Document Type



Collecting long-term satellite image series in high latitudes has been challenging due to its short growing season. For off-peak imagery, its reflective properties need to be corrected to maintain the spectral consistency. This study compares three statistical approaches to reconstructing a 30-year normalized difference vegetation index (NDVI) series for forest recovery assessment after the 1987 Black Dragon Fire in the Greater Hinggan Mountains Forest, Northeast China. To correct the off-peak NDVI to peak NDVI, the Landsat paired regression takes advantage of the scene-to-scene linear relationship between the two images, the GIMMS booster approach compensates the NDVI increment rate based on the 15-day AVHRR NDVI3g products between the two dates, and the climatic adjustment approach compensates the absolute NDVI change relying on the non-linear climatic influence on forest growth. The results find that the Landsat paired regression achieves the best performance with the image-wide residues within ± 0.2. The climatic adjustment picks the first-level NDVI under climatic control, while the GIMMS booster is heavily affected by the NDVI3g data quality. All approaches agree that the early-season (May-June) images are better sources for NDVI series reconstruction. The late-season images (especially October) are subject to fall senescence and early snowfall and therefore, are not recommended for satellite image series in high-latitude forests. The reconstructed NDVI series effectively corrects the off-season troughs along the trajectory. NDVI in burned forests increase rapidly in five years, and the heavily burned forests have the highest rate. Forest greenness could recover back to normal in ten years. This study confirms the feasibility of off-peak correction for building sparse image series. With advanced data availability such as the 5-day Sentinel-2 (10–60 m), daily MODIS imagery (500–1000 m), and hourly climatic reanalysis dataset (1–10 km), all three proposed approaches in this study could be improved for better application for post-fire monitoring of high-latitude forests.

Digital Object Identifier (DOI)


© The Authors. Published by Elesvier B.V. This is an open access article under the CC BY-NC-ND license (

APA Citation

Morgan, G. R., Wang, C., Li, Z., Schill, S. R., & Morgan, D. R. (2022). Deep Learning of High-Resolution Aerial Imagery for Coastal Marsh Change Detection: A Comparative Study. ISPRS International Journal of Geo-Information, 11(2), 100.

Included in

Geography Commons