I'm on a boat...evaluating the temperature sensor

How do we figure out the accuracy of a temperature sensor? By throwing it off of a boat! Actually, it's a little more complicated than that....

 

The CTD on the deck of the R/V Rachel Carson with Kim's Apple Watch Ultra strapped to the side of the water sampling cage.

Last week, we tagged along with scientists from the University of Washington on their semi-annual cruise to measure the health of Puget Sound. As they measured seawater temperatures with their highly-accurate and super-expensive CTD, we measured seawater temperature with our practical and not-so-expensive Apple Watch Ultra.

We did three vertical profiles in southern Puget Sound (profile = lowering and raising the instruments at one location to see the the vertical structure of seawater). By comparing the temperature difference between two instruments, we could estimate the accuracy of the temperature sensor on the Apple Watch Ultra.

 

On the left, the raw profiles of temperature versus pressure (aka depth) from the CTD (red) and the Apple Watch Ultra (blue). The difference between the two sensors is on the right and separated into the error for the downcast and the upcast. They upcast and downcast are different because of the long response times of the temperature sensor.

We found that the average error for the Apple Watch Ultra was -0.46° C, in other words, it's accurate to about 0.5° C. This is larger than the manufacturer's stated accuracy of 0.1° C.

Even though the error is larger than expected, it's not surprising or alarming. Here's why:

1. This is field testing, where the conditions are dynamic. Temperature is changing rapidly with depth. To really estimate the accuracy of the temperature sensor on the watch, we need to assess it in highly controlled conditions, AKA we need to head back into the lab and stick it in a calibration tank. But field tests are a great way to understand how the sensors are behaving in real world conditions.

2. It's been pretty clear from this experiment and the others in Lake Washington that the temperature sensor has a slow response time. That means that it reacts slowly to changes in temperature. Take your temperature at the doctor's office in one second by swiping your forehead with a thermometer? That's a fast response. Waited ten minutes with an old school liquid thermometer in your mouth? That's a slow response. The temperature sensor on the watch is more like the old school thermometer which is definitely causing some of the error.

1. We think this is all correctable with physics-based mathematical algorithms. I've got a series of tank tests planned based on the results from this cruise. We will characterize how the sensor behaves, develop some corrections, then apply them to the data.

So now we are onto the fun part (at least for Kim), where we figure out how to supercharge off-the-shelf sensors with software.

We want to give a huge shoutout to Jan Newton, Anna Boyar, the NANOOS science team and the captain and crew of the R/V Rachel Carson for having us onboard and the opportunity to test our equipment. THANK YOU!

 

The R/V Rachel Carson docked in downtown Tacoma. Mount Rainier is glowing at sunset in the background.