Premium
Using imaging spectroscopy to detect variation in terrestrial ecosystem productivity across a water‐stressed landscape
Author(s) -
DuBois Sean,
Desai Ankur R.,
Singh Aditya,
Serbin Shawn P.,
Goulden Michael L.,
Baldocchi Dennis D.,
Ma Siyan,
Oechel Walter C.,
Wharton Sonia,
Kruger Eric L.,
Townsend Philip A.
Publication year - 2018
Publication title -
ecological applications
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.864
H-Index - 213
eISSN - 1939-5582
pISSN - 1051-0761
DOI - 10.1002/eap.1733
Subject(s) - eddy covariance , environmental science , hyperspectral imaging , atmospheric sciences , shortwave , primary production , spatial variability , remote sensing , shrubland , carbon cycle , ecosystem , terrestrial ecosystem , flux (metallurgy) , partial least squares regression , ecology , chemistry , radiative transfer , geography , physics , biology , statistics , mathematics , organic chemistry , quantum mechanics
A central challenge to understanding how climate anomalies, such as drought and heatwaves, impact the terrestrial carbon cycle, is quantification and scaling of spatial and temporal variation in ecosystem gross primary productivity ( GPP ). Existing empirical and model‐based satellite broadband spectra‐based products have been shown to miss critical variation in GPP . Here, we evaluate the potential of high spectral resolution (10 nm) shortwave (400–2,500 nm) imagery to better detect spatial and temporal variations in GPP across a range of ecosystems, including forests, grassland‐savannas, wetlands, and shrublands in a water‐stressed region. Estimates of GPP from eddy covariance observations were compared against airborne hyperspectral imagery, collected across California during the 2013–2014 HyspIRI airborne preparatory campaign. Observations from 19 flux towers across 23 flight campaigns (102 total image‐flux tower pairs) showed GPP to be strongly correlated to a suite of spectral wavelengths and band ratios associated with foliar physiology and chemistry. A partial least squares regression ( PLSR ) modeling approach was then used to predict GPP with higher validation accuracy (adjusted R 2 = 0.71) and low bias (0.04) compared to existing broadband approaches (e.g., adjusted R 2 = 0.68 and bias = −5.71 with the Sims et al. [Sims, D. A., 2008] model). Significant wavelengths contributing to the PLSR include those previously shown to coincide with Rubisco (wavelengths 1,680, 1,740, and 2,290 nm) and V cmax (wavelengths 1,680, 1,722, 1,732, 1,760, and 2,300 nm). These results provide strong evidence that advances in satellite spectral resolution offer significant promise for improved satellite‐based monitoring of GPP variability across a diverse range of terrestrial ecosystems.