Public Lab Research note


Can open source beat this?

by cfastie | July 20, 2014 18:47 | 5,273 views | 11 comments | #10939 | 5,273 views | 11 comments | #10939 20 Jul 18:47

Read more: stable.publiclab.org/n/10939


Last summer, Jeremy Littell of the US Department of Interior Alaska Climate Science Center installed some environmental sensors at four of our study sites at alpine treeline in interior Alaska. Earlier this month we met Jeremy at our sites south of the Alaska Range as he retrieved and replaced the sensors and placed additional sensors at another treeline site. The sensors record only temperature and light, but strategic placement allows inference of critical environmental variables which could influence the success of tree seedlings. Our 15 year study at these sites suggests that white spruce (Picea glauca) is invading alpine tundra, but we have no local data to link this process to climate change.
.
Den1_0477.jpg
LogTag temperature loggers (upper right) and the parts to build DIY shelters for recording air temperature.
.
Den1_0469.jpg
HOBO Pendant temperature and light loggers. These were placed on the ground to record surface temperature and light levels for a year.
.
The sensors are proprietary devices and require proprietary hardware and software to collect and read the data. If the current field tests are successful, Jeremy will want to deploy hundreds of these sensors in Alaska. I suggested that some clever members of the Public Lab community might be able to devise an open alternative which could save money and allow better replication, reliability, and data accessibility.
.
Littell_Plot10WTC2k.jpg
A year’s worth of data from three temperature and two light sensors at a single location at treeline. Three temperature sensors recorded a) air temperature at two meters above ground, b) ground surface temperature, and c) soil temperature several cm below the surface. Light level was recorded at the ground surface. When light level is zero (November into April), the ground is snow covered. Soil temperature (red line) stays right below zero when the ground is snow covered. A late January thaw did not melt all the snow (no light was recorded) or thaw the soil. Surface frost occurred through April and May, and had started in late August.
.
Jeremy’s sensors record data at least every hour for a full year. The sensor data allow inference about the beginning and end of the winter snowpack, the period of surface frost, the freezing and thawing of soil, in addition to diurnal and seasonal patterns of temperature.
.
Den1_0472.jpg
The HOBO IR data transfer device ($70). The LogTag loggers also require a proprietary data transfer docking station ($50).
.
Den1_0483.jpg
Jeremy assembling the shelters for the LogTag loggers which will record air temperature in a tree.
.
Den1_0517.jpg
A LogTag logger in a waterproof case for recording soil temperature for a year.
.
Generally only one sensor is deployed for each measurement so there is no redundancy or replication at each location, although four replicate locations are instrumented at each site along a 0.5 km transect.

The temperature sensors are MicroDAQ LogTags. These record up to 8,000 readings with a range of -40° to 85°C and cost $25 for 30 or more units. The batteries can last three years, but are not replaceable.
.
Den1_0510.jpg
A lone tree near treeline is chosen and the air temperature logger is hung 2 m above the ground on the north side.
.
Den1_0543.jpg
The air temperature logger in place.
.
The light sensors are HOBO Pendant loggers which record temperature and light. These record up to 28,000 combined light and temperature readings and cost $64 apiece. Replaceable coin cell batteries last a year.
.
Den1_0556.jpg
A spruce seedling near the larger tree is chosen as the site of loggers for surface temperature, soil temperature, and surface light level.
.
Den1_0534.jpg
A LogTag in its case is buried several cm under the moss, and a HOBO Pendant temperature and light logger is placed on the soil surface near the spruce seedling.
.
Den1_0536.jpg
The HOBO Pendant records temperature and light at the ground surface for a year.
.

30_sensors.jpg
Possible configuration of replicate and redundant sensors with central microcontroller. The microcontroller requires a battery and SD card for storage.
.
One approach to reducing cost and improving replication and reliability might be to use a microcontroller circuit to record the data from multiple sensors. Instead of recording air, surface, and soil temperature with three separate devices, a single controller could be wired to all the sensors. The sensors themselves are inexpensive ($1 to $3?), so duplicate and replicate sensors could be included. For each station, there might be three duplicate air temperature sensors, and three replicate nearby substations each with three duplicate surface temperature sensors, three duplicate soil temperature sensors, and three duplicate light sensors. The 30 sensors might cost a total of $50 to $100, and the controller might be less than $20. Lots of cables would be required to connect the controller to the substations 10 m away. So for less than the cost of the three sensors now being used ($114) it might be possible to increase both replication (three substations instead of one) and duplication (three of each type of sensor at each substation instead of one). No proprietary docking station would be required to collect the data, which could be recorded to a micro SD card (do those work at -40° C?).

