Geoestatistical method of satellital image restoration Landsat in cloudy territories

Abstract

Satellite images are an important source of information in the study of the earth's surface. The result of their analysis can be influenced by atmospheric conditions in which they were taken. Cloudiness is the most common interference and its presence can be detrimental to processing. The objective of this study is to eliminate clouds on satellite imagery. Consequently, a surface estimate beneath the clouds is carried out. A method based on GNSPI is proposed. This is a geostatistical method originally devised for removal of voids (regions scanned by the satellite) caused by a fault in the SLC sensor Landsat-7 solution. It has been shown that there is a linear relationship between images taken at different dates. Therefore, we can make an estimate of the surface beneath the clouds from an image without clouds. The calculated estimate (temporal prediction) is not complete and does not consider local variations soil, so a second estimate is needed. The prediction of local variations is performed using a geostatistical method: Kriging. The main advantage of this geostatistical method over the deterministic is that it has good results on heterogeneous surfaces. We assume that similar pixels show similar patterns in the spectral variation between different dates. Therefore, one can get a good estimate if used exclusively similar pixels for prediction. The final estimate is the sum on the temporal estimate and the local estimate.