Applying FP_ILM to the retrieval of geometry-dependent effective Lambertian equivalent reflectivity (GE_LER) daily maps from UVN satellite measurements

Diego Loyola, Jian Xu, Klaus-Peter Heue, Walter ZimmerAtmospheric 2020 · DOI

Abstract

Accurate retrieval of atmospheric trace gases, clouds, and aerosols from satellite sensors requires precise knowledge of surface reflectivity. Traditional methods rely on static Lambertian equivalent reflectivity (LER) climatologies, which often lack spatial resolution and fail to account for viewing geometry or rapid surface changes like snow. This paper presents a novel algorithm using the Full-Physics Inverse Learning Machine (FP_ILM) to retrieve geometry-dependent effective LER (GE_LER) from UVN sensors. Applied to TROPOMI data, the algorithm demonstrates significant improvements in total ozone retrieval accuracy and spatial resolution.

Introduction: The Limitations of Static Climatologies

To measure trace gases such as ozone or $NO_2$ from space, scientists analyze the sunlight reflected from the Earth's surface back to the satellite. A critical parameter in this calculation is the Lambertian Equivalent Reflectivity (LER). Historically, researchers have used "climatologies"—static maps averaged over years of data from older satellites. However, these climatologies suffer from low spatial resolution, "geometry blindness" (ignoring how reflectivity changes with viewing angles, known as BRDF effects), and an inability to reflect real-time conditions, such as unseasonable snow cover.

Methods: The FP_ILM Framework

The researchers developed the Full-Physics Inverse Learning Machine (FP_ILM) to overcome these hurdles. This framework combines the accuracy of physical modeling with the speed of machine learning through a two-stage process.

The FP_ILM training phase flowchart, illustrating the integration of physical modeling and machine learning.
  1. Forward Modeling (Physics): Using the VLIDORT Radiative Transfer Model, the team simulated over 200,000 synthetic UV spectra. They employed "smart sampling" (Halton sequences) to ensure the training set comprehensively covered variations in ozone levels, surface pressures, and viewing geometries.
  2. Feature Extraction and Machine Learning: To avoid the computational burden of processing raw spectra, the team used DOAS (Differential Optical Absorption Spectroscopy) coefficients as "features." These coefficients compress complex light signals into meaningful parameters. A feedforward neural network was then trained to map these features to the GE_LER. Once trained, the retrieval is extremely fast, enabling real-time processing of massive satellite datasets.

Data and Validation

The algorithm was trained on synthetic VLIDORT data and applied to real-world measurements from the TROPOMI sensor on the Sentinel-5 Precursor (S5P) mission. Results were benchmarked against existing OMI and GOME-2 LER climatologies. Validation of total ozone columns was performed using CAMS (Copernicus Atmosphere Monitoring Service) reanalysis data.

Key Insights: Geometry and Gap-Filling

A major innovation of this work is the handling of surface anisotropy and temporal gaps. By accounting for the viewing angle (the "GE" in GE_LER), the model corrects for the fact that surfaces appear differently depending on the satellite's perspective.

The G3_LER daily map generation process for creating gapless global reflectivity data.

Furthermore, the researchers developed the G3_LER method to create "gapless" daily maps. This involves merging clear-sky GE_LER data from the current day with a rolling average of previous days, ensuring that a global map is always available, even for regions obscured by clouds. This dynamic approach allows the system to react to sudden surface changes, such as snowfall, which static climatologies would miss.

Results and Impact

The FP_ILM approach yielded several significant improvements:

  • Precision: The neural network retrieved reflectivity with a mean error of only 0.0016 ± 0.0018.
  • Ozone Accuracy: Using GE_LER for total ozone calculations reduced bias at high latitudes (80–60° S) from -2.61% (using OMI climatologies) to just 0.74%.
  • Spatial Resolution: The new maps offer a resolution of 0.1° x 0.1°, a five-fold improvement over the 0.5° resolution of previous OMI datasets.
  • Artifact Removal: Qualitative results show that the new method eliminates "ghost" features in ozone maps caused by incorrect snow/ice assumptions in regions like the Bering Strait.

Conclusion and Future Directions

While this study optimized the 325–335 nm spectral window for ozone, the FP_ILM framework is adaptable. Future work will extend this method to other trace gases like $NO_2$ and $SO_2$, and incorporate an "absorbing aerosol index" to mitigate interference from dust or smoke. The algorithm is poised to support upcoming geostationary missions like Sentinel-4 and TEMPO, providing high-frequency air quality monitoring across the globe.

Summary updated

Abstract

The retrieval of trace gas, cloud, and aerosol measurements from ultraviolet, visible, and near-infrared (UVN) sensors requires precise information on surface properties that are traditionally obtained from Lambertian equivalent reflectivity (LER) climatologies. The main drawbacks of using LER climatologies for new satellite missions are that (a) climatologies are typically based on previous missions with significantly lower spatial resolutions, (b) they usually do not account fully for satellite-viewing geometry dependencies characterized by bidirectional reflectance distribution function (BRDF) effects, and (c) climatologies may differ considerably from the actual surface conditions especially with snow/ice scenarios. In this paper we present a novel algorithm for the retrieval of geometry-dependent effective Lambertian equivalent reflectivity (GE_LER) from UVN sensors; the algorithm is based on the full-physics inverse learning machine (FP_ILM) retrieval. Radiances are simulated using a radiative transfer model that takes into account the satellite-viewing geometry, and the inverse problem is solved using machine learning techniques to obtain the GE_LER from satellite measurements. The GE_LER retrieval is optimized not only for trace gas retrievals employing the DOAS algorithm, but also for the large amount of data from existing and future atmospheric Sentinel satellite missions. The GE_LER can either be deployed directly for the computation of air mass factors (AMFs) using the effective scene approximation or it can be used to create a global gapless geometry-dependent LER (G3_LER) daily map from the GE_LER under clear-sky conditions for the computation of AMFs using the independent pixel approximation. The GE_LER algorithm is applied to measurements of TROPOMI launched in October 2017 on board the EU/ESA Sentinel-5 Precursor (S5P) mission. The TROPOMI GE_LER/G3_LER results are compared with climatological OMI and GOME-2 LER datasets and the advantages of using GE_LER/G3_LER are demonstrated for the retrieval of total ozone from TROPOMI.
BibTeX
@inproceedings{loyola2020applying,
  author = {Diego Loyola and Jian Xu and Klaus-Peter Heue and Walter Zimmer},
  title = {{Applying FP\_ILM to the retrieval of geometry-dependent effective Lambertian equivalent reflectivity (GE\_LER) daily maps from UVN satellite measurements}},
  booktitle = {Atmospheric},
  year = {2020},
  doi = {10.5194/amt-13-985-2020},
}
Figure