). Tall shrubs may have expanded throughout a larger area but there is insufficient precision in the MISR-based estimates to make an indisputable determination. Nevertheless, 94% of the locations that exhibited a robust change showed a positive trend toward an increase in shrub cover. The tall shrub fractional cover expansion rate varied between 0.006 yr−1 and 0.015 yr−1, with higher rates in the shrubby southwestern portion of the study area. •Tall shrub cover mapped with machine learning algorithm using MISR imagery•Mapping can only detect indubitable change in <1% of the North Slope over 2000–2010.•Tall shrub cover more than doubled in 97% of the area with an incontrovertible gain.•Shrub expansion is not homogenous and is likely influenced by local conditions.