Combining Airborne Hyperspectral and LiDAR Data Across Local Sites for Upscaling Shrubland Structural Information: Lessons for HyspIRI
Document Type
Article
Publication Date
9-15-2015
Abstract
Fine-scale variation of vegetation structure in dryland systems, such as the Great Basin in the western US, is critical to understanding ecosystem responses to changing land-use conditions. High resolution airborne hyperspectral (HyMap) and LiDAR datasets acquired across independent collection sites can reduce uncertainty in predictive ecosystem modeling and provide a basis for regional upscaling to satellite observations of structural metrics such as cover and height. In the first part of our study, we combined ground reference and airborne data collected at three sagebrush-steppe locations and used the statistical data mining tool random forests to identify remote sensing variables most relevant to estimating shrub cover. In the second part of our study, we hypothesized that vegetation indices derived from hyperspectral satellite observations would not only reliably predict shrub cover but also be relatable to shrub height; thereby augmenting the collection of vertical structure estimates from future satellite platforms such as ICESAT-2. To test this hypothesis, we simulated HyspIRI observations to derive variables to relate to LiDAR-based estimates of shrub cover and height. We generated the same hyperspectral variables as in the first part of this study but at coarser resolution (60 m) and we again used random forests to model shrub cover and height and identify predictors of greatest importance. Overall, combining LiDAR and HyMap datasets at the airborne scale improved shrub cover model results (r2 = 0.58) compared to LiDAR alone (r2 = 0.49). Primary shrub cover variables of importance were HIQR (the interquartile range of height of all LiDAR vegetation returns), HMAD (median absolute deviation from median height of all LiDAR vegetation returns), a narrowband index sensitive to anthocyanins, the ratio of LiDAR vegetation returns to total returns, and a red to green ratio. In addition, HyspIRI-simulated narrowband vegetation indices were relatable to LiDAR-derived shrub cover and height variables (r2 ranging from 0.63 to 0.71) with relatively low root mean square error.
Publication Information
Mitchell, Jessica J.; Shrestha, Rupesh; Spaete, Lucas P.; and Glenn, Nancy F. (2015). "Combining Airborne Hyperspectral and LiDAR Data Across Local Sites for Upscaling Shrubland Structural Information: Lessons for HyspIRI". Remote Sensing of Environment, 167, 98-110. https://doi.org/10.1016/j.rse.2015.04.015