A statistical model for latitudinal correlations of satellite data

[1] A new class of models is developed to represent the latitudinal correlation patterns of satellite-based measurements of ozone and temperature. We employ the monthly average series for each 5-degree latitudinal zone of the TOMS data from the Nimbus 7 satellite over 14 years from November 1978 to November 1992. For each latitudinal zone a temporal regression model that includes systematic seasonal variations, possible long-term trends, solar radio flux effect, and time dependency is fitted separately by maximum likelihood. From the residuals at all latitudinal zones, we observe strong contemporaneous latitudinal correlations, which decay at different rates depending on latitudes and become negative at moderate distances. A three-component model with two associated weight functions is developed that can represent the main features of the concurrent spatial correlations of the residuals. We use low-order ARMA models as components. A similar analysis is done on the latitudinal correlations of the residuals from the monthly average series of Climate Prediction Center temperature data at 30 mbar from January 1979 to December 1993. The new class of models can be applied to other remotely sensed geophysical data sets whose covariance structures exhibit similar patterns.