Cropland: Sudan Gedaref, West/South Darfur, Al Jazirah, Central Darfur - Feb 2022

[ivanzvonkov]

• Labeling projects created and links added below
• Set 1 Labeling: Abena, Adebowale, Gedeon
• Set 2 Labeling: Manthan, Bhanu, Alana, Diana, Hannah, everyone available
• Data added to repository
• Data added to intercomparison
• Model trained
• Map made
• Expert check

[ivanzvonkov]

@Gedeon-m-gedus created the sample used for the CEO projects. The sample contains 1196 points and is stratified by geographic boundaries within the ROI.
Looking at the GLAD mask this ROI seems to have a high proportion of crop and therefore I think this sample will be sufficient for a precise accuracy estimate (low standard error for crop UA, PA). In other words, I think it's not necessary to stratify using the cropland map for this sample.
@hannah-rae let us know what you think.

[hannah-rae]

@ivanzvonkov I think that will be the case too, so we can use the same sample for validation/accuracy assessment that we do for area estimation.

That said, if we want to compute change areas directly after making a second (pre-conflict) map, we would need to get a change strata sample. Potentially, if we use the same samples that are used here for 2022 for 2023, then we could use that for the change samples (but likely won't have good representation from change strata).

[ivanzvonkov]

Subset of data already labeled added to repository: #356
Initial model trained: #359

[Gedeon-m-gedus]

Addressing sample imbalance

Using dynamic-world cropmask, I have regenerated other sample points to be labeled.
Basically, under the assumption that we will find more positive samples (crop) from the crop category of the dynamic world map.