Motivated by a satellite remote sensing mission, this article proposes a multivariable errors-in-variables (EIV) regression model with heteroscedastic errors for relating the satellite data products to similar products from a well-characterized but globally sparse ground-based dataset. In the remote sensing setting, the regression model is used to estimate the global divisor for the satellite data. The error structure of the proposed EIV model comprises two components: A random-error component whose variance is inversely proportional to sample size of underlying individual observations which are aggregated to obtain the regression data, and a systematic-error component whose variance remains the same as the underlying sample size increases. In this article, we discuss parameter identifiability for the proposed model and obtain estimates from two-stage parameter estimation. We illustrate our proposed procedure through both simulation studies and an application to validating measurements of atmospheric column-averaged CO2from NASA’s Orbiting Carbon Observatory-2 (OCO-2) satellite. The validation uses coincident target-mode OCO-2 data that are temporally and spatially sparse and ground-based measurements from the Total Carbon Column Observing Network (TCCON) that are spatially sparse but more accurate. Supplementary materials for the article are available online.