Binary Mask Creation
In order to run the MSPA, we must first create our binary forest/non-forest mask. Forest corresponds to the foreground of the image, the target of interest. Non-forest corresponds to the background of the image.
- Binary mask creation: Example GEE Link
1. Select one Planet NICFI monthly mosaic from the dry period of the current year (e.g. Brazilian Amazon: July through September).
2. Collect training polygons/points for both forest and non forest classes.
3. Extract spectral values from mosaic pixels at polygon/point locations.
4. Apply a Random Forest classifier
- Specific parameters: 50 trees
5. Verify the results by visually analzying the mask against the imagery used for the classification.
- It is important to identify that holes within the canopy and cleared pathways are classified correctly. Different tree types and heights can have different spectral values and become missclassified, oftentimes being mistaken as gaps within the forest cover. If these areas are being misclassified, we collect more training polygons/points within these areas.
6. Apply a kernel smoothing function to remedy unnecessary speckling and noise.
7. Reclassify the forest mask.
- It is essential within the GTB that the background is valued as 1, the foreground is valued as 2, and any missing values are valued as 0. If these are not correct, we reclassify the image values accordingly.