A method for colocating satellite XCO2 data to ground-based data and its application to ACOS-GOSAT and TCCON
- 1Jet Propulsion Laboratory, California Institute of Technology, Pasadena, CA, USA
- 2California Institute of Technology, Pasadena, CA, USA
- 3Colorado State University, Fort Collins, CO, USA
Abstract. Satellite measurements are often compared with higher-precision ground-based measurements as part of validation efforts. The satellite soundings are rarely perfectly coincident in space and time with the ground-based measurements, so a colocation methodology is needed to aggregate "nearby" soundings into what the instrument would have seen at the location and time of interest. We are particularly interested in validation efforts for satellite-retrieved total column carbon dioxide (XCO2), where XCO2 data from Greenhouse Gas Observing Satellite (GOSAT) retrievals (ACOS, NIES, RemoteC, PPDF, etc.) or SCanning Imaging Absorption SpectroMeter for Atmospheric CHartographY (SCIAMACHY) are often colocated and compared to ground-based column XCO2 measurement from Total Carbon Column Observing Network (TCCON).
Current colocation methodologies for comparing satellite measurements of total column dry-air mole fractions of CO2 (XCO2) with ground-based measurements typically involve locating and averaging the satellite measurements within a latitudinal, longitudinal, and temporal window. We examine a geostatistical colocation methodology that takes a weighted average of satellite observations depending on the "distance" of each observation from a ground-based location of interest. The "distance" function that we use is a modified Euclidian distance with respect to latitude, longitude, time, and midtropospheric temperature at 700 hPa. We apply this methodology to XCO2 retrieved from GOSAT spectra by the ACOS team, cross-validate the results to TCCON XCO2 ground-based data, and present some comparisons between our methodology and standard existing colocation methods showing that, in general, geostatistical colocation produces smaller mean-squared error.