Uncertainty of Ocean Color Algorithms: A Monte Carlo Approach

The large global datasets used for the development of algorithms to estimate ocean color-derived biogeochemical parameters from space-borne sensors, such as chlorophyll-a (C_a), assemble observations from a variety of in situ optical sensors and analytical laboratory protocols, contributed over a period of years by dozens of investigators worldwide. Each algorithm-derived magnitude ideally must be ascribed a measure of uncertainty. In practice, that objective remains challenging. The establishment of uncertainty budgets for derived parameters through classical error propagation approaches is rendered impractical, if not intractable, by the complexity and heterogeneity within algorithm development datasets. An alternative approach is to use Monte Carlo statistical methods to mine the observation variability embedded in calibration datasets to establish measures of uncertainty. We present results from Monte Carlo estimation of uncertainty for C_a algorithms developed with the NASA bio-Optical Marine Algorithm Data set (NOMAD). NOMAD is a publicly available, global, high quality in situ bio-optical data set for use in ocean color algorithm development and satellite product validation activities. The dataset includes coincident observations of water-leaving radiances, surface irradiances, downwelling attenuation coefficients, and C_a concentrations contributed by the ocean color research community. Our results show how wavelength selection affects algorithm uncertainty and suggest ways to improve collection and reporting of in situ data for algorithm development. Results from this effort could become part of a standard pixel-by-pixel uncertainty assessment of future ocean color products.