Atmospheric correction of the strongly absorbing aerosol dataset includes three vertical distribution parameters, namely mean height (in km), height standard deviation (in km), and peak particle number concentration (in cm-3). This spatial resolution of dataset is 0.75° × 0.75° and covers the global region from January 2007 to December 2016. In addition, we established a global aerosol classification algorithm based on AERONET's Level-2 aerosol measurement data over coastal and open ocean globally. We constructed strong absorption aerosol model, including the particle size distribution of dust and smoke aerosols for the coarse and fine modes and the complex refractive index at the 412-865 nm wavelength band. On this basis, we applied a coupled ocean-air radiative transfer model to simulate the top-of-atmosphere (TOA) reflectance and remote sensing reflectance (Rrs(λ)）in the presence of absorbing aerosols. Based on a large number of radiation transfer simulations and machine learning models, we proposed a new atmospheric correction algorithm (OC-XGBRT) to reduce the impact of absorbing aerosols. OC-XGBRT considered the vertical distribution of absorbing aerosols and was used to retrieve the Rrs(λ) at the blue light band.
This dataset used CALIPSO aerosol product and ERA5 reanalysis meteorological product as input data. We established two different neural network models for both absorbing aerosols to retrieve the global vertical distribution parameters of dust and smoke aerosols based on Gaussian curve fitting. We produced the global monthly average aerosol mean height and standard deviation of aerosol height from 2007 to 2016. This dataset provides the aerosol vertical data support for atmospheric correction of strongly absorbing aerosols. On this basis, we developed the OC-XGBRT algorithm and applied it to the MODIS-Aqua ocean color sensor and validated it with in situ data from the SeaBASS and AERONET-OC stations. We compared the retrieval results of the OC-XGBRT with NASA SeaDAS, PLOYMER, and OC-SMART atmospheric correction algorithms in the presence of absorbing aerosols. The average absolute percentage deviation (APD) and root mean square error (RMSE) of OC-XGBRT are less than ~36.9% and ~5.5 ×10-4 sr-1, respectively. Results indicate that OC-XGBRT can provide more accurate remote sensing reflectance products in coastal and inland waters.
unfold
