I am trying to resample the MODIS (aqua) image from 1000m resolution to LandSat image resolution which is of 30m for applying ESTARTFM technique. So you need to do preprocessing (Resampling and Reprojection) before applying the algorithm. But when i read here it is suggested not to go from finer to coarser resolution. So, my first question why it is not recommended secondly how to achieve the same using QGIS 3.6. Also, I tried resampling MODIS image using pyMODIS using the following script

modis_convert.py -s "( 1 )" -o outputmodis.tiff -e 32630 -g 30 MYD11A1.A2019133.h17v03.006.2019150001731.hdf 

So, my final TIFF size increased from 2 MB to 3.4 GB. Is there something wrong I am doing here? Though When I resampled it to 500 m size increased from 2MB to 10 MB.

  • 2
    You increase the resolution thirtyfold and wonder why your filesize goes through the roof? You have much to learn, young padawan. Also, I doubt that increasing the resolution is that helpful, since you'll still end up with a washed out version of the 1x1 km data. – Erik Sep 16 at 8:08

First, your file sizes are as expected. When you go from 1000m per pixel, to 30m per pixel, you get 1111-times increase in size, which is close to what you see from 2 MB to 3.4 GB, once you also consider compression etc. The calculation is:

(1000 m/ 30m)^2 = 1111.1

It is squared due to the raster being a 2d array.

The whole point of STARFM, or ESTARTFM (a derived model), is to apply data fusion to get daily 30m data from MODIS. This means that you will get a much larger file size. The other thread that you linked to is not related to data fusion, so there, oversampling the data is not recommended, but for data fusion, such as STARFM, it is essentially the whole point.

  • Thanks Mikkel. I have one more question Is it possible to do the resampling of HDF using QGIS 3.6 or HDF-EOS To GeoTIFF Conversion Tool (HEG)? I tried but couldn't find any option – Naresh Sep 16 at 12:44
  • In QGIS, you can use 'Raster - Warp', and set the 'Target Resolution'. – Mikkel Lydholm Rasmussen Sep 16 at 12:47

If you have gdal, you can try gdal_translate -outsize 3333.33% 3333.33% -r bilinear input.tif output.tif

  • 1
    For data-fusion purposes, it should likely be '-r near'. – Mikkel Lydholm Rasmussen Sep 16 at 13:52
  • Is there any problem if i use cubic convolution. As i read it gives the smooth edges in the final output after fusion – Naresh Sep 16 at 14:13

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.