A Fast Machine Learning Emulator of the Community Radiative Transfer Model for Assimilation of Advanced Baseline Imager Observations

The continuous improvement in weather forecast skill over the past several decades is largely due to the increasing quantity of available satellite observations and their assimilation into operational forecast systems. Assimilating these observations requires observation operators in the form of radiative transfer models. Significant efforts have been dedicated to enhancing the computational efficiency of these models. Computational cost remains a bottleneck, however, and a large fraction of available data goes unused for data assimilation as a result. To address this, we used machine learning to build an efficient neural network (NN) based probabilistic emulator of the Community Radiative Transfer Model, applied to the GOES Advanced Baseline Imager (ABI). We will present results from the trained NN emulator which predicts CRTM results as well as the corresponding error with respect to CRTM. Across all channels, the RMSE of the predicted brightness temperature is 0.3 K. For clear sky conditions the RMSE is less than 0.1 K for 9 out of 10 channels. The error predictions are generally reliable across a wide range of conditions. Explainable AI methods demonstrate that trained emulator reproduces the relevant physics, increasing confidence that the model will perform well when presented with new data. Future work will involve integrating the trained emulator into the Unified Foreword Operator (UFO) within the JEDI framework. Observing system experiments can then be performed to evaluate the new operator.