Machine-learned gas optics with a focus on geostationary extended observations (GeoXO) for improving water vapor observations in the lower atmosphere
Solanki, Dhyey, author
Chiu, Christine, advisor
Kummerow, Christian, committee member
Jathar, Shantanu, committee member
In the grand scheme of the earth-atmosphere system, there are few constituents more vital and mysterious than water vapor. Vital because of its interwoven thermodynamic, radiative, and dynamic influence on the weather and climate of the planet, and mysterious because of our limited capacity in observing its time evolution in horizontal and vertical space. The advancements in the spectral and radiometric accuracy of next-generation infrared sounders are expected to bring unprecedented value to our observational capability with improved profiling of lower tropospheric water vapor where it is most abundant. Essential to performing satellite observations and their assimilation to dynamical models is the accurate and efficient radiative transfer calculations. In this process, calculating the atmospheric absorption by various gases is one of the most important steps. The 'line-by-line' approach of computing the influence of every absorption and emission line is operationally impractical for many observations that can contain hundreds of absorption lines. The existing radiative transfer models, therefore, use parameterized gaseous absorption using methods like pre-computed lookup tables or regression methods. The conventional methods compute channel values and can only be used for a specific sensor and channel. Here, we present a new method of performing gas absorption calculations using machine learning that can be applied to the spectral interval of any channel. With an example spectral interval of the new water vapor channel on the upcoming GeoXO infrared sounder, we train neural networks to emulate the line-by-line layer optical depths on a consistent grid of 100 atmospheric layers defined by 101 pressure levels spanning from 1100 hPa to 0.005 hPa. We sample a diverse set of 8640 profiles around the globe for the year 2014 from the Medium-Range Weather Forecasts (ECMWF) atmospheric reanalyses dataset (ERA5) and use 80% of these profiles as training data and 20% of the profiles as validation data. We test the performance of the emulators using a completely independent set of 83 profiles from ECMWF for the year 2006-2007, known as ECMWF83 profiles that have been widely used for training the atmospheric transmittance due to gas absorptions. The atmospheric optical depth used as the truth in all datasets is calculated from the line-by-line Monochromatic Radiative Transfer Model (MonoRTM). The evaluation results from the testing dataset show that the trained neural networks are able to predict line-by-line layer optical depths with a mean percent error of 0.47%. Radiative transfer models used for simulating satellite radiances, like Community Radiative Transfer Model (CRTM), require channel layer-to-space transmittance profiles for solving the radiative transfer equation. Transmittance profiles were calculated using the predicted line-by-line layer optical depths with a mean percent error of 0.02%. Further, the predicted values are also able to accurately calculate the channel weighting functions with the mean percent error of 0.13%. The results show the feasibility of utilizing neural networks in predicting line-by-line optical depths that can be applied for any spectral interval and can be highly useful for the designing of future sensors.
Includes bibliographical references.
Includes bibliographical references.
gas absorption parameterization