1

I am using the Near infrared and red bands from satellite images to research soil quality in southern Africa. The study site is about 20 km wide and 20 km long so almost square shape, and I am using 5 m resolution satellite imagery.

What is the best way to sample different locations so that I get enough samples and also to minimize spatial autocorrelation?

Is grid sampling method the best?

I am new to studying soil.

|improve this question
  • Should your samples be single pixels or a group of pixels (what shape)? – Michael Jan 26 at 8:14
  • Try to apply unsupervised classification to your image. That can help to define how many discernible classes you actually have according to the spectral variability. – Vadym Jan 26 at 8:28
  • What are you specifically trying to predict? Do you need to do the sampling in the field or can you expertly select classes on screen? – Aaron♦ Jan 27 at 23:09
1

First of all, I would remove all locations where there is some vegetation, because NIR and Red are strongly linked with the vegetation, so they cannot be used for direct soil characterisation if there is a vegetation cover. For a quick and dirty first "non vegetation" map, you could use NDVI < 0.1.

Then, it depends on how much "expert knowledge" you have on the area and what you want to do (estimate the area of each soil type, validate a map, characterize the different soil types, prospect in search for a given soil type...). If you have no idea at all about what you are looking for, simple random sampling is the most robust. The number of sample then depends on the expected accuracy of your map. For a precise validation, you will often need about 700 points. The paper from Olofsson et al is a good summary.

|improve this answer
  • Thanks for the suggestions @radouxju. My study area is quite sparse and not a lot of vegetation. I'll take a look at the Olofsson paper. – J. P. Jan 30 at 8:05

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.