研究目的
To introduce a methodology for quantifying the uncertainty in the retrieval of aerosol optical thickness (AOT), focusing on uncertainty due to aerosol microphysical model selection and uncertainty due to imperfect forward modelling, applied to OMI measurements.
研究成果
The Bayesian methodology improves uncertainty characterization in AOT retrieval by accounting for model selection and forward model errors. It demonstrates that uncertainties are often underestimated when using single models, and the approach provides more realistic error estimates. The method is general and applicable to other satellite instruments.
研究不足
The study focuses on over-land retrievals and does not address uncertainties from surface albedo assumptions. The model discrepancy covariance is estimated globally and may not capture location-specific variations. Prior distributions are weakly informative and uniform, which might not be optimal for all cases. The method is computationally intensive for higher-dimensional parameters.
1：Experimental Design and Method Selection:
The methodology involves Bayesian model selection and model averaging to weight aerosol microphysical models based on goodness of fit, and Gaussian processes to model systematic discrepancies between measured and modelled reflectances. The spectral correlation is determined empirically from residuals.
2：Sample Selection and Data Sources:
Uses OMI reflectance measurements from cloud-free, over-land pixels, with examples from specific orbits and dates (e.g., Greece forest fires, Russian wildfires, Sahara dust storms). Data from AERONET ground-based measurements are used for validation.
3：List of Experimental Equipment and Materials:
OMI instrument on NASA's Aura satellite, OMAERO algorithm, aerosol microphysical models (e.g., desert dust, biomass burning, weakly absorbing, volcanic aerosols), look-up tables (LUTs) for radiative transfer calculations.
4：Experimental Procedures and Operational Workflow:
Retrieve AOT by fitting models to observed reflectances using least squares minimization, incorporate model discrepancy via Gaussian processes, compute posterior probabilities and averaged distributions.
5：Data Analysis Methods:
Bayesian inference for parameter estimation and model comparison, numerical integration for evidence calculation, empirical semivariogram for covariance estimation.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
