From Validation Statistics to Uncertainty Estimates: Application to VIIRS Ocean Color Radiometric Products at European Coastal Locations

Uncertainty estimates are needed to assess ocean color products and qualify the agreement between missions. Comparison between field observations and satellite data, a process defined as validation, has been the traditional way to assess satellite products. However validation statistics can provide only an approximation for satellite data uncertainties as field measurements have their own uncertainties and as the validation process is imperfect, comparing data potentially differing in temporal, spatial or spectral characteristics. This study describes a method to interpret in terms of uncertainties the validation statistics obtained for ocean color remote sensing reflectance R-RS knowing the uncertainties associated with field data. This approach is applied to observations collected at sites part of the Ocean Color component of the Aerosol Robotic Network (AERONET-OC) located in coastal regions of the European seas, and to R-RS data from the VIIRS sensors on-board the SNPP and JPSS1 platforms. Similar estimates of uncertainties sigma(VRS) (term accounting for non-systematic contributions to the uncertainty budget) are obtained for both missions, decreasing with wavelength from the interval 0.8-1.4 10(-3) sr(-1) in the blue to a maximum of 0.24 10(-3) sr(-1) in the red, values that are at least twice (but up to 8 times) the uncertainties reported for the field data. These uncertainty estimates are then used to qualify the agreement between the VIIRS products, defining the extent to which they agree within their stated uncertainty. Despite significant biases between the two missions, their R-RS products appear fairly compatible.

Automated summary

Uncertainty estimates are crucial for assessing ocean color products and confirming the consistency between satellite missions. Despite significant biases between the two missions, their R-RS products appear to be fairly compatible based on the uncertainties associated with field data and validation statistics.