I've got this doubt on how to best proceed with the following situation: I would like to compare two datasets, one (model_data) is a raster (NetCDF format), with lat, lon and time as coordinates, being the spatial resolution of 0.5 degrees. The other dataset (observed_data) is actually a table that is geolocated, and I transformed it into a vector using a shapefile as reference (which is formed by multiple polygon representing counties).

My approach so far has been to rasterize the vectorized dataset at a very high resolution (0.0083 degrees), and then upscale it to the same 0.5 degrees found in the model_data. Therefore, it follows: vector -> raster (0.0083) -> raster (0.5). to upscale I have tried different techniques, from aggregate + mean to specific CDO techniques such as bilinear interpolation and first order conservative mapping.

However, I'm not entirely happy with the results so far, and I wonder if there is any flaws with this approach. I thought of perhaps trying the other way around, and converting the model_data into polygons or table and then doing the analysis at the county level.

Question: Are there any flaws in the approach described above and is there any alternative solution to the problem?

OBS: I am okay to use Python, R, and specific NetCDF tools such as CDO for this.

  • Welcome to GIS SE. As a new user, please take the Tour, which emphasizes the importance of asking One question per Question. We use a Focused question/Best answer model, so opinion-based/discussion-oriented questions are likely to be closed.
    – Vince
    Jul 26 '21 at 14:35
  • Thanks! I edited making the question more focused and objective.
    – Henrique
    Jul 26 '21 at 14:41
  • Please decide, which software you actually want to work with. It would also be helpful if you provided a screenshot of your polygons vs. raster, because right now it is unclear, whether you have one smaller polygon per raster cell, or a lot of smaller polygons.
    – Erik
    Jul 27 '21 at 11:18
  • What are the sizes of the datasets? If you have relatively few observation stations, it might be best to interpolate the model raster to the observations and compare with observations. Conceptually, it is probably easier to verify the model against the observations, and characterize the model error and bias at obs points, and then use that to extrapolate the data to the unobserved places.
    – Dave X
    Jul 30 '21 at 22:56

Your Answer

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

Browse other questions tagged or ask your own question.