Automatic Optical Depth Parametrization in Radiative Transfer Model RTTOV v13 via LASSO-Induced Sparsity for Satellite Data Assimilation

Abstract. The assimilation of satellite spectral sounder data requires fast and accurate radiative transfer models for retrieving surface and atmospheric variables. This study proposes a novel methodology to automatically parameterize atmospheric optical depths within the RTTOV scheme using statistical thresholds across pressure levels and LASSO regression to induce sparsity. Numerical experiments with VIIRS infrared channels demonstrate that this approach significantly reduces computational costs while maintaining accuracy. The sparsity also facilitates the automatic selection of absorbing gases and predictors by channel and pressure level, making it particularly effective for multispectral instruments with numerous atmospheric variables. These findings highlight the potential of sparse regression methods to enhance the efficiency of radiative transfer models for satellite data assimilation.