Using SkyTruth Alerts, we created a workflow for generating AOI-specific change detection using DW. While results will never match detections from a machine learning model that was built specifically for a given feature or condition, this workflow can often produce a first-line of detections that warrant further viewing.

This is version 1 of the workflow. We hope to add features and further explore DW capabilities in future versions of this workflow.

1. Open up alerts.skytruth.org in a browser. If you haven’t already, register for an account and login.
2. Once logged in, you can disable viewing NRC alert markers by unchecking that box. This will speed the process up for exploring.
3. Move the map to an area where you already know changes have occurred. This is your playground for testing various DW bands and thresholds. You can do this by moving the map directly to your Area of Interest (AOI). You can also outline your AOI or select from a pre-defined AOI. Here’s some help in getting you started with your own AOIs. Note that if it’s an AOI you’ll come back to again, you should save it to your list of AOIs.
4. Click on Layers in the left sidebar. Layers are separated by category and the first category is for DW layers. If you don’t see a list under Dynamic World, click it to see what’s available.
5. Select the DW: Explore change detection parameters layer.
6. Enter before and after date ranges. Again, the idea is to use date ranges that match when you already know there have been changes. That will help you play with the DW settings to match those known changes.
7. Alerts serves up Planet Basemaps–composite images for a month based on best pixels from daily images. By selecting the Compare using Planet basemaps checkbox, you’ll automatically bring up before and after basemaps in a comparison slider. This can be very helpful in evaluating results.
8. For Land Classification, start with All Land Classifications. A good starting value for Percent of Change is 25%, the default.
9. Click Run to view results.
10. By viewing Planet basemaps in the comparison, you can use the center slider to move from the left to right image. You can also adjust the layer’s Opacity to make it easier to see the underlying basemap. This is how you can identify false positives–detections that were picked up but should not have been–and false negatives–detections that were not picked up but should have been.
11. Click on the DW: Pixel Stats layer, then adjust the date range to cover the range of dates you entered in step 6. Pixel Stats work by clicking directly on the map (if you’re comparing maps, click on the right map). You’ll get to see exactly what DW is seeing and why detections are false positives or false negatives.Here’s an example of clicking a potential false positive pixel on the map:
12. Using what you discovered in step 11, there are three options available:
    1. If you see consistent changes in probabilities for a single band, try selecting that band in the Land Classification drop down menu. Once that’s done, the Percent of Change slider can also be move up or down.
       
    2. Is there a band with probabilities that can be compared to a certain percent to eliminate false positives or false negatives? That’s what the optional 2nd Land Classification is for. You can select either “Is greater than or equal to” or “Is less than or equal to” from the dropdown menu, then adjust the percentage you’re comparing to.
    3. Is there a band that should be filtered out because it has large seasonal changes in probabilities? That’s what the optional 3rd Land Classification is for. Select the band and adjust the  Saw a percentage change of. slider for the percentage in change you want to ignore. For example, consider how the Snow and Ice or Crop bands may vary.
13. Click Run again to view new results.
14. Repeat steps 10 through 13 until you see satisfactory results.

Future versions will allow you to save results you come up with for easy re-running on different begin and end date ranges.