XCO<sub>2</sub> estimates from the OCO-2 measurements using a neural network approach

Abstract. The Orbiting Carbon Observatory (OCO-2) instrument measures high-resolution spectra of the sun's radiance reflected at earth's surface or scattered in atmosphere. These are used to estimate column-averaged dry air mole fraction CO2 (XCO2) and pressure. official retrieval algorithm (NASA's Atmospheric Observations from Space retrievals, ACOS) is a full-physics has been extensively evaluated. Here we propose an alternative approach based on artificial neural network (NN) technique. For training evaluation, use as reference estimates (i) pressures numerical weather model (ii) XCO2 derived atmospheric transport simulation constrained by air-sample measurements CO2. NN trained here using real acquired nadir mode cloud-free scenes during even-numbered months then evaluated against similar observations odd-numbered months. evaluation indicates that retrieves pressure with root-mean-square error better than 3 hPa 1σ precision 0.8 ppm. statistics indicate representative set data allows excellent accuracy slightly algorithm. An spectrophotometer retrievals for ACOS estimates, skill varies among various stations. NN–model differences show spatiotemporal structures potential improving our knowledge fluxes. We finally discuss pros cons this processing OCO-2 other space missions.