<title>Distribution of cloud-free Landsat images in the Barents region</title>

Abstract

The utility of optical satellites depends heavily on latitude and cloud cover. In the north-eastern corner of the Fenno-Scandinavian landmass, extreme seasonal variations in sunlight and the clashing of weather fronts severely limit the season for optical satellites. In this paper, we examine the distribution of cloud-free Landsat images of the Barents Region during 1984-93. Based on cloud-cover data from ground station archives, we derive the observed frequency of cloud-free images (having less than 10 per cent cloud cover) of each of 10 scenes over 10 years. The data show that cloud-free images are very rare, and that they occur as isolated events in time. If the cloud-free images are essentially randomly distributed in space and time, they can be modelled by a Poisson distribution with unknown mean. Invoking the method of maximum likelihood, we estimate this parameter by the sample mean and hence compute the estimated expected frequencies. We find that they agree well with the observed frequencies. Next, we perform a X2 test to check the validity of the model. The computed test statistic falls below the critical one. Hence we conclude that the cloud-free images are randomly distributed in space and time.