1

My data

I downloaded climate data from the WorldClim website, provided as rasters at a global scale using WGS 84 and various resolutions. I'm using the 15min resolution and I focus on Mediterranean countries (North Africa, Southern Europe and Western Middle-East).

My aim

I want to analyze climate differences between polygons but for several reasons I don't extract climate data for each polygon (using the extract function from the raster R package or QGIS Zonal Statistics). Instead, I rasterized those polygons and I compare raster cells from one polygon with raster cells from another polygon.

My problem

Raster cells don't have the same area, which means that small raster cells have more weight than they should, which could distort my analysis.

How could I take account of this difference in raster cells area?

Can I use another projection? I've read here that reprojecting rasters might not be a good solution. Also I'm not sure that I can give different weights to those cells in the statistical analysis I use to compare cells.

Sorry if this topic had already been discussed, I did not find any solution on this website.


EDIT: I am seriously considering reprojecting my data using an equal-area projection like Lambert Azimuthal Equal-Area projection centered on my study area. Is this solution such a bad idea?

1 Answer 1

0

Using an equal-area projection will help you equalize the weight of different raster cells since the cells will have... well, equal area. However, doing so will necessarily introduce potentially much larger errors from projection distortion and raster interpolation.

The acceptability of the projection distortion depends on the size of your study area. For your case, LAEA with a well-chosen centre of projection will probably be OK; you might achieve less distortion by using a conic equal-area (Albers) projection since the area is elongated in a west-east direction.

Also, unlike vector data, most spatial transformations of raster data are lossy because values must be interpolated to the new cell locations, and you must choose the interpolation method with care. For WorldClim data (which are numeric and rather continuous - I would even suspect that they already are partially interpolated, which means the information loss due to projection will not be as severe), I would try examining bilinear, cubic or Lanczos interpolators.

Having said that, if you are concerned about high precision, I would also consider polygonizing the raster and working with vector data - it will be more computationally expensive but lossless.

1
  • I'll try Albers projection and I'll compare it with LAEA projection, thank you! Jun 2, 2020 at 9:13

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.