Clark, Matthew (Matt): Sonoma State University (Project Lead)
Project Funding:
2012 - 2017
NRA: 2011 NASA: HyspIRI Preparatory Airborne Activities and Associated Science
Funded by NASA, Other US Funding:
Abstract:
Land cover is an essential global climate variable and important for understanding coupled natural and socio-economic processes at work on a landscape. The science, policy, management, conservation, and development communities need recent land-cover information, with sufficient class detail, accuracy and wall-to-wall spatial coverage to analyze the multiple drivers of change that operate at different spatial and temporal scales. However, there is a surprising lack of land-cover data that meet these requirements at regional to global scales, partially due to our reliance on low- to medium-resolution multispectral satellites for mapping.
With its high spectral resolution, moderate pixel size, and multiple acquisitions within a year, the planned HyspIRI satellite will offer an unprecedented image stream at global scales. Studies with field spectrometers and imaging spectrometers have demonstrated how vegetation chemical-structural effects on hyperspectral properties can be used to discriminate land cover at leaf, crown and broader scales of floristic organization. The seasonal chemical-structural changes in vegetation types due to phenology, such as leaf flush, expansion and senescence, can have profound effects on spectral variation but have not been deeply explored in hyperspectral studies. In our proposed research, we ask two fundamental science-based questions: 1) Can simulated HyspIRI imagery and associated hyperspectral processing techniques produce better maps of natural vegetation than possible with imagery from a traditional, multispectral sensor? 2) How does seasonal spectral variation in natural vegetation types relate to underlying abiotic and phenological factors, and can this variation be harnessed to aid vegetation discrimination?
This project is based in California, a global biodiversity hotspot. It uses simulated HyspIRI and Landsat imagery derived from AVIRIS images covering three broad transects, with three seasons of imagery per year, over two years. We investigate spectral-based discrimination of natural vegetation at two levels of floristic organization, following the National Vegetation Classification scheme: formations at a broad scale covering all AVIRIS collections and finer-scale alliances in two core research sites. Reference data are collected by photo-interpretation of formation-level samples within an automated web-based tool and by sampling of alliances in the field. We use simulated HyspIRI imagery to generate a suite of hyperspectral metrics related to vegetation chemical absorptions, structure and physiology from the visible to shortwave infrared parts of the spectrum. Techniques include narrowband indices, absorption-feature fitting, derivative analysis and multiple-endmember spectral mixture analysis MESMA. The Random Forests and MESMA classifiers are compared for mapping natural vegetation using seasonal and between-year HyspIRI data. We also explore spectral-temporal variation within and among natural vegetation classes and its relationship to underlying environmental and phenological factors.
Our project provides a quantitative assessment of the potential of HyspIRI for mapping natural vegetation at levels of floristic organization needed by regional- to global-scale applications. As a cost-benefit analysis, we compare HyspIRI and Landsat map class and spatial detail, accuracy and temporal stability. Our proposed activities, questions and results will develop the foundation upon which we can ask broader HyspIRI Mission science questions, such as those related to the pattern, spatial distribution and seasonal activity of global biomes and their ecosystems. The project also has implications for assessing how spaceborne hyperspectral sensors can improve land-cover data needed by land-change science and national and international policies related to climate change.
Publications:
Clark, M. L. 2017. Comparison of simulated hyperspectral HyspIRI and multispectral Landsat 8 and Sentinel-2 imagery for multi-seasonal, regional land-cover mapping. Remote Sensing of Environment. 200, 311-325. DOI: 10.1016/j.rse.2017.08.028
Clark, M. L. 2020. Comparison of multi-seasonal Landsat 8, Sentinel-2 and hyperspectral images for mapping forest alliances in Northern California. ISPRS Journal of Photogrammetry and Remote Sensing. 159, 26-40. DOI: 10.1016/j.isprsjprs.2019.11.007
Clark, M. L., Buck-Diaz, J., Evens, J. 2018. Mapping of forest alliances with simulated multi-seasonal hyperspectral satellite imagery. Remote Sensing of Environment. 210, 490-507. DOI: 10.1016/j.rse.2018.03.021
Foukal, N. P., Thomas, A. C. 2014. Biogeography and phenology of satellite-measured phytoplankton seasonality in the California current. Deep Sea Research Part I: Oceanographic Research Papers. 92, 11-25. DOI: 10.1016/j.dsr.2014.06.008
Guidici, D., Clark, M. 2017. One-Dimensional Convolutional Neural Network Land-Cover Classification of Multi-Seasonal Hyperspectral Imagery in the San Francisco Bay Area, California. Remote Sensing. 9(6), 629. DOI: 10.3390/rs9060629
Clark, M. L., Kilham, N. E. 2016. Mapping of land cover in northern California with simulated hyperspectral satellite imagery. ISPRS Journal of Photogrammetry and Remote Sensing. 119, 228-245. DOI: 10.1016/j.isprsjprs.2016.06.007
More details may be found in the following project profile(s):
