Fusion of MODIS and Landsat data to allow near real-time monitoring of land surface change

Abstract

A new methodology for fusion of MODIS and Landsat data improves monitoring of land surface change and snow mapping. This fusion method is based on prediction of MODIS data using a time-series of Landsat data. An underlying hypothesis is that the predicted MODIS images will form a more stable basis for comparison with new MODIS images than previous MODIS images. Correlations between predicted and observed MODIS images are higher than for successive days of MODIS data, confirming our hypothesis. Differences in the spectral signatures between predicted and real MODIS images become the "signal" used detect land surface change.

Tests of the fusion method to detect forest clearing show producer's and user's accuracies of 86% and 85%, respectively. Cleared patches of forest as small as 5-6 ha in size can be detected, a considerable improvement over current published results. Additionally, the fusion method can be used to map snow cover on a daily basis and is more accurate than current operational MODIS snow products. The encouraging results indicate that the fusion method holds promise for improving monitoring of land surface change in near real-time.