A new model for simulating microwave radiation
The UDL-RT framework uses neural-network parameterizations to simulate all-sky microwave radiative transfer.
Researchers created the Unified Differentiable and Learnable Radiative Transfer model, or UDL-RT. This framework is for all-sky microwave radiative transfer.
The model couples neural-network parameterizations with an explicit radiative transfer solver. It uses a unified differentiable computational graph.
Radiative transfer models simulate radiances and quantify sensitivities to surface and atmospheric conditions. Parameterizations represent complex nonlinear relationships. Bulk scattering properties describe hydrometeor properties. Ocean surface emissivity describes how the ocean surface emits radiation.
The neural-network gas-absorption parameterization reduces brightness-temperature errors by 60-85% in water-vapour channels. UDL-RT uses automatic differentiation to compute Jacobians with respect to input variables.
Using Microwave Humidity Sounder-2 (MWHS-2) observations, the model optimizes the snow bulk scattering properties parameterization. This optimization improves agreement between simulated and observed brightness-temperature distributions.
The framework supports future remote sensing retrieval and data assimilation applications. It integrates radiance simulation, sensitivity analysis, and parameterization optimization.
Key facts
- The UDL-RT model reduces brightness-temperature errors in water-vapour channels by 60-85%.
- The model uses Microwave Humidity Sounder-2 (MWHS-2) observations to optimize snow bulk scattering properties.
- UDL-RT integrates radiance simulation, sensitivity analysis, and parameterization optimization.