Day 4

Detailed paper information

Back to list

Paper title Uncertainty characterization and mitigation of inland water remote sensing reflectance using an autonomous profiler: The case of Lake Geneva
Authors
  1. Abolfazl Irani Rahaghi Eawag, Swiss Federal Institute of Aquatic Science and Technology Speaker
  2. Camille Minaudo EPFL
  3. Alexander Damm-Reiser University of Zurich
  4. Remika Gupana Eawag, Swiss Federal Institute of Aquatic Science and Technology
  5. Daniel Odermatt Eawag, Swiss Federal Institute of Aquatic Science and Technology
Form of presentation Poster
Topics
  • A7. Hydrology and Water Cycle
    • A7.06 EO for monitoring water quality and ecological status in inland waters
Abstract text Inherent Optical Properties (IOPs), such as absorption and scattering, link the biogeochemical composition of water and the Apparent Optical Properties (AOPs) obtained from satellites, including remote sensing reflectance (Rrs). The so-called optical closure analysis between radiometrically-measured AOPs and simulated AOPs from measured IOPs and light-field boundary conditions is crucial for assessing and, ideally, minimizing the uncertainties associated with AOP-to-IOP inversion algorithms. However, this step is complicated due to several factors, e.g., the unknown bias and random errors in the individual measurements, limitations in the sampling of the Volume Scattering Function (VSF) and fluorescence emission, and uncontrolled environmental effects, causing uncertainties in the AOPs and water constituents retrieval.
In this study, we used in-water bio-optical data acquired by an autonomous profiler (WetLabs Thetis) several times a day, as well as Sentinel-3 OLCI radiance and reflectance products to quantify, characterize, and mitigate the uncertainty of Rrs estimates. Various bio-optical sensors, as well as a Conductivity-Temperature-Depth (CTD) probe, are mounted on this profiler. Hyperspectral downwelling irradiance and upwelling radiance (Satlantic HOCR; 189 channels between 300-1200 nm), hyperspectral absorption and attenuation (AC-S; 81 channels between 400-730 nm), backscattering at 440, 532, 630 nm at 117° (ECO Triplet BB3W), as well as backscattering at 700 nm at 117° and Chlorophyll-a fluorescence (ECO Triplet BBFL2w) measured at an offshore research platform in Lake Geneva (Switzerland/France), called LéXPLORE (https://lexplore.info/), were used to address the scientific objectives. The in situ dataset includes 294 high vertical resolution daily profiles for the period between 10/2018 and 5/2020. The quasi-concurrent Sentinel-3 data (within ±2 hr of the in situ measurements) were used to assess the performance of the proposed uncertainty characterization and mitigation. The POLYMER atmospheric correction was used to obtain Rrs. We tested two bio-optical models available in POLYMER: (i) the globally optimized model by Garver, Siegel and Maritorena (GSM01), and (ii) the model proposed by Park and Ruddick (PR05). 41 and 31 matchups are available for the GSM01 and PR05 models, respectively.
The Hydrolight (HL) radiative transfer model was employed to obtain Rrs from the measured IOP profiles. We used a combination of different metrics based on the residuals of IOP-derived and radiometrically-measured Rrs to quantify and characterize the optical closure. Using the raw IOP profiles, our closure study indicated 33% of profiles with both error and bias of < 15% (i.e., good closure), and 18% with an error or bias of > 30% (i.e., poor closure). We then investigated the effect of scattering corrections for AC-S measurements, which only slightly improved the results (38% good and 21% poor closure). Next, we evaluated a simple single-step backscattering ratio (Bp) optimization method based on Rrs residuals, which significantly improved the Rrs optical closure (99% good, and 0% poor closure). The resulting optimized Bp shows a plausible seasonal variation ranging from ~0.005 during winter to ~0.024 during the end of spring and the beginning of summer. Our study confirmed that Bp, or more generally the VSF, is the most sensitive parameter in estimating AOPs from IOPs.
We further investigated the effects of uncertainty characterization (i.e., profile clustering) and uncertainty mitigation (e.g., IOPs correction) on the in situ-derived and Sentinel-3 Rrs matchup analysis. The latter showed a similar pattern to pure in situ analyses, i.e., slight enhancement using AC-S scatter-corrected profiles, and recognizable improvement implementing backscattering optimization. To avoid any overfitting by the backscattering optimization, the AC-S scatter-corrected profiles were used for investigating the effect of profiles clustering on matchup analysis. The results revealed that the uncertainty clustering based on the in situ profiler optical closure exercise can be used for Sentinel-3 matchup analysis, i.e., profiles with good closure indicated better performances based on different metrics as compared with poor closure. The satellite-derived Rrs using both PR05 and GSM01 models showed similar patterns in analyzing the effect of uncertainty characterization and mitigation with only slightly better results employing PR05.
Ultimately, we used profiles with good optical closure in other wavelengths to estimate phytoplankton fluorescence quantum yield in the emission region (670-700 nm). By relating these estimates to irradiance and pigment concentration, we managed to derive realistic diurnal estimates of non-photochemical quenching (NPQ) across the euphotic layer. In doing so, we can explain the limitations of fluorescence-based Chlorophyll-a retrieval algorithms for oligo- to mesotrophic lakes, and characterize the impact of photoinhibition on daily integrated primary production estimates.
Our results, in general, highlight the potential of using autonomous optical profiling as an alternative for automated ground-truthing of AOPs, with the added value of simultaneous IOP measurements. Further research is needed to investigate if improved VSF measurements and hence better estimates of Bp, or the consideration of full polarization in radiative transfer simulations enable improved conclusions from optical closure assessments.