Testing drones and computer vision for acquiring glacier melt observations

Climate change is affecting glaciers worldwide, leading to unprecedented melt rates. In this context, establishing systems that provide near-real-time glacier information can be of high interest. However, the effort for acquiring real-time, in situ glacier observations is large.

In a previous study, we investigated the potential for automated acquisition of real-time mass balance readings by using optical cameras installed in-situ and computer vision techniques. The setup proved to be useful for obtaining melt rates with a temporal resolution of 20 minutes. However, it is not feasible to cover an extensive portion of a glacier with such a setup.

In our contribution, we present a method to acquire glacier mass balance readings with a custom drone equipped with a camera. The principle is to acquire images of a color-coded stake, from which surface mass balance can be determined via the glaciological method. To autonomously approach and read the stake, we exploit a combination of computer vision techniques and geometrical triangulation.  The results of off-glacier test flights, as well as four flights on Rhonegletscher, Switzerland, prove that the system is successful in detecting the stake in the videos captured by the drone. The determined stake position has uncertainties of 2.4 - 4.6 m, thus being sufficient to safely approach the stake. We investigate the main factors influencing the performance of the method in more detail, and discuss potential future developments of the system.