KEDR system

What is KEDR system?
The KEDR System is a digital platform created to facilitate forest monitoring. The system is based on an interactive map that visualizes all kinds of data needed to protect forests from illegal logging.
A user of the KEDR system, for example, a forest inspector who plans to patrol a certain area, can get all the available and necessary information right inside the system.
Thanks to this, the forest inspector is freed from a large amount of paperwork and can collect all the necessary information for verification in just a few clicks.
But the main feature of the KEDR system is not even the ability to quickly collect the necessary information, but the presence of signals about changes in the forest. The signals of the KEDR system are generated by a neural network that collects and processes satellite images from different times, and then compares them to identify new "clearings" in the forest canopy. For the analysis, free images of the Landsat satellite with a resolution of 30 meters are used, i.e. one pixel of the image corresponds to an area of 30 by 30 meters.
Thus, forest inspectors planning field raids can see in advance where there have been potential changes in the integrity of the tree canopy and the possibility of illegal logging, which undoubtedly increases the effectiveness of field and raid activities.
How the system work?
Neural networks and the system
Neural networks are data processing systems that are structurally similar to the networks of nerve cells in the human brain. Their main feature is the ability to capture the smallest patterns and relationships (so-called patterns) that may be invisible to the human eye or a traditional computer program; this allows neural networks to classify the analyzed information with amazing accuracy.
Like humans, neural networks are capable of self-learning. For example, if a person is shown several cards with the image of a tree, then in the future he will be able to recognize (classify) other images of trees, even if they differ significantly from the images on which the training was conducted. A similar process occurs with neural networks.
How does it differ from satellite surveillance?
How are new signals formed?
The KEDR System automatically checks the updates of satellite images of Landsat-8 and Landsat-7, if new images are available, the process of finding changes is launched. At the first stage, images are prepared: the system, using a special algorithm, "cleans" images from clouds, water bodies and shadows that can interfere with analysis.
In the figure, an example of a cloud mask created. Red shows the contours of clouds and shadows cast by clouds. These areas are excluded from further analysis
In the next step, the system prepares the stack. To do this, using the so-called harmonic model based on historical data (images of past years taken in a given area in similar conditions), the system builds forecast values of brightness and indices for each section of the analyzed area based on the position of the satellite camera and the azimuth of the sun.
After that, using a neural network, a pixel-by-pixel comparison of the attributes of new images with the predicted results obtained using a harmonic model (stack) takes place.
When comparing each of the pixels, the system assigns it one of three possible classes:
- «No changes», in the event that the values of brightness and pixel indices in the new image do not differ from the predicted values
- «New changes», in case when the pixel value has changed during the analyzed period
- «Old changes», in case when the pixel value in the new picture differs from the forecast values, however, this change was recorded in the last period.
In the figure, an example of a raster layer with pixels marked as forest changes. Red color corresponds to a high probability of forest changes, orange color - medium.
Loggings that have not been highlighted belong to the old changes.
The result of the neural network is a raster layer in which each pixel is assigned a value from 0 to 100. This figure reflects the confidence of the neural network that a change has occurred in a given pixel. For example, a value of 100 denotes 100 percent confidence of the neural network that a change has occurred in a given pixel.
Area of operations
The KEDR system was launched and officially put into operation in the Primorsky and Khabarovskiy provinces. The territory of the analysis in Primorye is currently represented by three test forest districts (Roshchinsky, Dalnerechensky and Ussuriysky). And until 2021, it is planned to expand the operation zone by three more forest plots (Chuguevskoye, Sergeyevskoye and Kavalerovskoye). In the Khabarovskiy province, the system operates on the territory of six test forest districts (Avanskoye, Bikinskoye, Oborskoye, Khorskoye, Mukhenskoye, and Sukpayskoye).
In total, since the launch of the system (taking into account illegal logging identified during system testing), mobile groups created to work with the CEDAR system, as well as WWF employees, have identified 108 illegal logging with a total volume of about 7 442 m3 and economic loss of 856,2 million rubles.
More specifically:
- At the stage of testing the system in 2015-2017, 37 illegal logging during the growing season and 13 illegal logging during the snowy/leafless period were detected. The total volume of illegal logging amounted to 3 903 m3, and the damage amounted to more than 639,8 million rubles.
- At the stage of using the system in Primorsky provncein the period from 2017 to 2020 44 illegal loggingwas recorded, with a total volume of 1129,72 m3, which caused damage in the amount of 92,5 million rubles.
- At the stage of using the system in the Khabarovsk province for the same period, 14 illegal logging was detected, with a total volume of 2409,27 m3, which caused damage in the amount of 123,9 million rubles.