Hi everyone! We are using NDVI to build a time series based on images from Landsat 5 and 8. However, the values from these two satellites are slightly different. Does anybody know about a procedure (i.e. a correction or pre-processing) that generates equivalence between the NDVI values of these two sources?
For different satellite data, sometimes minor differences are expected due to different projection methods and the satellite positions. However, if difference is more, then you need to re-check the pre-processing including the projection method.
Hi João,
Firstly, you probably want to make sure you are calculating the NDVI based on the surface reflectance images. After that, if you still find differences, you can run a regression with the overlapping period of the sensors to get multiplicative and additive factors for calibration for each band or for the NDVI.
Best,
João Custódio Fernandes Cardoso this is a known issue that some have tried to tackle, however there is still room for improvement. You can follow the articles by Roy et al. (2016) and Villaescusa-Nadal et al. (2019) for ideas from the following discussion:
https://www.researchgate.net/post/Is_there_any_aggregated_Spectral_Band_Adjustment_Factor_SBAF_for_Landsat_5_and_8
You may want to look at:
Preprint A Model of Color Vision for Remote Sensing 1
It is better to use the Level-2 product of the Landsat which has already been pre-processed, i.e, surface reflectance (SR). According to USGS, SR products from different Lansat sensors are similar. Again, studies such as Li et al. (2013) and Flood (2014) reveal a standard deviation smaller than one of the time-series of the spectral curves for the different Landsat sensors, therefore, there are no significant effects on NDVI calculation or classification results when data from multiple sensors of Landsat are combined.
- Badges
- Science topic
- Similar topics
- Engineering
- Modeling