I need to convert raster data to point data so I can use the points for interpolation via kriging. My problem is that my resolution is 5 ft and I'm working on a whole city. The data I am converting to points is grouped by zone with No Data values between; most/all of the points in each zone have the same value. I need the high resolution because some zones are linear, just 2-3 cells wide and about 20 cells long, while others are huge behemoths. I need to capture the smaller zones but I don't need 2000 points in the larger zones; because the values are all the same, I might need 10 or 20. I experimented with lowering the resolution to 50 and 100 ft but I lose much of the smaller zones. My raster is floating data so converting to polygons to do any other vector stuff isn't possible, I don't think.

Am I doomed to have a billion points and manually remove ones from the big data chunks?

What kind of processing time am I looking at if I do 5 ft Raster to Point and have one point per cell and I interpolate those? It already took several hours to interpolate between 10 points via kriging over the whole area.

I would not like to sacrifice my data resolution for the small zones if I can help it. They are very important and represent the variability I need to capture.

  • Have you thought about using a TIN? Commented Mar 2, 2015 at 17:26
  • what about a raster calculation to remove cells you are not interested in. Alternatively, you could convert your rasters to polygons, and use the polygons as a means to select/delete points (ie: no data, or other land use). Commented Mar 2, 2015 at 17:38
  • Why do you think that polygonize does not work with float data?
    – AndreJ
    Commented Mar 2, 2015 at 18:43
  • @AndreJ - I'm guessing the number of decimal places is such that you'd never get two adjacent cells with exactly the same value. Commented Mar 2, 2015 at 19:40
  • I certainly wouldn't lower the resolution. If you have 5ft cells and want to interpolate values between them, starting out by resampling to 50ft seems totally illogical. Your data is going in the wrong direction, enough to make kriging it back again irrelevant I would think. You may as well just stay in raster-land and interpolate with bicubic resampling. Commented Mar 2, 2015 at 19:57

2 Answers 2


Using GDAL & Python, you could split the city zones into separate rasters, and then convert to comma-separated xyz points using gdal2xyz.py. The skip option allows you to control your sampling frequency (i.e. how many raster cells are converted to points). For example, a skip factor of 3 means that you are converting one in every three pixels:

gdal2xyz.py -skip 3 -csv InputRasterName OutCSVName

You could then use pandas to remove the nodata points:

import pandas

InputCSV = 'xyz.csv'
OutputCSV = 'xyz_v2.csv'

# Read CSV
Data = pandas.read_csv(InputCSV, sep=',', header=None, index_col=None)

# Remove no data (e.g. zero) points in your z values 
NewData = Data[Data[2] != 0] 

# Export sorted data to CSV
NewData.to_csv(OutputCSV, header=['x','y','z'], index=False)

You can then import the CSV into QGIS or ArcMap and convert the data to an ESRI shapefile or any other vector format.

Something similar can be achieved in R by using the raster package:

# Open raster
r <- raster("ImageName")

# Convert raster cells > 0 to points
points <- rasterToPoints(r, fun=function(r){r>0}, spatial=FALSE)

# another example, raster cells = 1 to points
points1 <- rasterToPoints(r, fun=function(r){r==1}, spatial=FALSE)

# another example, raster cells = 2 to points
points2 <- rasterToPoints(r, fun=function(r){r==2}, spatial=FALSE)

# take a random sample of 100 points/pixels
sample <- points[sample(1:nrow(points), 100, replace=FALSE),]
sample1 <- points1[sample(1:nrow(points1), 100, replace=FALSE),]
sample2 <- points2[sample(1:nrow(points2), 100, replace=FALSE),]

# Export points to csv
write.table(sample, file="Points.csv", row.names=FALSE, na="", col.names=TRUE, sep=",")
write.table(sample1, file="Points_1.csv", row.names=FALSE, na="", col.names=TRUE, sep=",")
write.table(sample2, file="Points_2.csv", row.names=FALSE, na="", col.names=TRUE, sep=",")

Using stratified random point sampling will allow you to say "10 points in this class", eg in QGIS. You might need to convert to polygons first, I am not sure.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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