SeattleRibbitNetwork

I will be adding to this story as the project progresses, with more anecdotes and lessons learned. Stay tuned...

1. Electronics in the rain (and snow)

In the summer of 2021, I received a "version 1" Frog sensor (which may actually have been the first "production" build of the version 1 to be deployed). I'd had prior experience setting up environmental sensors and data loggers, and therefore thought I could help develop some of the documentation and instructions for deploying sensors. My initial task was therefore to follow the existing setup instructions, make notes and edits where those instructions were confusing or missing information, and deploy the sensor to begin logging observations!

The existing instructions were mostly sufficient, and any small edits I had were incorporated on the project's github page for the version 1 hardware. The sensor was up and running!In late summer of 2021 the weather was dry, but I knew that the rain (and snow) that persists throughout much of our fall, winter, and spring was on its way. I had some concerns about how well this Stevenson screen (the green outer protective covering) would actually protect the internal electronics from moisture. Additionally I was concerned about my power cord setup, where the USB-C Frog sensor power cord plugged into an extension cord outside. But this was all an opportunity to learn about deploying the Frog sensors to inform future iterations and best-practices.

My first attempt at "waterproofing" the power cord connection was very crude, plastic bags wrapped around and held in place with rubber bands. While this seemed to work through the onset of rainy fall weather, I also tried out using a second Stevenson screen to house the connections, mounted below the sensor on the same PVC pipe.

This was an awkward setup that didn't quite fit inside the enclosure given the 90 degree angle that the USB cord connector made with the extension cord, so I returned to my plastic-wrapped temporary fix for the time being. Surprisingly, this lasted through months of rain and snow without power issues!

Only later in the spring of 2022 did my crude "waterproofing" fail, which I then replaced in true MacGyver fashion with a plastic food container and duct tape. But my hasty replacement had a fatal flaw! While waterproof to rainfall on the container itself, I did not account for water running along and down the surface of the power cords and into the box. The sophisticated Microsoft Paint diagram below illustrates my mistake.

Since I've corrected this mistake and the latest food container enclosure only has holes for cables to enter from below, this seems to be my most robust setup and one that I would recommend to others. Perhaps I should make a step-by-step DIY budget enclosure guide in the near future.

2. What are these Frog sensors measuring?

The main purpose of the Frog sensor is to measure the amount of carbon dioxide (CO2, a major anthropogenic greenhouse gas) in the air. However, in order to get accurate results, a few other measurements are needed: air tempreature, pressure, and humidity. All of these variables can be seen on the Ribbit Network live data dashboard.

How do the frog sensors measure CO2 concentration?

Carbon dioxide is measured using what is called a nondispersive infrared (NDIR) sensor. These sensors are a type of spectroscopic instrument, that is they measure light and how that light interacts with a material at specific wavelengths. In the case of measuring CO2, light at a wavelength in the mid-infrared part of the electromagnetic spectrum is emitted, passes through a small chamber that is open to the outside air, and a detector then measures the intensity of that light. Light at the mid-infrared wavelength of 4.26 μm (4.26 x 10^-6 meters) in particular is absorbed by CO2 molecules. That means that a portion of the emitted light will effectively be blocked by CO2 in the air in proportion to the amount of CO2. The difference between the intensity of the emitted light and the intensity of the detected light can be used to determine the CO2 concentration. (If our eyes could see light at these infrared wavelengths, high concentrations of CO2 in the atmosphere would make the air look hazy as it blocks the light, whereas the air would look clearer at low concentrations of CO2.)

This plot from the National Institute of Standards and Technology shows the transmittance of light (y-axis) through CO2 across a range of wavelengths (x-axis). Note the narrow dip in transmittance down to nearly zero left of the 5 μm mark. This is the 4.26 μm absorption band of CO2 that the NDIR sensors are looking for.

These measurements need to take into account air temperature and pressure, which is why the frog sensors also measure these properties. The ideal gas law, in the form PV = nRT describes the relationship between the pressure (P), volume (V), temperature (T), and amount of a gas substance (n) with the ideal gas constant (R). We are interested in measuring the concentration of CO2 in ppm (parts per million) which is a ratio of the number of CO2 molecules to the total number of molecules in a sample volume of air.

The NDIR sensor however, is measuring through spectroscopy the amount of CO2 (n) in a particular volume (V). Rearanging the ideal gas law equation we can get n/V = R(P/T), showing that concentration per volume (n/V) will be proportional to pressure (P) and inversely proportional to temperature (T). This means that as air pressure increases more air molecules will be within the sample volume and we can expect the amount of CO2 detected in that volume to be greater even though the ratio of CO2 to total air molecules remains the same. Conversely as temperature increases and air molecules move around faster fewer air molecules will be within the sample volume and we can expect the amount of CO2 detected in that volume to be smaller. Read more details about how temperature and pressure affect NDIR CO2 measurements here.

What "should" the CO2 data look like?

So what does CO2 concentration data plotted over time look like? Why does it vary so much over the course of the day? if you are looking at data on the dashboard you might be wondering how to interpret what you're seeing.

