This story begins last Christmas (2015), which is when I received a nice gift from my wife. She gave me a solar-powered Acurite weather station for our house (Figure 1).
She knows that I am very interested in data and weather, so this seemed to me to be the perfect gift.
If you have read this blog before, you probably already know my motto:
I treat all data as guilty until it is proven innocent
This article is a perfect example of why I am part lawyer, part data scientist, part data dork, and why I believe in that motto. What you will see might surprise you, but then again, maybe it won’t if you are naturally skeptical.
In the beginning of my career, I spent a couple of decades deploying all types of electronic data collection systems for the environmental studies I was involved in. I installed pressure transducers to record water level responses, weather stations to collect precipitation, temperature, humidity, and atmospheric pressure, and many other types of systems.
In other words, I’ve been around the block a time or two when it comes to recording, analyzing and using electronically-collected data. I know a lot about what can go right, and more importantly, what can go wrong. One example of what can go wrong is having lightening strike your recording equipment. I can tell you that pressure transducers and recording systems do not like lightening strikes!
So when my wife gave me the weather station gift back in December, I was pretty happy to install the system and start recording the data. I had a feeling that a blog post was being born on the day that I installed the unit. I had visions of Tableau visualizations and all kinds of insights related to my weather data.
The weather station installation was a simple thing to do. There were no calibrations necessary – all I had to do was level the unit and screw it down onto the railing of my deck, as shown in Figure 2. Afterwards, I had to plug in the inside unit and my weather data career was off and running. Wireless connections exist between the solar-powered weather station and the interior unit.
Once data collection began, all I had to do was connect my laptop to the inside unit to download the data. The data is recorded and stored every 12 minutes, which means that there are 5 readings per hour. There are 15 pieces of data being collected, as shown in Figure 3.
An example of the data collected at my house is shown in Figure 4. Initially, I did not understand that the data logger only held 2 weeks of data. Therefore, by the time I did my first data download, the data remaining in the logger started on Feb 15, 2016.
Using Tableau to Visualize the Weather Data
As with most projects like this, I loaded the data into Tableau to visualize the time series history of the data being collected. I thought that after a few data downloads, everything was looking good. I saw the outdoor temperatures starting to climb as springtime approached, as shown in Figure 5.
I got all geeky and set-up parameter driven scatter plots so I could investigate any variable against any other variable, as shown in Figure 6. This link contains the Tableau packaged workbook for this data. You have to download the file and rename it from *.zip to *.twbx. I had to do this to fool the wordpress content management system. On the page called scatter plots, you will find the parameter driven scatter plot approach.
At this point, I was feeling pretty good, almost like a real weatherman! That feeling lasted until the morning of March 21, 2016. That is the morning that reality slapped me upside the head.
Upon Returning From the Beach
On the previous night (March 20, 2016), I happened to have returned home from a weekend at the beach in the South Carolina Low County. I also happened to be wearing a tee-shirt and pair of shorts at the time I opened the car door when I arrived home. When the door was opened, I was hit with big wind gusts, cold rain and 40 degree air temperatures. It was about 6 in the evening. I made a mental note about how cold it was after having been to the beach.
The next morning, I downloaded the latest weather station data. I almost immediately recognized a big problem. Even though the interior weather station screen was reporting morning temperatures accurately (37 degrees), the downloaded data was not even close to being correct. After checking on the ability to calibrate temperature readings as shown in Figure 7, I knew I was in trouble because the weather station data appeared to be inaccurate.
The Outdoor Temperature Inaccuracy
I decided to do a full quality assurance (QA) check to determine the accuracy of the Acurite weather station data. To do this, I went to the weather underground and found the nearest weather station to my house. I went through a laborious process of downloading about a weeks worth of hourly data to be used as a comparison to my data. If you want to see my real-time description of how I did this, you can watch the dorky video shown in Figure 8.
