Improve Matching Variable: Agricultural Soil Suitability

[Jo-Schie]

Background: Soil aptitude for agriculture is dependend on different factors, not solely clay as we have it in our model here. Textual factors comprise clay, silt and sand particles and an optimal mixture can be found e.g. here. Other variables are e.g. cation exchange capacity (CEC), total nitrogen as well as soil organic carbon density and soil organic carbon stock.

There are two approaches to create a better indicator:

• create a better estimation based on the soilgrid data that the mapme.biodiversity package is able to process. This would require: Defining an adequate model, reprocessing the data and implementing a routine. It would also require working already on the raster level (i.e. changes to the package) because it does not make too much sense to do that within an AOI where mean values already distort the measures.
• search for another dataset that already shows agricultural suitability. This might be conceptually easier but the routine is not implemented in the package so I would be not so easly reproducible and also technically more work.

As for the time being, I will only implement a routine that defines soil aptitude based on clay content within certain boundaries (i.e. from 10-40% as optimal to fair aptitude and beyond or below as low aptitutde.

Definition of Done. A new indicator is defined and conceptually sound. Data is processed and integrated into the matching frames.

[Jo-Schie]

fyi: @yotaae and @melvinhlwong . I think this issue will be more complicated. So as for the time beeing I will try to implement a better estimation based on the available data and later depending on ressources we should change this. I will also create a new issue in mapme.biodiversity for this.

Image Source

[Jo-Schie]

Probably it is easier and more sound to user another data-product with an estabilshed methodology see this issue in the biodiversity repo.

[fBedecarrats]

This is an old reference(see Soil quality section) : https://www.fao.org/soils-portal/data-hub/soil-maps-and-databases/harmonized-world-soil-database-v12/en/ But from what I've found, soilgrids 2.0 seems to be the state of the art right now (and I understand that it is not very satisfying). Just an idea: why not train a model to predict which soilgrids characteristics' combinations are the best predictors for deforestation?