| yesteryear forever |

MOUND — A Backyard Sensor Array

MOUND deployed in the garden
MOUND deployed in the garden.

MOUND is a small solar-powered sensor array watching the backyard from a post in the garden. It logs temperature, humidity, pressure, light, soil conditions, and its own battery health every ten minutes, and after dark, a little microphone under the garage eave listens for bats — not because the data is particularly useful, but because paying attention is its own reward. Simplicity is deliberate; ambiguity is a feature.


live — Central Ohio  ·  loading...

--   ·   --   ·   --

→ full sensor data, graphs, and archive
→ every bat it's heard


Deliberate Simplicity
We're inundated with powerful technology nowadays; the glass computers we carry in our pockets are practically indistinguishable from magic. But this project isn't about pushing humanity forward. To the contrary, it's about slowing down and paying attention. It's about integrating clever, human-made tools with nature as a deliberate way to observe and document... and simplicity is the point.

I could've spent ~$1,200 on a pre-built weather station, or even just downloaded a weather app on my glass pocket computer, but the journey's at least partly the point. The other part is interpretation and observation as a sign of respect and admiration.

Original version of MOUND
Original version of MOUND.

Basic Avatars
I've always loved the idea of human space flight and exploration, but as I've gotten older, I've grown to love our robot surrogates even more. Unfortunately, human space travel is insanely difficult and mostly impractical. We can send robots in our stead to explore quicker, faster, cheaper. The downside, though, is comparatively limited data acquisition; we only get a select few data points to collect and scrutinize depending on the onboard sensors. In order to build a complete picture, we must extrapolate...

Ambiguity as a Feature, not a Bug
The scariest monster isn't the one shown on screen, it's the one in your imagination. The corporate weather app or the fancy store-bought weather station would've provided much more detailed information, but ambiguity is intriguing — who doesn't love a good mystery?! Creating a story out of scraps of data is so much more fulfilling. A picture from the surface of Mars tells us SO much about the environment and what it would be like to visit and explore, but a word-picture is richer and more compelling; it's why the book is always better than the movie. MOUND's simple text data files and graphs are deliberate, and only half the story. The other half is the imagined world MOUND inhabits.

Latest version of MOUND
Latest version of MOUND.

Imagined Reality
Think of the last nature documentary you watched or a memorable National Geographic photo... those are real places. Those places actually exist. Pick out a rock in the foreground — that rock is probably still there! Imagine sitting next to it: Maybe it's early evening... sun setting, light breeze, quiet. Pick it up, feel the weight in your hand. The dirt is warm and comforting. Scan the horizon. Notice the patterns in the clouds and the trees in the distance. The rock experienced this for eons and will continue to lie there patiently for eons after you've left.

Those dark scenes at the bottom of the ocean with the roaming crabs on the whale carcass are just as real and are playing out right now. So are the cliffs of the Valles Marineris — the Grand Canyon on Mars. A rock just slipped from the top and tumbled down even though no one was there to hear it. It still happened.

The select few data points from this humble weather observatory set the stage for a real place on an actual planet in a solar system in a genuine galaxy.
“That's here. That's home. That's us.” ...and these sparse data points from MOUND start to tell the story of this tiny, remote outpost here on Earth.


the build

Practicality vs. Play
I could have easily thrown all these components in a box where everything is accessible with plenty of room for upgrades and adjustments, but where's the fun in that? I love practicality, but art and play is where the soul of life lives. Instead, I had Tatooine mechanisms on my mind while designing the enclosure; weird devices with questionable purposes integrated with the planet because... it's cool! We all need a little more whimsy in our lives.

MOUND prototyping and testing
Prototyping and testing.

Intelligent Evolution
I know. It's messy. It's technically "poorly designed". Why would I wire that there? Well, I'm a tinkerer, not a designer. I build and create and add... and whatever comes out at the end is anyone's guess. Building is more fun that way! I like to think of it almost like gestation; parts get added and tested, then things get left like vestigial organs. It infuses character and uniqueness instead of boring standardization. Feels more alive, and "mine".

What's in a Name?
The names of a few of my previous projects were inspired by indigenous peoples of this area: Adena & Hopewell. While brainstorming to continue the trend, I realized the street I live on is called "Mound", as in, effigy mounds built by Native American cultures. Seemed too fitting to pass up.

