Estimating optically relevant components of the water column from ocean-color satellite missions
Sundarabalan V Balasubramanian, University of Maryland Baltimore County, sundarv1@umbc.edu (Presenter)
Nima Pahlevan, SSAI / NASA GSFC, nima.pahlevan@nasa.gov
Ryan O'Shea, SSAI / NASA GSFC, ryan.e.o'shea@nasa.gov
Mixture Density Networks (MDNs) are a class of deep learning models, successfully demonstrated for the accurate retrieval of surface concentrations from coastal and inland waters. In this study, we developed a robust MDN-based model to retrieve 10 different optically relevant variables, including chlorophyll-a (Chla) and total suspended solids (TSS), and the absorbing components of inherent optical properties (IOPs) from the remote sensing reflectance (Rrs). The MDN is trained and validated on in situ spectra from GLORIA, augmented with globally distributed in situ IOP measurements (N=10,050). For training and validation, the hyperspectral in situ radiometric and absorption datasets were resampled, via the relative spectral response functions of MODIS, MERIS, and VIIRS, to simulate the response of each multispectral mission. The retrieved parameters from the validation dataset have variable uncertainty represented by the Median Symmetric Accuracy (MdSA) for each parameter and sensor combination. Of the 10 retrieved variables from the MDN, MERIS retrieves acdom(443) with a minimum MdSA, while MODIS and VIIRS retrieve aph(443) with the minimum MdSA. TSS was the parameter with the largest MdSA for all three sensors (MODIS, VIIRS, and MERIS). The average MdSA over all 10 variables was 26.7%, 31.5%, and 28.5% for MERIS, VIIRS, and MODIS, respectively. The overall performance of the MDN model presented here was also analyzed for the near-simultaneous images of MODIS and VIIRS as well as MODIS and MERIS to understand the multimission robustness and consistency in retrieved variables. In summary, the developed MDN is capable of robustly retrieving 10 water quality variables for monitoring the coastal and inland water from multiple multispectral satellite sensors (MODIS, MERIS, and VIIRS).
Associated Project(s):
Poster Location ID: 2-58
Presentation Type: Poster
Session: Poster Session 2
Session Date: Wed (May 10) 5:15-7:15 PM
CCE Program: OBB