By comparing my outdoor temperature data to the weather underground data as shown in Figure 8, I can see that the Acurite weather station data is NOT ACCURATE! I also compared other variables such as humidity, and these were also inaccurate. Therefore, I conclude that the Acurite weather station data is guilty (i.e., it is inaccurate), and I cannot find a way to prove it to be innocent. I have stopped downloading the data and I only use the system to see real-time outdoor temperatures and/or other extreme weather events as shown below.
The question remaining for me is this: How can the data logging system be recording numbers that are different than those being shown on the indoor screen? I wonder what data is actually being stored? This is very perplexing.
The Windstorm That Tossed and Trampled the Trampoline
A couple of days ago (April 7th), a big windstorm came and slammed our house. Jett’s trampoline got lifted and tossed across the yard like a giant 3D-frisbee. I actually saw it flying through the air. Here is some weather station information about that day.
Figure 9 shows Jett in his trampoline before the wind storm hit. It was a perfectly good trampoline.
Shown below is a video recorded just after the peak gusts hit, which are shown in the Figure 10 video to be 33 mph. The weather underground monitoring station also recorded a peak gust speed of 33 mph on that day. It seemed like the peak gusts were higher than that because the house felt like it was vibrating. Typically in Knoxville, there is very little wind blowing throughout the year.
Finally, Figure 11 shows the aftermath of the storm. The trampoline may be irreparably damaged.
If a weather station system is constructed to record and store data, how can it record such bad data? I’m considering asking Costco for a refund on this unit. I’m very disappointed, but once again, Tableau allows me to get the job done with the greatest of ease!
I returned the unit to Costco a couple of months ago. I miss having the real-time outdoor temperatures displayed in my kitchen. Those were always accurate, as well as the indoor temperatures, ever since I installed the unit.
After watching dozens of people report problems with their units via the Acurite forum, I realized that my problems were not isolated to my unit. My problems were not due to how the unit was installed, either. If you read the comments below, George is insistent on the installation being the problem with the data that I showed in this report and he thinks I wrote this article to “sell my Tableau services”. Well, anyone who is a fan of this blog knows that I am not trying to sell services.That is simply not my mission.
Understanding data and teaching people how to comprehend data is my mission. George ignores the fact that I have multiple degrees in geology and have spent over 20 years installing environmental monitoring systems. He also ignores the fact that I am a specialist in advanced analytics. He also ignores the data shown in Figure 8. He ignores the OK correlation between the weather underground site and my unit between March 13 – 17th and then he ignores the fact that my unit went off the deep end after March 19th. To me, this appears to be some sort of system failure. Any time data begins deviating to this degree, there is a malfunction or mis-calibration, neither of which I could fix.
Even though the unit was reporting the correct temperature on the panel inside the house throughout the time frame of interest, what was being stored in the unit in the csv file was incorrect. As I stated in the article, I cannot understand how a 12-minute aggregation of temperatures around 37 degrees could be stored in the file as an average of over 70 degrees, as shown in Figure 8. Either the programming is bad or the system developed a malfunction.
I could have gone further back in time to see if my home system correlated to the data reported by nearby weather underground system. However, I didn’t have time because it took a long time for me to download page after page of weather underground data. I had enough data to know that my system developed a problem. What the problem was, I don’t know. I’ll simply get another weather station one day and return to watching the weather at my house.
George gave me a great reference today to download a Citizen Weather Observer Program (CWOP) guidance document, which I did and I read, for installing the systems. I will re-read it, understand it, and be sure to use it when I install my next system. George and I can continue to disagree on what caused my problems, but it really doesn’t matter anymore since I returned the unit. So thanks George, for helping me try to understand your point of view.
Finally, I contacted Acurite directly and asked them to send me another unit to replace the one I had. I told them I would do whatever they wanted me to do with respect to installation. I told them I would test it and write another report to discuss my findings. To this day, I haven’t heard back from them.
Next Article: A final letter to George