Cycles and patterns:

Plants take up CO2 for use in photosynthesis during the day. Your CO2 sensor may pick up on this daily (also called diurnal) cycle! In areas with lots of vegetation, you may see CO2 concentrations rise during the nighttime when plants are not performing photosynthesis, but as the sun rises and plants "wake up" the CO2 levels may decrease. You can see evidence of this cycle in the plot above.

This plant-driven cycle also changes seasonally, with deciduous vegetation active in the summer lowering CO2 concentrations, and inactive in the winter allowing CO2 concentrations to rise. In fact, because most of Earth's forests are located in the Northern Hemisphere, this seasonal cycle can be seen in global estimates of atmospheric CO2. During the Northern Hemisphere summer global average CO2 is less than that in the Northern Hemisphere winter. Read more about the differences in CO2 patterns by latitude in this article from the Scripps Institution of Oceanography.

You may also see diurnal patterns caused by human activity. If your sensor is in an urban or suburban residential area you may see increases in CO2 concentration during peak hours of car traffic, such as from morning and afternoon commutes as people drive to and from workplaces. In the plot above, there are some rises in CO2 concentration (not the narrowest spikes though) around 8 am and 5 pm on the 17th that may be caused by car traffic.

For more examples, see this research paper (Imasu & Tanabe, 2018) which used NDIR CO2 sensors at different locations around Tokyo, Japan to look at CO2 concentration patterns.

Noisy data or spikes:

You may see "noisy" data where the CO2 measurements bounces up and down minute to minute creating jagged lines on the plots (such as in the plot above). This is to be expected somewhat, and can likely be attributed to the accuracy of the NDIR sensors. These relatively low-cost sensors have a manufacturer stated accuracy of +/- 30 ppm. This means that even if the local CO2 concentration is not changing, the measurements may bounce around by this amount.

Do you see "spikes" in the CO2 concentration where it rises steeply much more than typical noise, then drops back down over the course of several minutes? This is likely not noise, but a true measurement of air with elevated CO2! Think about the placement of your sensor, is it near a road or parking lot where passing or idling cars can create a "cloud" of exhaust that blows by the sensor? These spikes are signatures of nearby CO2 sources before the gas has dispersed. The plot above has several big jumps in CO2 concentration in the morning and in the evening, likely due to the sensor being installed near a parking lot, it could be picking up on local emissions from cars coming and going from the parking lot.

3. Detecting a Volcanic Eruption!

The eruption of Hunga Tonga-Hunga Ha'apai in 2022 and resulting tsunami was a disastrous event for the people of Tonga, and unfortunately resulted in fatalities, injuries, and other damages. I mention this because I don't want to ignore the tragedy of the event and the very real human impacts when discussing environmental observations of the eruption. Though not a natural disaster connected to climate change, it does serve to remind me that while working in the climate/environmental monitoring space what I might see as an interesting technical or scientific problem is in reality a disaster or tragedy for another person. And more often than not, those affected the most are from communities that did little to cause our current climate crisis.

On January 15th, 2022, the Hunga Tonga-Hunga Ha'apai submarine volcano erupted and was one of the most powerful eruptions recorded in the modern era. Many scientific instruments recorded the impacts of the event, from seismic sensors to satellite imagery and even the Frog sensors on the Ribbit Network.

Two Ribbit Network sensors picked up the the atmospheric pressure wave following the volcanic eruption in their measurements of barometric pressure. The plot below shows pressure over time for a sensor in Auckland, NZ and one in Seattle, USA.

A rough calculation of wave travel time / distance gives wave speeds ~315-320 m/s. What does this tell us about the atmosphere between the volcano and these two sensors if the speed of sound is about 340 m/s at sea level?

This was an unexpected observation for the Ribbit Network, but demonstrates that the ancillary environmental measurements used in computing CO2 concentrations may have other applications we haven't yet imagined.

4. Over 6 months of (nearly) continuous Ribbit Network observations

Though I've had my single frog sensor running (as continuously as I can manage with intermittent losses to power) for over a year now, I only have a data archive going back just over 6 months. See the plots below, the grey dashed lines are the 1-minute samples, and the black lines are the 12-hour average value.

For the temperature time series we can see the daily zig-zag of cold nighttime temperatures to warm daytime temperatures in the grey line, while the black line gives us approximately a nighttime average and daytime average temperature. We can also see the slow warming of air temperatures from the start of the time series in May 2022 into a peak around late August, then the cooling of air temperatures into the fall months.

It is apparent that CO2 concentrations fluctuate quite a bit given the dramatic ups and downs of that 1-minute data line. Taking an average over a time window helps to smooth out those short term fluctuations and maybe see some longer term trends. Unlike the temperature time series, I don't yet see a seasonal trend in CO2. Perhaps that will only be apparent when we have more than a year of observations to see the seasonal changes in plant uptake of CO2 (or trends in local car traffic).

Barometric pressure has much less variation on short timescales, our 1-minute and 12-hour averages line up very closely. And finally, relative humidity has day/night fluctuations like we saw in the temperature time series, and it looks like we can see a slight decrease in relative humidity in the summer months.

This was a short update, but stay tuned for more!

I should also note that we've added a data disclaimer on the Ribbit Network Dashboard, which I'll share here for transparency: "The real-time Ribbit Network observations displayed here, and available for download as comma-separated value (CSV) files, are raw data values that have not yet undergone any quality control. See further information in the FAQ."