Miliaresis, GC (2012). Elevation, latitude and longitude decorrelation stretch of multi-temporal near-diurnal LST imagery. INTERNATIONAL JOURNAL OF REMOTE SENSING, 33(19), 6020-6034.
A method is presented for elevation, latitude and longitude decorrelation stretch of day (1:30 pm) minus night (1:30 am) monthly averaged land surface temperature (LST) differences captured by the Moderate Resolution Imaging Spectroradiometer (MODIS) which correspond to near-diurnal LST (NDL) estimates. Principal components 1 and 2 (PC1 and PC2) account for 88.9% of the variance evident in NDL imagery. Multiple linear regression analysis of PC1 and PC2 quantifies the variance explained by elevation, latitude and longitude. Predicted and residual images of PC1 and PC2 account for 45.37% and the 43.53% of the variance, respectively. If dimension reduction (omit PC1 and PC2) is applied, then 43.53% of the variance that is independent of elevation, latitude and longitude will be subtracted from the reconstructed NDL data set. Selective variance reduction reconstructs the NDL imagery from the PC1 and PC2 residual images as well as the remaining PCs and thus only the portion of variance that is not related to elevation, latitude and longitude is taken into account. The reconstructed imagery shows the amount by which the standardized NDL value per pixel deviates from the predicted elevation, latitude and longitude. Clustering of the reconstructed data set identified positive and negative NDL anomalies in the USA and Mexico.