nasaharvest / crop-mask

End-to-end workflow for generating high resolution cropland maps
Apache License 2.0
94 stars 26 forks source link

Sudan South regions first intercomparison + template #360

Closed ivanzvonkov closed 5 months ago

ivanzvonkov commented 6 months ago
image

Only uses test points so it is comparable to the LSTM metrics (below) however this results in very large standard errors.

"test_metrics": {
            "accuracy": 0.8429,
            "f1_score": 0.4762,
            "precision_score": 0.4545,
            "recall_score": 0.5,
            "roc_auc_score": 0.8113
        },
review-notebook-app[bot] commented 6 months ago

Check out this pull request on  ReviewNB

See visual diffs & provide feedback on Jupyter Notebooks.


Powered by ReviewNB

ivanzvonkov commented 6 months ago

@Gedeon-m-gedus I also created instructions for generating this notebook in the wiki: https://github.com/nasaharvest/crop-mask/wiki/Intercomparison-Accuracy-Assessment#4-push-new-intercomparison-notebook-to-github

@hannah-rae the GEE script at the bottom of the notebook currently doesn't work because the admin zone is not properly specified, could you share the admin zone code from that Sudan GEE app?

Also, geemap.ee_to_gdf(extracted) failed with internal error for worldcover-v200 so I skipped it for simplicity for now.

Gedeon-m-gedus commented 6 months ago

@Gedeon-m-gedus I also created instructions for generating this notebook in the wiki: https://github.com/nasaharvest/crop-mask/wiki/Intercomparison-Accuracy-Assessment#4-push-new-intercomparison-notebook-to-github

@hannah-rae the GEE script at the bottom of the notebook currently doesn't work because the admin zone is not properly specified, could you share the admin zone code from that Sudan GEE app?

Also, geemap.ee_to_gdf(extracted) failed with internal error for worldcover-v200 so I skipped it for simplicity for now.

@ivanzvonkov the crop proportion seems to be low, I was discussing with Hannah about how we could increase positive labels (we can discuss this tomorrow during our scheduled call).

Regarding the RoIs boundary, I think they are fetched from FAO

var rois = ee.FeatureCollection("FAO/GAUL/2015/level1")
            .filter(ee.Filter.inList('ADM1_NAME', 
                                     ['Gadaref', 
                                      'Western Darfur',
                                      'Southern Darfur',
                                      'Al Jazeera']));
ivanzvonkov commented 5 months ago

Thanks @Gedeon-m-gedus I've updated the notebook so that the GEE script includes that ROI

Gedeon-m-gedus commented 5 months ago

@ivanzvonkov I realized that the FAO boundaries are wrong! here is the correct boundaries: https://code.earthengine.google.com/?asset=projects/ee-gmuhawenayo/assets/rois

I created a PR for it https://github.com/nasaharvest/crop-mask/pull/363