Estimation of NDVI Images Using Geostatistical Methods

The objective of this article is to examine the effect of NDVI (Normalized Difference Vegetation Index) images estimation using four geostatistical algorithms: Ordinary Kriging (OK), Universal Kriging (UK), Indicator Kriging (IK), and Sequential Indicator Simulation (SIS). Using the undersampled NDVI data obtained from a NOAA/AVHRR image, both estimates and variances from the three Kriging methods are obtained. It is found out that the images of estimates from Ordinary Kriging and Universal Kriging can both successfully restore the overall trend of the original image, but the image of estimates from Indicator Kriging is not good enough. It has also been found out that the images of variances from Ordinary Kriging and Universal Kriging only reflect the sampling configuration because they are independent of the local data values. However, owing to the fact that variances from Indicator Kriging are conditional on the local data values, they show the errors of estimates perfectly, and the values of variances from Indicator Kriging are consistent with the uncertainty of the original NDVI data. Using the undersampled data of NDVI, the multiple realizations from SIS exhibit strong variability, and the mean image from multiple realizations has a clear smoothing effect, with a low exactness. However, the image of variances from multiple realizations can successfully show the distribution of data uncertainty.