Zotta et al. (2026) Improving AMSR2 vegetation optical depth retrievals via land parameter retrieval model parameter optimisation

Identification

• Journal: Remote Sensing of Environment
• Year: 2026
• Date: 2026-02-05
• Authors: Ruxandra-Maria Zotta, Richard de Jeu, Nicolas Francois Bader, Thomas Frederikse, Wouter Dorigo
• DOI: 10.1016/j.rse.2026.115286

Research Groups

• Department for Geodesy and Geoinformation, TU Wien, Vienna, Austria
• Transmissivity B.V., Alphen aan den Rijn, the Netherlands
• Planet Labs, Haarlem, the Netherlands

Short Summary

This study improves Vegetation Optical Depth (VOD) estimates from AMSR2 X-band observations by optimising key Land Parameter Retrieval Model (LPRM) parameters (surface roughness, effective temperature, single scattering albedo) through minimising brightness temperature residuals, demonstrating enhanced VOD-LAI seasonal agreement, especially in forests, but revealing trade-offs with soil moisture retrieval skill.

Objective

• To improve Vegetation Optical Depth (VOD) retrievals by reducing model residuals through a better parameterization of surface roughness (ℎ), effective temperature (𝑇eff), and single scattering albedo (𝜔) in the Land Parameter Retrieval Model (LPRM) using AMSR2 X-band data.

Study Configuration

• Spatial Scale: Global, 0.25° latitude-longitude grid.
• Temporal Scale: Daily observations from 2012 to 2021 (optimisation period: 2012-2014; validation period: January 2015-December 2021).

Methodology and Data

• Models used: Land Parameter Retrieval Model (LPRM v6.1), Powell algorithm (for parameter optimisation), Sobol method (for Global Sensitivity Analysis - GSA), ANOVA High-Dimensional Model Representation (HDMR).
• Data sources:
  • Satellite observations: AMSR2 (10.7 GHz X-band and 36.5 GHz Ka-band brightness temperatures), MODIS LAI (MOD15A2H, Collection 6.1).
  • Reanalysis data: ERA5-Land (skin temperature, volumetric soil water content in layer 1).
  • Ancillary data: ESA CCI Landcover v2.
  • In situ observations: Plate Boundary Observatory (PBO) Normalised Microwave Reflectance Index (NMRI), International Soil Moisture Network (ISMN) soil moisture measurements (0-2 cm depth).
  • Benchmark products: AMSR2 IB X-VOD (Wang et al., 2021), LPDR v3 VOD and θ (Du and Kimball, 2021).

Main Results

• A global sensitivity analysis revealed strong interactions between LPRM parameters (surface roughness, effective temperature, single scattering albedo), necessitating their simultaneous optimisation.
• Both optimisation scenarios (Microwave and ERA5) significantly reduced brightness temperature (𝑇B) residuals compared to the Default LPRM v6.1, with the Microwave scenario achieving lower residuals (below 0.06 K) than the ERA5 scenario (below 0.26 K).
• Both scenarios enhanced the temporal agreement between VOD and MODIS LAI seasonal cycles. The Microwave scenario improved Spearman's 𝑟 in 56.9% of pixels (mean gain +0.072), while the ERA5 scenario showed larger gains in 78.31% of pixels (mean gain +0.200), particularly in forests.
• For VOD anomalies, the Microwave scenario maintained or improved agreement with LAI more broadly, while the ERA5 scenario degraded in arid regions but retained gains in forests.
• Comparison with in situ Normalised Microwave Reflectance Index (NMRI) showed higher correlations for the Microwave scenario, indicating better tracking of day-to-day vegetation water content (VWC) variability, whereas the ERA5 scenario failed to capture this.
• Optimised parameters (ℎ1, Δ𝑇, 𝜔) differed significantly from default values, with higher average ℎ1 values suggesting a weaker dependence of ℎ on soil moisture at X-band.
• Improvements in VOD did not necessarily translate into better soil moisture (SM) estimates; 𝜃 skill decreased in both optimised scenarios, confirming known VOD–SM trade-offs. The Microwave scenario showed a median ISMN 𝜃 correlation of 0.43 (down from 0.51), while the ERA5 scenario showed 0.12.
• Benchmarking against other products showed ERA5-scenario VOD had the highest LAI correlation over 57% of land pixels (dense vegetation), Microwave-scenario VOD was best in parts of Europe, Africa, and Australia, and LPDR VOD dominated in short-vegetation biomes. Microwave scenario VOD showed the highest correlation with NMRI.

Contributions

• First study to perform joint, pixel-wise optimisation of three key LPRM parameters (surface roughness, effective temperature, single scattering albedo) for AMSR2 X-band VOD retrieval, guided by a global, pixel-wise Sobol' variance-based sensitivity analysis.
• Quantified singular and interaction effects of parameters on simulated brightness temperatures and VOD, explicitly demonstrating the need for joint rather than one-at-a-time tuning.
• Derived and provided global maps of relative standard error (RSE) estimates for the retrieved parameters (ℎ1, 𝜔, 𝑇eff), which are not available in other similar studies.
• Systematically diagnosed how two contrasting effective temperature (𝑇eff) configurations (microwave-derived vs. reanalysis-derived) propagate into VOD–LAI agreement, VOD–NMRI relationships, and soil moisture skill.
• Demonstrated the impact of reanalysis-independent 𝑇eff on VOD and soil moisture, recommending the microwave configuration for future VODCA products to avoid dependence on reanalysis priors.
• Recommended decoupling VODCA and ESA CCI Soil Moisture developments due to observed SM–VOD trade-offs.

Funding

• Wouter Dorigo acquired funding for this research. No specific projects, programs, or reference codes were listed. Article publishing charges were provided by TU Wien Library.

Citation

@article{Zotta2026Improving,
  author = {Zotta, Ruxandra-Maria and Jeu, Richard de and Bader, Nicolas Francois and Frederikse, Thomas and Dorigo, Wouter},
  title = {Improving AMSR2 vegetation optical depth retrievals via land parameter retrieval model parameter optimisation},
  journal = {Remote Sensing of Environment},
  year = {2026},
  doi = {10.1016/j.rse.2026.115286},
  url = {https://doi.org/10.1016/j.rse.2026.115286}
}