Estimation of chromophoric dissolved organic matter and non-algal particulate absorption coefficients of seawater in the ultraviolet by extrapolation from the visible spectral region

Extending the capabilities of optical remote sensing and inverse optical algorithms, which have been commonly focused on the visible (VIS) range of the electromagnetic spectrum, to derive the optical properties of seawater in the ultraviolet (UV) range is important to advancing the understanding of various optical, biological, and photochemical processes in the ocean. In particular, existing remote-sensing reflectance models that derive the total spectral absorption coefficient of seawater, a(lambda), and absorption partitioning models that partition a(lambda) into the component absorption coefficients of phytoplankton, a(ph)(lambda), non-algal (depigmented) particles, a(d)(lambda), and chromophoric dissolved organic matter (CDOM), a(g)(lambda), are restricted to the VIS range. We assembled a quality-controlled development dataset of hyperspectral measurements of a(g)(lambda) (N = 1294) and a(d)(lambda) (N = 409) spanning a wide range of values across various ocean basins, and evaluated several extrapolation methods to extend a(g)(lambda), a(d)(lambda), and a(dg)(lambda) = a(g)(lambda) + a(d)(lambda) into the near-UV spectral region by examining different sections of the VIS as a basis for extrapolation, different extrapolation functions, and different spectral sampling intervals of input data in the VIS. Our analysis determined the optimal method to estimate a(g)(lambda) and a(dg)(lambda) at near-UV wavelengths (350 to 400 nm) which relies on an exponential extrapolation of data from the 400-450 nm range. The initial a(d)(lambda) is obtained as a difference between the extrapolated estimates of a(dg)(lambda) and a(g)(lambda). Additional correction functions based on the analysis of differences between the extrapolated and measured values in the near-UV were defined to obtain improved final estimates of a(g)(lambda) and a(d)(lambda) and then the final estimates of a(dg)(lambda) as a sum of final a(g)(lambda) and a(d)(lambda). The extrapolation model provides very good agreement between the extrapolated and measured data in the near-UV when the input data in the blue spectral region are available at 1 or 5 nm spectral sampling intervals. There is negligible bias between the modeled and measured values of all three absorption coefficients and the median absolute percent difference (MdAPD) is small, e.g., < 5.2% for a(g)(lambda) and < 10.5% for a(d)(lambda) at all near-UV wavelengths when evaluated with the development dataset. Assessment of the model on an independent dataset of concurrent a(g)(lambda) and a(d)(lambda) measurements (N = 149) yielded similar findings with only slight reduction of performance and MdAPD remaining below 6.7% for a(g)(lambda) and 11% for a(d)(lambda). These results are promising for integration of the extrapolation method with absorption partitioning models operating in the VIS.