Aquaverse: An Aquatic Inversion scheme for remote sensing of fresh and coastal waters
Akash Ashapure, SSAI / NASA GSFC, akash.ashapure@nasa.gov (Presenter)
Nima Pahlevan, SSAI / NASA GSFC, nima.pahlevan@nasa.gov
William Wainwright, SSAI / NASA GSFC, william.wainwright@nasa.gov
Brandon Smith, SSAI, b.smith@nasa.gov
Arun M Saranathan, SSAI / NASA GSFC, fnu.arunmuralidharansaranathan@nasa.gov
Pengwang Zhai, University of Maryland Baltimore County, pwzhai@umbc.edu
The primary challenge in aquatic remote sensing is developing a robust atmospheric correction (AC) method to estimate remote sensing reflectance (Rrs) products. Here, we describe the utility of Mixture Density Networks (MDNs) in an AC framework, termed Aquaverse, for processing Landsat-8/-9 (L8/9) and Sentinel-2A/B (S2A/B) images over inland and nearshore coastal waters. First, a coupled ocean-atmosphere radiative transfer (RT) model was adapted to simulate top-of-atmosphere reflectance spectra (rhot) for various imaging geometries and environmental conditions. Second, the hyperspectral pairs of rhot- Rrs spectra (@ 5 nm) were resampled with sensor spectral response functions and used subsequently to train MDNs to transform rhot to Rrs with the associated uncertainties. The trained model was applied to L8/9 and S2A/B imagery for an extensive matchup assessment obtained from AERONET-OC products (N ~ 800) and other field-measured datasets (N ~ 700). Compared to the state-of-the-art AC processors, our matchup analyses showed performance improvements from 20 to 100% depending on the spectral band and statistical metrics. Median estimated errors for L8/9 and S2A/B matchups combined were ~ 45%, 35%, 20%, and 30% for the 443, 482, 560, and 655 nm spectral bands, respectively. Aquaverse-generated maps for several images exhibited similar spatial patterns as those produced by other processors, and uncertainty maps conformed to the model’s expected performance. We demonstrate that Aquaverse can be utilized as one alternative data processing technique in fresh and coastal waters.
Associated Project(s):
Poster Location ID: 1-17
Presentation Type: Poster
Session: Poster Session 1
Session Date: Tue (May 9) 5:00-7:00 PM
CCE Program: OBB