Abstract
The increasing use of hyperspectral optical data in oceanography, both in situ and via remote sensing, holds the potential to significantly advance characterization of marine ecology and biogeochemistry because, in principle, hyperspectral data can provide much more detailed inferences of ecosystem properties via inversion. Effective inferences, however, require careful consideration of the close similarity of different signals of interest, and how these interplay with measurement error and uncertainty to reduce the degrees of freedom (DoF) of hyperspectral measurements. Here we discuss complementary approaches to quantify the DoF in hyperspectral measurements in the case of in situ particulate absorption measurements, though these approaches can also be used on other such data, e.g., ocean color remote sensing. Analyses suggest intermediate (${\sim}5 $∼5) DoF for our dataset of global hyperspectral particulate absorption spectra from the Tara Oceans expedition, meaning that these data can yield coarse community structure information. Empirically, chlorophyll is an effective first-order predictor of absorption spectra, meaning that error characteristics and the mathematics of inversion need to be carefully considered for hyperspectral data to provide information beyond that which chlorophyll provides. We also discuss other useful analytical tools that can be applied to this problem and place our results in the context of hyperspectral remote sensing.
Highlights
In many instances, in both science and life, light provides a wealth of information about our environment
Derivative Analysis Another commonly used technique, and one that is often cited as motivation for acquiring hyperspectral data, is derivative analysis, i.e., taking spectral derivatives of signals and analyzing/comparing these derivatives rather than the original signals (e.g., [53,54])
Smoothing filters are typically applied before taking derivatives, which removes any high-frequency variability due likely to uncorrelated noise
Summary
In both science and life, light provides a wealth of information about our environment. In a hypothetical limit case where all of the wavebands were perfectly correlated among themselves such that there was no variation in spectral shape in the whole ocean (but only of intensity), the DoF of any measurement would be one, regardless of the spectral resolution Using such data to infer more than one variable would be fraught, and would not be able to provide independent estimates of the quantities being inverted for. We employ two simple and complementary analyses—information content analysis (ICA) and principal component analysis (PCA)—to address this question as applied to hyperspectral particulate absorption (ap(λ)) There is both a great deal of global ap(λ) data available, as well as a substantive body of work decomposing the ap(λ) spectra into that of non-algal particles (NAP; aNAP(λ)) [22] and absorption by phytoplankton pigments (aφ(λ)) and further from aφ(λ) to different sizes of plankton or to different pigments [23,24,25]. We present our approaches and findings in sections defined by analysis type; Section 2 addresses the ICA, Section 3 addresses the PCA, and Section 4 reports on other types of analyses and conclusions
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
