An NDVI synthesis method for multi-temporal remote sensing images based on k–NN learning: a case based on GF-1 data

Abstract

ABSTRACTGaofen-1 (GF-1) satellite data has the advantages of having high temporal resolution and wide coverage. Therefore, normalised difference vegetation index (NDVI) data collected by GF-1 can provide an accurate assessment of vegetation coverage. Because of its limited range and the interference of cloud cover, NDVI data should be synthesised by using multi-day data, which creates a more reliable dataset. However, NDVI data synthesised by existing methods have poor continuity and reliability. To overcome these problems, an NDVI synthesis method for multi-temporal remote sensing images based on k-nearest neighbour (k-NN) learning is proposed in the present study. Based on a k-NN learning algorithm and the continuity in spatial and temporal aspects of NDVI data, multi-temporal remote sensing image data was screened and classified to remove cloud cover. Then, each image was assigned a weight, based on which the data weighting fusion could be achieved. Compared with the Maximum Value Compositing and the Average Compositing methods, the k-NN method proposed in this study was found to remove the mutation points more effectively, ensuring better spatial continuity of NDVI data and improving the reliability of the results.