Averaging overlapping rasters with gdal

Question

I am working on creating a class which will merge several georeferenced rasters into one using different strategies, essentially taking average, max, min where the images are overlapping.

So far I've tried using gdalwarp with --resample parameter set to average.

gdalwarp -srcnodata 0 -r average a.tif b.tif output.tif

But gdalwarp just overlaps the images. I've tried other approaches with gdal_merge.py and gdalbuildvrt but they also simply overlap images, without taking average.

Reading gdal dev list I've seen people taking following approach:

reproject images to same dimensions, filling the rest with no data value
filling no-data values with zeroes
using gdal-calc to take max or average on images

I wanted to try this approach but stumbled on a problem of changing dimensions of image with adding no-data value, i. e. the following command changed the whole image, instead of just inserting extra no-data pixels.

gdalwarp -ts 1591 1859 a.tif r1.tif

So my question are:

Is there any other approach on how this averaging could be done?
Is there any utility, preferably with gdal, that could change dimension of the image by adding no-data value pixels to it?

Note: you can find sample files here https://drive.google.com/drive/folders/1cm8Y4WX03wn4XrNKOifYBhd13GqVNGdb?usp=sharing

Владислав Мокроусов · Accepted Answer

The following approach worked pretty well. First I build virtual raster. gdalbuildvrt raster.vrt -srcnodata 0 -input_file_list paths.txt paths.txt is file with following content: a.tif b.tif Then I add a pixel function to it, as showed here https://lists.osgeo.org/pipermail/gdal-dev/2016-September/045134.html. Pixel function is written using numpy, basically it sums all images and divides each pixel by the number of overlapping images for that particular pixel. Raster before adding pixel function. GEOGCS["WGS 84",DATUM["WGS_1984",SPHEROID["WGS 84",6378137,298.257223563,AUTHORITY["EPSG","7030"]],AUTHORITY["EPSG","6326"]],PRIMEM["Greenwich",0],UNIT["degree",0.0174532925199433],AUTHORITY["EPSG","4326"]] -3.0531428271702840e+01, 3.7890083929483308e-02, 0.0000000000000000e+00, 6.7079735828607269e+01, 0.0000000000000000e+00, -3.7890083929483308e-02 0 Gray a.tif 1 0 b.tif 1 0 Raster after adding pixel function. GEOGCS["WGS 84",DATUM["WGS_1984",SPHEROID["WGS 84",6378137,298.257223563,AUTHORITY["EPSG","7030"]],AUTHORITY["EPSG","6326"]],PRIMEM["Greenwich",0],UNIT["degree",0.0174532925199433],AUTHORITY["EPSG","4326"]] -3.0531428271702840e+01, 3.7890083929483308e-02, 0.0000000000000000e+00, 6.7079735828607269e+01, 0.0000000000000000e+00, -3.7890083929483308e-02 average Python 0 Gray a.tif 1 0 b.tif 1 0 And finally, transform it to raster using gdal_translate and gdal python option set to 'yes': gdal_translate --config GDAL_VRT_ENABLE_PYTHON YES raster.vrt raster.tif A result image for this example.

aparrar · Answer

Thank you very much for posting your workflow, this helped me with a similar issue I was having. In case this might be useful to somebody else, I used different python functions for my raster mosaic. In my case, the no data value for the VRT was 255 and because my data only goes from 0 to 100, I masked all the values in my VRTs greater than 100 before calculating min, max, or mean values, and then reset the value of the masked pixels to 255.

VRT mean function

import numpy as np

def average(in_ar, out_ar, xoff, yoff, xsize, ysize, raster_xsize,raster_ysize, buf_radius, gt, **kwargs):
    x = np.ma.masked_greater(in_ar, 100)
    np.mean(x, axis = 0,out = out_ar, dtype = 'uint8')
    mask = np.all(x.mask,axis = 0)
    out_ar[mask]=255

VRT max function

import numpy as np

def average(in_ar, out_ar, xoff, yoff, xsize, ysize, raster_xsize,raster_ysize, buf_radius, gt, **kwargs):
    x = np.ma.masked_greater(in_ar, 100)
    out_ar[:] = np.ma.max(x, axis = 0, fill_value=0)
    mask = np.all(x.mask,axis = 0)
    out_ar[mask]=255

VRT min function

import numpy as np

def average(in_ar, out_ar, xoff, yoff, xsize, ysize, raster_xsize,raster_ysize, buf_radius, gt, **kwargs):
    x = np.ma.masked_greater(in_ar, 100)
    out_ar[:] = np.ma.min(x, axis = 0, fill_value=100)
    mask = np.all(x.mask,axis = 0)
    out_ar[mask]=255

nraj · Answer

The answer above is very helpful.
Just so that others coming here don't fall into this trap, I spent a lot of time trying to figure this out, in the end realizing that I was missing: subelements for VRTRasterBand (whose subclass specification must be set to VRTDerivedRasterBand). Setting the VRTDerivedRasterBand subclass made everything work!
It is 4.30 AM but it was worth it!

Averaging overlapping rasters with gdal

3 Answers

Add your own answers!

Ask a Question