Estimating and removing sensor‐induced correlation from advanced very high resolution radiometer satellite data

Abstract

In order to estimate the spatial covariance structure of sea surface temperature (SST) using advanced very high resolution radiometer (AVHRR) satellite data, autocorrelation induced by the sensor itself must be determined and then removed before the correlative properties of SST per se can be found. Sensor‐induced autocorrelation arises from (1) partially redundant sampling and (2) a nonideal impulse response. Simulation techniques are used to model the correlation structure produced by the AVHRR. Results indicate that an autocorrelation of 0.39 at lag 1 arises from the overlap of adjacent pixels in the alongscan direction. Autocorrelations of 0.51 (alongscan) and 0.43 (alongtrack) arise from the sensor impulse response at lag 1. The autocorrelation at lag 2 is still significant (0.07) due to the sensor impulse response in the alongscan direction. Overall, an autocorrelation of 0.46 occurs at lag 1 due to the combined effects of pixel overlap and the sensor impulse response, at nadir. Procedures for removing this sensor‐induced correlation are presented. Oceanographically, these results indicate, for example, that the gradients associated with sharp oceanic fronts will be suppressed. Application of the results to observed AVHRR satellite data indicate that correlation length scales for SST over one portion of the Gulf Stream are on the order of 20 km. The results also indicate that the effective spatial resolution of the AVHRR is roughly 1.25 km at nadir, somewhat lower than the commonly accepted value. Finally, it is concluded that the accuracy of the calculations is limited by (1) an incomplete specification of the AVHRR system modulation transfer function (MTF) and (2) the sine wave response function that is used to approximate the system MTF.