Death Valley, a Year Later

About a year ago (May 2021), I installed a weather station next to the National Weather Service’s (NWS) official station at Furnace Creek in Death Valley National Park (California, USA). To familiarize yourself with this project, read my previous article. The purposes of this installation and the collaboration with both NWS and the National Park Service (NPS) were to provide additional temperature measurements for comparison with the official temperature and to co-locate comparable sensors that would be possible replacements for the discontinued CS215-L Digital Air Temperature and Relative Humidity Sensor currently in use on the official station.

This collaboration was initiated because of the high temperature measurement recorded in August of 2020. While the sensor that was in use at that time is still being evaluated, there are some interesting results of the collaboration that I can share.

What about the co-located data?

As mentioned in my previous article, the station I installed includes sensors in both active (fan-aspirated) radiation shields and passive shields. Two of those sensors were HygroVue™5 Digital Temperature and Relative Humidity Sensors—one in a passive shield and one in an active shield. This sensor is the direct replacement for the CS215 (also in a passive shield) installed on the NWS station. In addition, a 109 Temperature Probe was installed in a passive shield, and three replicate thermistor beads were installed in an active shield. The beads and 109 compared well to a high-accuracy platinum resistance thermometer (PRT) in a liquid bath at 50, 55, and 60 °C. My thinking was that the beads—mounted in an active shield—would provide the highest accuracy measurement of air temperature against which to compare the rest of the measurements. This thinking was not wrong, but there are some key caveats.

What qualifies as an extreme?

Before going into any detail on the data from the summer of 2021, I want to go over some bigger-picture ideas.

One consideration is what might be called “representativeness.” A sensor may measure the conditions immediately around it very accurately, but if those conditions don’t represent a broader area of interest, absolute accuracy may be less relevant. Carefully selecting the location (or siting) for a weather station and mounting the sensors at or near two meters above ground help address the question of representativeness.

There are several factors that contribute to the accuracy, or lack thereof, of an air temperature measurement. The most important of these are exposure to solar radiation and the impacts of wind. As I mentioned in my previous article, a standard (passive) radiation shield reduces the effects of solar radiation on the measurements, but during times of low wind speed, an upward bias can result from lack of air movement around the sensor. Therefore, a fan-aspirated or active radiation shield is considered the best available method for an accurate air temperature measurement. Of course, at the level of electronics, the sensor and data logger contribute to uncertainty in the measurement as well. However, these uncertainties are typically on the order of tenths of a degree Celsius rather than the several-degree bias that can result from solar radiation.

Comparing measurements in space or over time is complicated by all these factors, as well as one more that varies from sensor to sensor: how quickly the sensor equilibrates to changes in ambient conditions. This is quantified with something called the time constant. In brief, the time constant for a given temperature sensor is quantified in a lab setting by subjecting the sensor to an instantaneous step change, usually of several degrees. The time constant, represented by the Greek letter τ (tau), is defined as the time for the measurement to come to 63% of that step change at some known wind speed^1,2,3. Due to the nature of that equilibration curve, it takes about three times τ for the sensor to adjust to 95% of that change. The World Meteorological Organization (WMO) recommends a time constant of 20 seconds at 1 m/s wind speed for air temperature measurements¹.

What about processing data for recording extremes?

Remarkably—given the importance we see as a society in extremes—there do not seem to be widely accepted standards for what it takes to be an extreme or how to quantify it. Most data loggers, unless explicitly programmed otherwise, will simply record maxima and minima as the highest or lowest single measurement over a given interval (hourly, daily, etc.). Note that one might be tempted to call these “instantaneous” measurements, but the time constant makes this incorrect. If an accurate sensor has a fast response (short time constant), then its maximum recorded temperature over a given hour may be accurate, but what does it actually represent? Is that measurement important? That really depends on how those measurements are to be used. For reference to what a human would perceive, it may not be very important.

Some level of data processing for recording extremes seems appropriate for measurements such as wind and air temperature, among others. Indeed, the WMO^1,4 does have a processing recommendation for temperature that is based on their recommendation of a sensor with a 20-second time constant. This recommendation is to sample the sensor four times per time constant and average those measurements once every minute. This then becomes the sample. This implies that maxima and minima are then calculated from those one-minute averages. Indeed, using this one-minute average for maxima is recommended by the WMO’s Automatic Weather Station (AWS) Tender Specification⁴. While this makes sense, there are many sensors in use around the world, particularly temperature and relative humidity sensors, that have a time constant that is much greater than 20 seconds. Neither the station I installed nor the NWS station at Furnace Creek does this kind of processing; however, all the raw data are stored, so we can do post-processing.

