Alert weather watchers (do you guys ever take a day off??!!!??) noted that the site went a little sideways yesterday. The main thermometer thought the high and low for the day was 32°F, the winds and other basic parameters were not right, and some of the supplementary pages failed to load or generated a pageful of errors. Hopefully, it’s working better today.
A junior web operator who shall remain nameless attempted to check the web sites’ configuration files into Git and instead loaded in the entire site, including dynamically updated data files and cached files. This means Git will pretty much always report the repository is out of date, and make cloning amd redeployment a serious hazard. In fixing the issue, the problem compounded: it turns out the “git rm” command does not remove the file from git as one might suspect, but instead OBLITERATES THE WORKING FOLDER OF PRODUCTION DATA and marks the file retained in (for historical purposes), but no longer active in the repository. Removing live data and cached files while the website is running had predictably disastrous results. Management has been notified and stern lectures delivered about diddling with the live production site without making three backups while clicking your heels together.
The management wishes to apologize to anyone who just wanted to know what the weather was outside. Yesterday, you would have had to go out IRL and look for yourselves; today, you can check on your phone again.
Updates to the website
Alert readers (I know you’re out there!) have noted that the Wunderground forecasts stopped updating recently. I had updated the API key provided to me by WU in exchange for posting my station’s data on their site. but I missed a portion of the update. The website continued onward, requesting updates from WU, logging the failed access, and republishing the out-of-date forecasts, I had no clue this was stuck, and I’ll look at finding a better way to monitor this. Currently, the script just logs errors into comments in the resulting web page, rather than writing out a log visible on the separate status /results page. I’ll see if I can patch this in.
While debugging this, I noted a couple of scripts were recently updated:
The Sun/Moon astronomy page has been updated with slicker graphics and a pretty neat graph showing moon and sun factoids (azimuths, elevation, zenith, etc.).
GW1100 Antenna Placement
I’ve been doing some experiments in finding the optimal placment of the GW1100 gateway. The gateway receives radio signals from Ecowitt weather instruments in the field: anemometer, UV, light, air & soil temperature, humidity, rain as well as in-house temperatures, barometric pressure and indoor pm2.5, does a few local calculations, hosts a small set of web pages, and makes the data available over the in-house intranet. If the signals from the instruments are too weak or attenuated (by distance, metal screens, water walls or dense construction). that data is lost.
I noticed when I added the meadow soil temperature instrument (an Ecowitt WN34S) last week, that the readings were erratic, so I checked on the strength of the signal, and eventually graphed all of the signals on our in-house WeeWx web page. I first moved the GW1100 to an upstairs rear window. Anemometer signals were better, but rain and T&H outside worse. I tried a back room on the lower floor and that may turn out to be optimal. I’ll continue to test.
Signal strength is reported by the GW1100 in a scale of zero to four. and the WeeWX software graphs an average value The nearby in-house “Upstairs” and “Basement” temperature/humidity sensors (two Ecowitt WH32) always have a strong signal. The outdoor T&HP sensor (Ecowitt WH31_EP) was good when the GW1100 was in north-facing windows at the front or back of the house, but signals were poor around 1600 when I tried the upstairs placement. The anemometer (an Ecowitt WH68) has never been very strong, as it is the furthest away, and behind a metal-roofed garage. When the GW1100 was upstairs the signal appeared the strongest, but the rain gauge and outdoor T&H were notably weaker.
Alert readers (I see you out there!) will wonder why I didn’t just use the internal web pages of the GW1100 that display a “signal” column, shown above. The problem with those is that the signal strength appears to be averaged over time, and not as immediately responsive as the graphs I generated with WeeWX. Overnight, all of the signal scores caught up, and all are displaying four bars in the morning.
Fall upkeep
The autumn season turned startlingly fast with morning temperatures dropping into the 40s, so it’s time to do the upkeep and maintenance on the outside instruments before winter weather makes them inaccessible. Today, I replaced the AA battery in the anemometer. I use lithium batteries for the outdoor instruments, as they are long-lasting, and tolerate a wider range of temperatures.
A glitch in the matrix
An update to the GW1100 firmware, version 2.3.1, enabled the gateway to “see” the WH32 (indoor temp, humidity and pressure) sensor instead of the built-in T&H sensors in the GW1100, and made that option the default. After the update, the calibration settings for the barometer appeared to change on their own, and the resulting reported barometric pressure was off. A quick search of the great posts on wxforum revealed this was a known issue and provided a fix: disable reading the newly-available WH32. Done!
Technical Difficulties
Due to operator error (that’s me!), the weather page was wrong from around 730 pm EST last night until 6:20 am EST this morning. The site generator was inadvertently set to “Simulator” mode. We regret the error.
Cleaned WH41 PM 2.5 sensor
When reviewing my annual logs, I noticed the PM 2.5 detector, an Ecowitt WH41 model, dropped off in readings in August, and was pretty consistently low for the rest of the year. I visited the WxForum and found an article suggesting that the air chamber could be clogged and needed regular clearing. Sure enough, I took the device apart and found debris in side, and what might have been a thin leaf stem stuck in the fan. Cleaned and blown out, the device returned to service, readings immediately started to show more interesting behavior. Hope that fixes it!