Questions and next steps

Assuming that 50 of these stations would be needed, and each station includes 21 temperature sensors and 9 light sensors, this would require 1050 temperature sensors and 450 light sensors.

Is my estimate of the price for sensors in the right ballpark? The precision of the sensors has to be about 1° C, and the range of the air temperature sensor has to be -40° C to 40° C.

If the substations are 15 m from the microcontroller, what type of cable is required to carry the signal from nine sensors?

Is there an existing microcontroller capable of doing the logging? Would it be cheaper to design and manufacture 50 custom controllers?


11 Comments

Oh man, HOBOs. It's been awhile since I've had to use HOBO dataloggers. It sucks that they use a proprietary format.

"Open source" refers to software, whereas "open hardware" refers to the same concept with hardware. Unfortunately, while the open hardware community can make temperature and light sensors, the difficult engineering problem is ruggedizing the units for long term data collection. This is not something you'll usually find coming out of makerspaces or hackerspaces. Furthermore, for scientific data collection, meeting NIST standards is a must. I don't know any open hardware person who bothers with NIST. It costs a lot of money and takes a lot of engineering to ensure a device is NIST compliant. With HOBOs, you really are paying for that resilient case and electronics that have been endurance tested. In summary, yes we could make your sensor units, but no I don't think we'd be creating devices that are NIST-approved or rugged enough to withstand Alaskan winters without damage.

However, back in my day, we used to reverse engineer the HOBOs so that we could collect data from them without using the proprietary software. If you can put a HOBO datalogger in the hands of a few choice hackers, they might be able to get you software which can grab your data. The units will still be expensive, but you'll have more freedom to acquire the data if you can make your own connection.

The state of your HOBO units look very different from the ones we used to use. They used to have a standard 1/8th inch microphone cable into a special adapter to interface with a computer for some ungodly reason (proprietary hardware at its best).

Reply to this comment...


Bryan, maybe this is what you remember. We deployed a bunch of these in the tundra 15 years ago:
glassTemp-483-17.jpg glassTemp-483-24.jpg
.
They don't seem that rugged to me. I wonder if certain construction techniques will withstand -40° C better than others. Does an Arduino have an operating temperature range?

These also have to be protected from water, but that's a separate issue. It's not rocket science to put something in a waterproof case. But unlike the stand-alone loggers, my proposal calls for cables connecting each sensor. So that requires careful design where the cables pierce the enclosure, but it should not be much of an obstacle. Each microcontroller needs a weatherproof case with four cables attached. Each sensor is a tiny component that would need a tiny waterproof case at the end of a cable, and the light sensors need a transparent window. But maybe just a glob of silicone would suffice for each sensor (or cluster of redundant sensors).

Is there really any reason to suspect that a simple circuit board in a waterproof enclosure (with some silica gel) would not survive the winter under a tree in Alaska?

Is this a question? Click here to post it to the Questions page.

Reply to this comment...


Yeah the old HOBOs weren't very rugged, but that one in the image was one I remember using.

There are definitely temperature range issues with a lot of ICs. Generally speaking, microcontrollers in the popular markets aren't suited for extremes. However, a friend of mine has told me that AVR has milspec chips in deployment. So if you can find some AVRs that are milspec, you've got a start.

Waterproofing a system isn't necessarily all that is required. As the devices cool, water vapor trapped inside can condense on your electronics. Even if it is okay once or twice, direct sunlight can potentially evaporate trapped water only to let it condense again in the evening, making water a moving target inside the device. Most of these mass produced things have some amount of air control systems in place as they are constructed so that if they come off the line watertight, little water vapor will make its way inside.

But when I worked at a place that used HOBOs, the real problem was NIST. We had people in our group that were making new sensors to be bid off by the government for mass production. I learned a lot about NIST compliance from those folks. When we did data collects, we had to use sensors which were guaranteed to hold up to NIST standards for measuring. This was for the government. I would be both surprised and annoyed if USGS decided NIST did not matter and accepted data collects from sensors that were not rigorously tested against gold standard devices and tested for the maximum useful ranges. No matter how good of an enclosure, scientific data for government studies should be collected by NIST compliant devices. As I said before, I have yet to see open hardware makers do anything but cobble together a minimum viable product. NIST compliance might as well be walking on the moon.

