UC Santa Barbara
Remote sensing of plant species using airborne hyperspectral visible-shortwave infrared and thermal infrared imagery
- Author(s): Meerdink, Susan Kay
- Advisor(s): Roberts, Dar A
- et al.
In California, natural vegetation is experiencing an increasing amount of stress due to prolonged droughts, wildfires, insect infestation, and disease. Remote sensing technologies provide a means for monitoring plant species presence and function temporally across landscapes. In this his dissertation, I used hyperspectral visible shortwave infrared (VSWIR), hyperspectral thermal (TIR), and hyperspectral VSWIR + broadband TIR imagery to derive key observations of plant species across a gradient of environmental conditions and time frames. In Chapter 2, I classified plant species using hyperspectral VSWIR imagery from 2013 - 2015 spring, summer, and fall. Plant species maps had the highest classification accuracy using spectra from a single date (mean kappa 0.80 – 0.86). The inclusion of spectra from other dates decreased accuracy (mean kappa 0.78 - 0.83). Leave-one-out analysis emphasized the need to have spectra from the image date in the classification training, otherwise classification accuracy dropped significantly (mean kappa 0.31 – 0.73). In Chapter 3, I used hyperspectral TIR imagery to determine the extent that high precision spectral emissivity and canopy temperature can be exploited for vegetation research at the canopy level. I found that plant species show distinct spectral separation at the leaf level, but separability among species is lost at the canopy level. However, species’ canopy temperatures exhibited different distributions among dates and species. Variability in canopy temperatures was largely explained by LiDAR derived canopy structural attributes (e.g. canopy density) and the surrounding environment (e.g. presence of pavement). In Chapter 4, I used combined hyperspectral VSWIR and broadband TIR imagery to monitor plant stress during California’s 2013 – 2015 severe drought. The temperature condition index (TCI) was calculated to measure plant stress by using plant species’ surface minus air temperature distributions across dates. Plant stress was not evenly distributed across the landscape or time with lower elevation open shrub/meadows, showing the largest amount of stress in June 2014, and August 2015 imagery. Plant stress spatial variability across the study area was related to a slope’s aspect with highly stressed plants located on south or south-southwest facing slopes. Overall, this dissertation quantifies the ability to temporally study plant species using hyperspectral VSWIR, hyperspectral TIR, and combined VSWIR+TIR imagery. This analysis supports a range of current and planned missions including Surface Biology and Geology (SBG), Environmental Mapping and Analysis Program (EnMAP), National Ecological Observatory Network (NEON), Hyperspectral Thermal Emission Spectrometer (HyTES), and ECOsystem Spaceborne Thermal Radiometer Experiment on Space Station (ECOSTRESS).