SARTA ANALYTIC JACOBIANS

Abstract

SARTA is a fast Radiative Transfer code with TwoSlab Cloud scattering ability [De SouzaMachado et al., 2018] that has been extensively tested for clear sky accuracy against e.g.,kCARTA [De Souza-Machado et al., 2020] and clouds [Aumann et al., 2018, 2023, De Souza-Machado et al., 2018]. It currently takes about 4 minutes to generate radiances for one granule of AIRS L1C data (2645 channels × 12150 FOVS), given input atmospheric profiles (thermodynamic and cloud) supplied by e.g.,radiosondes or Numerical Weather Prediction (NWP) model fields that are put into a 100 layer atmosphere model. This is more than adequate for processing observations for comparison purposes (given that line by line codes such as kCARTA take about 40 seconds to process one spectrum!). But if used for retrievals where estimates of jacobians are needed, the times become very large (for example would need to run it 100 times for T(z) jacobians, 100 times for WV(z) jacobians and another 100 times for O3(z) jacobians, or about 20 hours for all channels; more if you need e.g.,CH4, N2o etc jacobians). For these purposes, an analytic SARTA jacobian code has been written. This document outlines the changes to the SARTA code, and the ideas behind propagating the thermodynamic (temperature and constituent gases) analytic derivatives at layer j to the Top of Atmosphere (TOA); cloud jacobians still need to be done by finite difference.