An unstandardized selective variance reduction script for elevation, latitude and longitude decorrelation stretch of multi-temporal LST imagery

Abstract

A selective variance reduction (SVR) script is presented that applies linear regression models to the principal components (PCs) of multi-temporal night monthly averaged land surface temperature (LST) imagery, in an attempt to spit the variance associated to elevation, latitude, longitude. The recently released version 6, MODIS LST data (MYD11C2) with spatial resolution 0.05° is used while the method is applied in SW USA. The innovation relies on the use of un-standardized PCs. Thus, the reconstructed LST should express the deviation in degrees Celsius, from the elevation, latitude, longitude predicted LST. The reconstructed data quantifies the temporal and spatial patterns of thermal anomalies. The modeling of the frequency distributions of the reconstructed LST imagery indicate that possibly snow melting and variations in water table depth are associated to the negative thermal anomaly observed during the Spring. The GNU OCTAVE 4.0 SVR software implementation is available at http://selective-variance-reduction.sourceforge.net for testing and evaluation.