Using In-Situ Aircore Measurements to Enhance a Satellite-based CO2 Neural Network Retrieval Algorithm

We introduce a hybrid approach to a neural network (NN) trace gas retrieval algorithm that integrates the JPSS NUCAPS CO2 trace gas product with satellite data inputs with in-situ AirCore CO2 measurements (using CrIS Cloud-Cleared Radiances as input). Our aim is to enhance existing CO2 retrieval models, particularly focusing on the lower troposphere and the planetary boundary level and providing improved total-column accuracy with respect to in-situ measurements. Initial results of our NN algorithm show promise (at first with the AirCore measurements only), exhibiting a Total Column mean absolute error bias of 1.7% and RMSE of approximately 2.1% (depending on which is optimized on) in comparison to ground-based AirCore profiles, although the NUCAPS CO2 Trace Gas product displays a lower raw bias of -0.7% and RMSE of 0.8% (which will improve the algorithm when added). This hybrid approach hopes to constructively combine the advantages of the satellite’s global observations with the accuracy of the local in-situ measurements (which has been used to [1]). This work aims to contribute to the advancement of monitoring of the dynamic and highly variable lower troposphere CO2 levels for ecological, air quality, climate, and weather applications.