Process Mining on Ocean Science

Computers are usually associated with the use of social media or the next travel booking. In this article, I will show you how computers can also be used in unusual disciplines such as process mining on ocean science.

Who doesn´t know this situation: We spend our typical vacation at the beach and enjoy the cooling water. But what influence do currents and wind have on the water temperature and how can we use computers to find this out?

Researchers have defined a so-called “upwelling” process, which shows a connection between windstress, horizontal and vertical currents and water temperature. This processes is visualized in the following image:

With traditional analysis researchers are using linear regression and correlation measurements to prove the upwelling process. But every step in this analysis has to be done manually and you need domain knowledge to understand everything in detail.

And how does Process Mining on Ocean Science work?

This is where computers come in. The idea is to teach the computer what the domain experts know and to discover new relationships between the measurements, mentioned above.

Upwelling is a well-known phenomenon. But it is unclear, whether the upwelling process is really going the way we think. Process Mining can give new insights into physical ocean processes. Therefore I implemented some algorithms to transform ocean data to a process model.

What have I found out?

Beside the expected relationship between the physical variables (see Upwelling), the process model also shows additional behavior. Although the wind does not change, the westward current occasionally increases. In the process model this is the path from “Windstress stays” to “Westward Current raises”. Although the upwelling drops, the surface temperature occasionally does not increase. In the process model this is the path from “Upward Current drops” to “Temperature stays”. Both insights cannot be spotted with traditional analysis in ocean science. Thus, process mining provides an added value.

And how can we continue?

The subsequences, which are identified as outlier, could be analyzed further. Therefore, I identify the exact time periods and the behavior of the measured values to understand the upwelling process in more detail and to explain, why I identified these subsequences as outlier.