Assessing the uncertainty of non-change in national-scale vegetation mapping using 3D wavelet transformed NDVI time series

The national or international mapping projects have usually long interval to update. To complement this long interval, satellite imagery obtained periodically and continuously is useful. This study is a preliminary step to generate annual spatial datasets comparable to the existing vegetation cover (EVC) layers using satellite imagery. We introduced and tested an approach based on 3D wavelet transform to analyze the multi-temporal satellite images. 3D wavelet transform based analysis was experimentally applied to NDVI images where identifying non-changed pixels. 3D wavelet transform can extract important patterns such as long term trends and short term events and analyze spatial and temporal relations with neighboring pixel. As a result, 3D wavelet transformed LLL values were much more stable across time than those of the original NDVI in non-changed areas. This study can provide fundamentals for image correction to ensure image comparability across years.