Spatiotemporal correlation analysis of satellite-observed CO2: Case studies in China and USA

Observations of atmospheric carbon dioxide (CO₂) from the Greenhouse gas Observation SATellite (GOSAT) provide us new data sources for the global carbon research. However, the available GOSAT observations have gaps and are irregularly positioned. Geostatistics can be used to fill the gaps. The correlation modeling is one of the critical steps in geostatistical prediction (Kriging) and it is important to choose a suitable correlation model for Kriging. In this study, the spatio-temporal correlation structure of CO₂ data from GOSAT is estimated and modeled using the spatio-temporal variogram models. China and USA are selected as the study areas and compared during their variogram modeling process. Three different spatio-temporal variogram models, including the product model, the linear model and the product-sum model, are fitted to the empirical variogram surface of GOSAT observations in China and USA. Both weighted mean square errors (WMSE) and cross-validation are adopted to evaluate the modeling by the three models. As a result, the product-sum model performs the best in modeling and prediction accuracies, and the flexibility of using the product-sum model is also highlighted.