These people — real people with stories and experiences of their own — were naturalists too. They built mounds to observe the sky, to mark seasons, to pay attention to the land. MOUND is doing something quietly similar — sitting on that same Ohio earth, watching the sky, and marking time. The data files are just a different kind of mound... layers of accumulated observation, one reading at a time.


what it sees

On June 11th, 2026 — one of the first full days of uninterrupted logging — a storm rolled through Central Ohio in the early evening. MOUND watched the whole thing unfold in real time:

Temperature dropped sharply from 90°F down to 82°F right at the rain event.
Humidity spiked dramatically — jumping from ~45% back up to 67% instantly.
Pressure plummeted to its lowest point of the day right at that moment — the storm system arriving.
Light — already near zero as the storm clouds rolled in.
Soil temperature — barely flinched, all that thermal mass absorbing the shock.
Soil moisture — dry all day around 15000, then the rain hit and it shot up to 18000+ instantly and kept climbing.

That's a complete weather event captured in six sensor channels. You can literally watch the storm arrive, hit, and the aftermath unfolding in real time.

MOUND sensor graph — June 11, 2026
June 11, 2026 — six sensor channels on a shared time axis. Storm hits at 20:00.

what it hears

On October 6th, 2026, at 7:40 in the evening, something flew past the garage and said hello about ten times a second. Nobody in the house heard a thing. MOUND did.

The next night, at 7:40 — same bat, same commute, right on schedule.

Spectrogram of the first bat MOUND heard, October 6, 2026
The first bat. Time runs left to right, pitch bottom to top. Each stripe is one call.
The same bat, slowed down ten times. Those chirps are its sonar.

A Sound Too High to Hear
Bats see with sound. They shout and listen for the echo, the way a ship pings sonar off the sea floor, and they do it fast enough to snatch a moth out of the dark. The catch is pitch: human hearing tops out around 20,000 vibrations a second (20 kHz) on a good day, and that gets lower every birthday. Ohio's bats shout somewhere between about 20 and 50 kHz. The whole backyard could be full of screaming hunters on a summer night and it'd sound like crickets and a car somewhere far off.

To catch a sound, you have to sample it at least twice as fast as its highest note. That's a rule with a name (Nyquist) and no wiggle room. MOUND's microphone samples about 146,700 times a second, so in theory it hears up to about 73 kHz, well into bat territory. In practice, the cheap mic gets hissy above 30 kHz or so and drowns out completely past 50. Ambiguity, as ever, is part of the deal.

Reading a Spectrogram
Since nobody can actually hear these recordings, MOUND draws them instead. A spectrogram is a picture of sound: time runs left to right, pitch runs bottom to top, and brightness is loudness. A bat call shows up as a short, steep swoosh sliding down from high to low, a chirp, really, repeated over and over as the bat closes in. A katydid looks like a fuzzy smear. Hiss looks like snow. Once you've seen a few, you can spot a bat from across the room.

The Ear
The microphone is a ridiculous little thing: an SPH0641 MEMS mic, smaller than a pencil eraser and cheaper than lunch, the same kind of part that lives inside phones. Unlike phone software, though, it doesn't throw away everything above human hearing. It's wired to the same Raspberry Pi that serves this website, which was already sitting there in the garage with nothing better to do at night.

The first recordings were mostly static. The fix came from an unlikely place: a scrap of Cat6 network cable. Pairing each signal wire with its own ground on a twisted pair and the noise just... went away. Old tricks are the best tricks.

Bats Come in Bursts
Every ten minutes through the night, MOUND records two minutes of ultrasound and hands it to BatDetect2, a neural network trained to pick bat calls out of noise. It's clever, and it's also easily fooled. The first nights, it confidently reported bats that were actually katydids, crickets, dew dripping off the gutter, and plain microphone hiss.

The fix wasn't a smarter computer; it was paying attention to the bats. A hunting bat doesn't call once. It calls five to ten times a second, so a real bat always shows up as a burst of confident detections, while noise shows up as lonely one-offs. New rule: no burst, no bat. Checked against every clip from the first three nights by eye, the rule caught 10 of 12 real bats and was fooled by exactly zero of 175 bugs, clicks, and hisses. The two it missed were faint and far away. I'll take it.

