How could I quantify areas of soil without vegetation cover?

[erichnagy]

Hello everyone.

I am working on a project where we want to quantify the area of bare soil or with little vegetation cover in meadows and pastures used in livestock farming. This as a basis to later be able to compare with the past if it improved or worsened. It is also the idea of improving forage production in these areas and avoiding further degradation. These bare areas are still unclear as to what caused them.

The area that I am going to monitor is approximately 500,000 hectares. Located in country: Paraguay around the city of Filadelfia (https://maps.app.goo.gl/dhX3zR2wFbdo57o18).

My question is what would be the best way to do it?

I attach a Google Earth image of what we are looking to measure.

What did I already try:

1) Perform automatic classifications with sentinel images of the various areas.

2) With sentinel images (10x10 meters) calculate the "modified Soil Adjusted Vegetation Index" (MSAVI) but as expected the resolution area is very large.

What do you think of this? What other options do you suggest?

Some cons are:

There aren't many satellite sensors to choose from.

[MarkDann]

Interesting use case!  I'm curious, have you given the Segment Anything Models a try for this?  I'm curious if the resolution would work to capture the soil condition and the scale you're looking at?  Here is the link to the SAM tools if you haven't seen them before...  

https://www.arcgis.com/home/item.html?id=9b67b441f29f4ce6810979f5f0667ebe

[DavidWright]

Yes this looks like an interesting project. I think using Classify Pixels Using Deep Learning would be a good approach. You would need to train the model with base soil samples. It should work well based on results I’ve observed in other applications of this tool in ArcGIS.

As you already pointed out you'll likely want to use some higher resolution imagery. Check into what's available from crewed or uncrewed aerial sources, or perhaps satellite imagery from Airbus. Maxar or Planet.

[erichnagy]

Thanks for the answers @DavidWright  & @MarkDann . I will try it with the mentioned methodologies. The main impediment is the low resolution of the images I have. I'll check what options I can get. Thank you