Logging roads on Malaita Island
Last updated
Was this helpful?
Last updated
Was this helpful?
Malaita, the most populous island in the Solomon Islands, is currently facing severe threats to its critical ecosystems due to unregulated logging, driven by foreign demand and population growth. To protect its rich biodiversity and fragile environment, we want to use Earth Index to identify and monitor logging roads on Malaita. This will help us assess the extent of the harmful logging activities and inform necessary action.
Let’s start by brainstorming key features to label to help Earth Index identify what I am looking for:
Next, I find an example of a logging road and add in positive labels.
I can continue this process with other roads to input more verified labels, or I can try to run predictions to see how Earth Index will do with this amount of data.
By generating a high number of predictions, it is more about its accuracy in aggregate vs. an individual square. Where predictions are clustered, it shows an area that would be worthwhile to investigate. Using the “Config” button, however, allows you to adjust the number of predictions that are generated (1-500). Adjusting the slide bar from low to high confidence can also customize the predictions displayed.
At quick glance, when I look at my results, there are too many non-road objects identified. I will label a few more positive examples and update the predictions.
This time, I spot check some areas with a large number of clustered predictions. Sure enough, the results have improved and other logging roads are identified:
I also find areas labeled that I am not interested in - which I change from predictions to negative labels. As villages, they show the clearing/deforestation that is common with logging roads, which is why they were predicted as something I might be interested in, but it is ultimately not what I am looking for.
As I look closer, I see some dirt roads that are not logging roads were identified. This is where I need to “teach” Earth Index how to differentiate between them. This will involve a bit of trial and error, but I know I want to focus on the color differences (logging roads are reddish) and the deforestation.
Another tool for validating the data is the “Quick Validation” button. Using the “filter by” drop down menu is helpful in targeting the type of data you want to work with to improve the search. High Confidence data will be quicker in multiple positive labels, the reverse for low confidence. Medium allows you to add a human touch to help the predictions with more ambiguous labels. All have their use depending on what you are seeing the tool do.
As you edit your data, use the update features to continue to refine the labels.
When are you done? It’s up to you and the goals of the project you are working on. When you have data that you feel will be helpful, click on the three gray dots to expand to “Export GeoJSON” data.
Watch this video if you're curious to learn more!
While the change is small, after editing several predictions that identified villages and hitting the data points continue to refine, with outliers disappearing and clusters refining over the logging roads.