OK, but really, what about the co-located data?

Finally, we’re getting to some of the results from the 2021 summer data collection!

As you might imagine based on the above information, there are a lot of nuances and caveats to comparing all the data we have from the two co-located stations. I presented a poster (“Measurement and Uncertainty in Death Valley Temperatures”) at the American Meteorological Society Annual Meeting in January 2022 that goes into more detail, including figures and tables that show the station setup, but here are some highlights.

On July 9, 2021, at approximately 4 p.m. local time, the NWS station reported a maximum temperature of 54.4 °C (130 °F), which, like the temperature reported in August of 2020, would be a record. Looking more closely at those data (see Figure 1a), there is a relatively short-duration spike in temperature of about 2 °C.

Figure 1a: Click the graph for a larger image.

If we only had the data from the NWS station, this would be suspicious. However, a spike of similar magnitude is shown in the data from the sensors (all seven of them!) on the station I installed. In addition, that spike is also evident after processing the raw data to ten-minute averages (Figure 1b).

Figure 1b: Click the graph for a larger image.

Therefore, the spike seems to be a real phenomenon, though I’m unsure of the cause and am interested in any ideas. (Please don’t hesitate to add comments to this article or contact me directly). All that said, the measurements on the station I installed were noticeably lower than that on the NWS station. The closest was the 109 probe at 53.6 °C (128.5 °F) comparing raw, one-minute measurements. So, the record-breaking number is not supported by the co-located measurements.

More generally, looking at a full year of data beginning in mid-May 2021, there are some interesting results. As mentioned earlier, the temperature and relative humidity sensor model that has been used by the NWS (CS215) has been retired, and the direct replacement is the HygroVue 5, which fits in the aspirated radiation shield (TS100SS Aspirated Radiation Shield). Comparing the overall average difference and its standard deviation (SD) between the CS215 on the NWS station and the HygroVue 5 in a passive radiation shield shows strong agreement, but with an interesting difference at night (Table 1). At night, the HygroVue 5 is about 0.25 °C warmer, on average, than the CS215. This difference is puzzling, and I don’t have a clear explanation for it.

Table 1: Click the table for a larger image.

Another interesting comparison is the two HygroVue 5 measurements (in passive and aspirated radiation shields) and the thermistor beads in an aspirated shield. I used the average of those three beads as what I’m considering the measurement of the highest accuracy. At night, unsurprisingly, all three of these compare very well (Table 1). During the day, the aspirated HygroVue 5 shows an upward bias, but within a generally acceptable range. The HygroVue 5 in a passive shield, however, shows a notably larger positive bias, which is similar to that of the CS215. The variations of these differences are the greatest of all the comparisons in Table 1.

What’s next?

Originally, I had planned to take the station down after the 2021 summer season. In conversations with my collaborators, we decided to leave it up for at least another summer. I returned to Death Valley in early April 2022 to check on the station, do any necessary maintenance, and make two key changes.  One change was to add a thermistor bead to the passive shield where the 109 probe is located. This gives us a direct comparison in terms of time constant between the aspirated and passive shields. The other change was to store raw five-second measurements on the NWS station so we can do deeper evaluations related to time constants.

I hope you’ve found this article informative, and I’m looking forward to seeing the results of this 2022 season. If you have any ideas about the cause of the noted temperature spike, please post them below or contact me directly at dbaker@campbellsci.com.

References

¹CIMO Guide 8. WMO Guide to Meteorological Instruments and Methods of Observation. https://community.wmo.int/activity-areas/imop/cimo-guide 

²Instrument Engineers’ Handbook: Process control and optimization. https://www.google.com/books/edition/Instrument_Engineers_Handbook_Volume_Two/TxKynbyaIAMC?hl=en&gbpv=1&printsec=frontcover 

³Electronics Tutorials: Tau – The Time Constant: https://www.electronics-tutorials.ws/rc/time-constant.html 

⁴IOM-136. Generic Automatic Weather Station (AWS) Tender Specifications. https://library.wmo.int/records/item/57830-generic-automatic-weather-station-aws-tender-specifications