Who's Out There?
Here's the fun part: the software was trained on British bats. Every bat it hears gets a confident British name, which is a bit like asking a Londoner to identify Ohio birds. So MOUND sorts them by the one thing it can actually measure: how low the call goes.

The 7:40 regular calls down around 25 to 30 kHz with shallow sweeps; almost certainly a big brown bat, the most common bat in Ohio and the one most likely to be living in somebody's attic. On October 9th, a second voice showed up, higher at 38 to 50 kHz with steep, hooked sweeps. Best guess: an eastern red bat, migrating south for the winter and passing through for a snack. And once in a while something calls lower than even the big browns, which just might be a hoary bat, Ohio's biggest, also passing through in fall.

Could be wrong on all counts. That's fine. I'd rather have a good mystery than a confident wrong answer.

last bat heard  ·  listening...

--

→ every bat it's heard, with spectrograms

Ohio's bats go quiet from about late October until April, either hibernating in caves and attics or long gone south. MOUND will keep listening through the winter anyway, to a whole lot of nothing, waiting for the 7:40 commuter to wake up.


the long dark

For a long stretch this summer, MOUND died every night. Not metaphorically... the batteries would simply run out somewhere between dusk and dawn, and the little station would go dark until the sun found it again, usually mid-morning. Some nights it made it to 2am. Some nights barely past 8pm. A few nights, for no obvious reason, it made it through entirely which was almost more unsettling than the failures, because it meant the system was teetering right on an edge, and I had no idea which side of that edge any given night would land on.

I chased ghosts for weeks. Rewrote scripts. Killed rogue desktop services that had no business running on a headless garden Pi. Fixed a stale SSH connection that could hang forever instead of failing cleanly. Moved a wireless access point twenty feet closer. Every fix was real and worth making, but none of them was the actual answer. The actual answer, it turned out, was stubbornly physical: three overgrown trees were shading the panel through most of the morning and evening, and the battery chemistry I'd chosen simply couldn't survive being chronically, quietly shortchanged, day after day.

There's something fitting about a project built on "vestigial organs" and imperfect gestation having a whole limb that quietly failed every night for a season. MOUND doesn't hide its scars. The fix, in the end, was almost embarrassingly simple; real lead-acid batteries instead of finicky lithium cells, a proper charge controller that actually protects them, and two solar panels facing different directions to catch whatever light made it through the branches. One of those trees has since come down entirely, courtesy of the power company. MOUND hasn't gone dark since.


how it works

All sensors are wired to a Raspberry Pi Zero W mounted inside a weatherproofed enclosure on a pole at the back of the garden. Every ten minutes a Python script reads all sensors and appends one line to a flat text file — no database, no framework, just a timestamp and a handful of numbers. Another script generates graphs using gnuplot and publishes everything to this site automatically.

Sensors: BME280 for temperature, humidity, and barometric pressure. TSL2591 for light and lux. DS18B20 waterproof probe for soil temperature. Capacitive sensor for soil moisture. FC-37 rain sensor. INA219 for battery voltage and charge/discharge current. Raspberry Pi Camera Rev 1.3 for periodic photography — a live shot every half hour, with one kept permanently each day around noon.

Bats: An SPH0641LU4H-1 PDM MEMS microphone under the garage eave, wired over Cat6 to the I2S pins of the Raspberry Pi 4 that runs this website. From 6pm to 7am, every ten minutes, a shell script records six 20-second clips at 146,739 samples a second, runs them through BatDetect2, keeps any clip with a burst of at least two confident calls, and draws its spectrogram with sox. Everything lands in a plain text log, one line per bat.

Power: Two solar panels of different sizes, facing different directions to catch light through the tree gaps at different times of day, feed a PWM charge controller through blocking diodes so a sunlit panel never backfeeds a shaded one. The controller charges a pair of sealed lead-acid batteries and powers the Pi directly off its regulated USB output, set to stay always on... no toggle switch anymore, no manual intervention needed after a power blip. Simple, stubborn, and (finally) reliable.

→ full wiring reference, for the curious


gallery


- At the edges where nature and technology meet, something unexpected happens... not contradiction but a conversation. MOUND is a handful of cheap sensors on a pole in a Central Ohio garden, patiently logging the world in sparse numbers every ten minutes. The data is only half the story. The other half belongs to imagination.

"It's at the edges where the most interesting things appear."


→ live sensor data, graphs, and daily archive

2026-10-20