Clouds and Shadow detection is one of the most important tasks in analytical remote sensing solutions.
If the whole satellite image is 90% clean, there is still a 10% chance that your field will be under clouds/shadows area. The reverse is also true – many systems do not process images with clouds above 70% – although the remaining 30% can help the agronomist or farmer to make the right decision during the season.
At GeoPard, we solve this problem with the help of several machine learning algorithms that work with very high accuracy.
We define a cloud mask and a shadow mask at the level of the whole image, and for each field-image pair, we consider what percentage of the field is covered by clouds or shadows from the clouds.
In automatic analytics, we take only completely cloudless images for a specific field, which allows the user to be confident in making decisions based on multi-year analytics.
The GeoPard user has the opportunity to view and even run the analytics himself, even on partially cloudy images.
This can still be very useful, for example, during the season to apply Variable Rate spraying with crop protection products based on the latest satellite image, in which 10% of the field’s area is under the clouds. This means that for 90% of the field’s area, the decision will be based on verified data.
Also, users of GeoPard can easily check source images on a regional level in Near-Infrared view, which helps to distinguish clear land from clouds and shadows (see picture attached).
On UI cloud filter is located in the top menu of “Satellite monitoring” module, see screenshots attached.
Keep in mind that digital ag companies can still struggle with the correct detection of clouds and shadows. In the attached screenshot one such a company creates VR fertilizer map based on clouds and shadows data. So the wrong map leads to wrong agronomic decisions and wrong outcomes in the end.
GeoPard detects clouds and shadows with a high level of accuracy and doesn’t propose to make your Variable Rate decision based on clouds. You can always look at the source images in different views and indices and zoom out to see the picture around.
Not every solution is the same, choose the best options available on the market!Remote sensing