I downloaded TM (2006-2011) and OLI (2013-2016) Surface Reflectance Higher-Level Data Products (atmospheric corrected) based on L1TP Landsat images (radiometrically calibrated, orthorectified). And I got problems wtih preparing data for time series change analysis. Thereafter in this analysis I'll want to detect changes in area damaged by bark beetle as only supporting information for our project. All images are taken from end of July to early September with respect to ecology and behaviour of given species. Study area is mountainous region in Georgia.
According to USGS documentation (https://landsat.usgs.gov/landsat-collections) :
"Tier 1 includes Level-1 Precision and Terrain (L1TP) corrected data that have well-characterized radiometry and are inter-calibrated across the different Landsat instruments."
I assume, that key pre-processing steps are done with this data and now I have absolute surface reflectance data. So if I create time series of some vegatation index (e.g. NDVI) from images from one sensor type (taking into account, that values may vary across sensors) and pick some time consistent pixel (e.g. impervious built-up surface), the values will be more or less similar, or vary sligthly. But it looks like, that it does not work this way.
I processed the images with R packages 'raster' and 'RSToolbox'
1. Batch clip all data to area of intertest.
2. Create individual raster stacks for each image
3. Topographic correction with topCor() (srtm data + image metadata)
4. Claud masking with BQA band
5. Calculate NDVI indices
6. Create raster stack with NDVI outputs
7. Checking NDVI values of given pixel in time-series stack (see results in attached files, first example is from build-up area which does not change over time)
As you can see the pixel value changes drastically across time within TM (2006-2011) and OLI (2013-2016) separately.
Although it may not make much sense for me, I also try matching images by pifMatch() or histMatch() and I also try the process without cloud masking and topographic correction steps, all without visible improvement.
My question is:
Can I suppose that Surface Reflectance Higher-Level L1TP is ready for time-series analysis in way I did it, than my processing steps are correct or I am missing something crucial?
I am new to remote sensing, so I will be grateful if someone points to what I am doing wrong, or for giving me any suggestions. I am comfortable with QGIS, R, Python, GRASS or bash/GDAL.
My pre-processing approach was based on this paper (and few case studies)
A survival guide to Landsat preprocessing (Young et al. 2017)
https://landsat.usgs.gov/landsat-collections
Article A survival guide to Landsat preprocessing
Seems reasonably consistent .
a few questions
Are you presenting NDVI values for a single pixel? What about taking mean values for buffer instead?
Did you evaluate how landcover changed for the vicinity of that pixel?
Did you evaluete rainfall pattern for that period?
As mentioned before the rainfall pattern, rainfall amount and intensity at the time and days before the time of the image aquisition play an important role, as well as the specific lithologic setting and the soil properties influencing soil moisture conditions and, thus, vegetation growth conditions.
Thank you very much for your comments.
What I forgot to mention (and it may be important) is that the area is mountainous region in Georgia. And in one my example are values exactly from build-up area which does not change over time.
@Paulo Eduardo Cardoso
"Are you presenting NDVI values for a single pixel?"
Yes, in my examples the NDVI representig data only from one pixel, but I tried the mean values in buffers before (that was my original approach), the result was similar.
"Did you evaluate how landcover changed for the vicinity of that pixel?"
Sorry I don't understand this very well, visually (RGB) the example pixels and their surroundings doesn't changed, but I am not sure how to test it since the images looks uncoparable for me.
@Paulo Eduardo Cardoso, @Barbara Theilen-Willige
"Did you evaluate rainfall pattern for that period?"
Maybe this is the thing, I didn't read any study that mention this yet, so I omit this at all. Thank you for clue, I am going to do some more research about it. But can this affect also the build-up areas/non-vegetation areas which does not change over time?
My pre-processing approach was based on thise paper (and few case studies):
A survival guide to Landsat preprocessing (Young et al. 2017)
http://onlinelibrary.wiley.com/doi/10.1002/ecy.1730/full
Article A survival guide to Landsat preprocessing
Landsat L1TP images are orthorectified with ground control points and they have the highest quality suitable for pixel-level time series analysis. As you've downloaded surface reflectance high level L1TP images you can use these images without atmospheric and topographic preprocessing.
In addition to surface reflectance Images you can download NDVI calculated using surface reflectance data from ESPA website. You can use them directly in your studies or you can download 1 or 2 NDVI images to check the compare the values regarding to your NDVI values.
You have mentioned that this 2 pixel represent a non-changing area. But do you say this based on your ground truth? Otherwise these pixels may present a non-evergreen area or you may be seeing the effects of a dry season.
https://espa.cr.usgs.gov
- Badges
- Science topic
- Similar topics
- Geoscience
- Remote Sensing