Selective variance reduction of multi-temporal LST imagery in the East Africa Rift System

Abstract

A new method for elevation and latitude decorrelation stretch of multi-temporal land surface temperature (LST) in the East Africa Rift System (EARS) from MODIS 2008 monthly average night imagery and Globe digital elevation model (DEM) is presented. Multiple linear regression analysis of principal components images (PCAs) quantifies the variance explained by elevation and latitude. Selective variance reduction (SVR) reconstructs the multi-temporal LST imagery from the residual images and selected PCAs by taking into account the portion of variance not related to elevation and latitude. Clustering of the reconstructed imagery identifies two major thermal anomalies a) in the Afar Triangle, and b) a new one in between the Ethiopia and Kenya. These regions present LST values higher than the elevation and latitude predicted ones through out the year. It is assumed that the new thermal anomaly corresponds to a triple junction formed in between the Ethiopian Rift and the Eastern and the Western branches of the EARS, in an area where active volcanoes and mantle plume activity concentrate. SVR is expected to assist tectonic and volcanic zones characterization on the basis of their thermal response.