I have a raster with resolution 0.05x0.05 deg. I want to extract the values of the raster to a point grid with resolution roughly 14 km (mean latitude is 47 deg), which I have averaged to 0.1575 deg. My concern is that each point represents the whole 14x14 km cell, but extracting values from a finer resolution might not be representative.

Is it advisable to resample the raster to a coarser resolution before extracting the values - even though the data won't fit perfectly anyway due to a different projection - or would it bring even more inaccuracy?

EDIT: my working domain is within 35N 5E (SW corner) to 60N 37E (NE corner). The raster data is in WGS-84 GCS, the point grid has a custom Lambert Conformal Conic projection. I need to assign land cover characteristics to each grid point - I should note that this data is not discrete, it is stored as a multiband raster with percentage values for each band.

EDIT2: Ok, I just noticed that the tool which I am using for extraction (Extract Multi Values to Points) has the option of Bilinear interpolation of values at point locations. Would it possibly be better to use this option instead of interpolating the raster in advance?

  • What type of analysis are you doing? How large a geographic area are you working with?
    – Aaron
    Aug 23, 2016 at 18:28
  • 1
    What type of value do you have ? Quantitative (e.g. temperature) or qualitative (land cover) ? Will you handle your extracted values as points or as surfaces ?
    – radouxju
    Aug 24, 2016 at 7:06
  • @radouxju As I stated in the edit, I would say that it is a quantitative representation of qualitative data. I'm not sure what exactly you mean by the second part of your comment, though. I will be passing the gridded data to a model and the extracted values will represent the 14x14 km surface, not just the exact point at which the values were extracted.
    – Janina
    Aug 24, 2016 at 7:29

1 Answer 1


Weither or not you aggregate strongly depends on the type of measurement that is stored in your pixel. There are two cases :

-point : the value is measured on a regular grid at the center of each pixel. This is what you obtain by systematic sampling using a local measuring instrument (e.g. a thermometer placed every km) -area : the value represents the whole cell. This is the most common. Typically you get this by remote sensing when all he surface of the cell (and sometimes more) contribute to the estimated reflectance.

In your case, the proportions are obviously valids for the area of the pixel. Therefore my first idea would be to aggregate when comparing with data of a coarser resolution. However, if one of the data that you used at coarser resolution in your model is of "point" type, then it could better not to aggregate (especially if the "point" data has a poor spatial autocorrelation).

This is not an absolute rule, but the best way to aggregate quantitative values is very often the average. However, this will work best with normally distributed variable and should be avoided in some case (e.g. DEM aspect).

In practice, you can achieve it

1) by upscaling the grid data using an averaging rule the converting grid centers to points.

2) applying a local average filter on the high resolution data then extracting the values at pixel location.

This might yield slightly different results because of the actual neighbourhood being used (aware of the projection in (1), constant shape in (2)) and the resolution is different, but conceptually they are the same.

You can then extract the grid value at point location, preferably with a bilinear interpolation.

  • Thanks, but rather than how to aggregate the data I needed to know IF it is neccessary or even a good idea.
    – Janina
    Aug 24, 2016 at 9:03
  • I've edited my answer. Considering the information that you provide (proportions of land cover), I would says "yes, it is better to aggregate", but there is no universal answer to this question without knowing the type of variable.
    – radouxju
    Aug 24, 2016 at 10:20

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.