Linking vegetation cover and seasonal thaw depths in interior Alaska permafrost terrains using remote sensing

REMOTE SENSING OF ENVIRONMENT

卷:233

Article

Anderson, John E.¹  Douglas, Thomas A.²  Barbato, Robyn A.³  Saari, Stephanie²  Edwards, Jarrod D.¹  Jones, Robert M.³ 

[1] US Army, Gerospatial Res Lab, Corbin Field Stn, 15315 Magnet Lane, Woodford, VA 22580 USA

[2] US Army, Cold Reg Res & Engn Lab, POB 35170, Ft Wainwright, AK 99703 USA

[3] US Army, Cold Reg Res & Engn Lab, 72 Lyme Rd, Hanover, NH 03755 USA

关键词: Hyperspectral imagery;    Spectral reflectance;    Permafrost;    Vegetation;    Seasonal thaw;    Active layer;    Classification;   

DOI  :  10.1016/j.rse.2019.111363

来源: Elsevier

【 摘 要 】

Permafrost in Interior Alaska is protected against summer thaw by an insulating layer of moss and mixed vegetative cover that regulates seasonal thaw and the end-of-summer season permafrost active layer depth. Thaw depths are laborious point scale measurements that can be difficult to translate regionally. Since disturbances are present on the landscape across many temporal and spatial scales they can greatly affect the soil and vegetation regime. Furthermore, many areas are denied full assessment due to terrain complexity or limited accessibility. As such, a remotely sensed means for estimating surface thaw based on insular vegetation composition would be advantageous. A synoptic evaluation of this insulating layer could eventually benefit regional mapping of areas where vegetation cover helps regulate thaw depths during the local growing season. Herein, we present multi-year data collected from three terrain types in Interior Alaska that relates seasonal thaw depth to vegetative cover type. Field samples for spectral reflectance, vegetation, soil, elevation and seasonal thaw depths were obtained from surveyed 1 m(2) quadrats during the local growing season (late July, each summer from 2014 to 2017) across three lowland boreal landscapes. Statistical relationships between vegetation and samples were explored using CCA and showed vegetative cover distributions were highly correlated in two dimensions with the principal variables represented almost evenly by the soil variable pH and thaw depth. Class maps representing vegetation and associated thaw depth were derived from hyperspectral imagery using field and imagery training data. Map accuracy assessment, conducted using random points to establish truth data, yielded overall accuracies of > 85%. Regression analysis and root mean square error testing of the predictive capacity of the vegetation classes and thaw depth was variable but encouraging, ranging from 5 to 11 cm or between an 8 and 37% chance of error. We feel the results are strong enough to stimulate more study in the evaluation of vegetation and thaw depth mapping during the local growing season.

【 授权许可】

Free