Wildlife monitoring with the help of distributed artificial intelligence

Network coverage

Animal swarms tend to occur in remote areas beyond the reach of terrestrial communications infrastructure. But large amounts of data can often only be transmitted within days or even weeks in poorly connected areas. Therefore, it is necessary to connect existing networks via satellite to enable global connectivity.

Data rate

Sensors make it possible to measure data on animals and in their environment, from which scientists can derive important information on the state of ecosystems. The amount of data produced is problematic, because in principle more data can be collected than transmitted. Instead of risking incomplete data sets, local preprocessing of sensor data can reduce the amount of data to be transmitted. This means that less data rate is required for the transmission.

Latency

The limited data rate and the lack of terrestrial network coverage result in high latency, a delay in the transmission of data. This in turn can result in delayed data evaluation. In situations where fast action is required, such a delay can be problematic. Local preprocessing by AI enables a reduction in the amount of data to be transmitted and thus lower latency. Therefore, the connection of local networks to the satellite also plays an important role here in order to transmit the information generated in the local AI network quickly and efficiently.

Hardware limitation

In order to be able to process data directly on local hardware, the computing power available there must be sufficiently dimensioned. This can become a challenge, especially for mobile data acquisition by transmitters attached to small animals. Since vultures can only carry a small amount of their own body weight without being affected in their behavior, a strong limitation of the hardware is necessary. This is because computational power is always correlated to the resulting energy consumption and therefore battery size. The larger the battery, the heavier it becomes. The execution of complex AI algorithms directly on the animal transmitter is consequently only feasible to a limited extent, as the hardware required for this would be too heavy and too large.