The other question is changing your data collection system in a long term data collect. If you've been using HOBOs for 15 years, even if the model has changed, you're at least being held to a similar QA standard from a single company which produces a well-known and well-trusted sensing device. Simply walking away from that could make the data uncomparable before and after the change.

Reply to this comment...


@btbonval is probably right, the best short-term solution is to try to read the IR serial signal coming off of the HOBO devices. That seems to be the point of greatest cost savings immediately and without calibration.

That said, I don't think the weatherproofing is really that tough. [Rick Shory's Greenlogger] is just painted with a coat of clear nail polish. Its a problem that is being addressed by @donblair and the Riffle team-- we can bake'em (evaporate all the moisture) and then brick'em (pot them in epoxy) its very effective. You can always pour silica desiccant into the box if necessary.

For temperature, at least, the NIST calibration looks pretty doable. I could Here is a lab calibrating DS18B20's (~$3) against a NIST thermometer

Reply to this comment...


Would either of these help? https://www.spark.io/ http://www.digi.com/xbee/

Is this a question? Click here to post it to the Questions page.

Reply to this comment...


Rs485 can be used. Only 3 lines and 31 slaves with max 4000ft distance.

Reply to this comment...


Apparently Onset manufactures HOBOs to be NIST traceable, but they charge more money for units that pass a calibration test. Jeremy buys the less expensive units and tests them in an ice bath to ensure accuracy at the freezing point. The same could be done for DIY sensors. @btbonval has already done something like this. Inexpensive sensors probably have to be tested at temperatures throughout their range, and we don't really know how many would fail.

The published operating temperature range for Arduinos is -40° C to 85°C. A custom board like the Riffle with an Atmel chip might have to be tested at -40° C.

My proposal to cable sensors to a central microcontroller might not scale well. For the 50 installations I propose, about 2000 feet of cable would be needed. That much good cable like the RS485 that @gpenzo suggests would cost about $1000 ($5 per sensor cluster). And somebody would have to solder 1500 sensors (x 2 leads), etc, etc.

Using radios as @amysoyka suggests instead of cables would require a radio (plus battery) at each sensor cluster (which at about $20 per Xbee) would probably eliminate any cost saving over proprietary stand-alone loggers.

Because Jeremy already has the proprietary data transfer devices, hacking the data transfer does not provide much benefit. Buying the transfer devices was a trivial part of the investment so far.

Reply to this comment...


In short yes, I think your price estimates are good and opensource can beat this. It will be challenging running an Arduino for a year on a coin cell battery, but two people do run cheap arduino clone loggers for 2-3 years on two AA batteries with proper usage of sleep modes.

Reply to this comment...


This thread was part of my inspiration to build a few long strings of DS18B20 sensors:

http://edwardmallon.wordpress.com/2015/03/01/using-ds18b20-sensors-to-make-a-diy-thermistor-string-pt-1-the-build/

I am trying to improve the calibration of these sensors (from their default +-0.5C) using the ice & steam points of distilled water. I suspect that I will still have to obtain a $500 NIST pt100 to know for sure if the process has worked, but I want to know just how far you can get with DIY approaches. If I can get to +- 0.1 degree accuracy, then the answer to the original posters question about open source is definitely yes.

P.S. I recently deployed a couple of these multi sensor loggers underwater between 10 & 15 m depth. I will be going back to get them in ~6 months, so will be able to say something about the power consumption of these long strings of DS18b20's then.

Reply to this comment...


Any update on your sensor setup?

Is this a question? Click here to post it to the Questions page.

Reply to this comment...


After two successful deployments of our temperature chains I can confirm that my heat-shrink and epoxy sensor mounting method goes the distance under water. The only failures we've seen so far have been wire breaks inside the cabling due to excessive flexing when we stuff them in the bags during deployment/retrieval dives.

But the more interesting news is that I think I finally have a handle on calibrating cheap DS18B20 sensors to the point where they will at least give HOBO class sensor networks a run for their money:

DS18B20 Calibration: We finally nailed it!

Adapt the size of the holes in the dry well and the method should work for other temperature sensors.

Reply to this comment...


Login to comment.