The engineering story behind spatial detection, told for pilots who want to understand how it actually works (without needing a PhD).
A pilot asked me recently: "How can you possibly detect a thermal before my vario does? What kind of black magic is this?"
No magic. Just physics, good sensors, and asking a different question than everyone else has been asking for the past 50 years.
Let me show you how it works.
The Question Nobody Asked
For decades, everyone in the soaring world has been trying to build better variometers. Faster barometric sensors, more sophisticated filtering algorithms, better total energy compensation. Millions of dollars of development, thousands of engineering hours, all focused on one question:
"How can we detect vertical motion faster?"
And they've done amazing work. Modern varios are incredible pieces of technology. But they're all still fundamentally limited by the same constraint: you can't detect vertical motion until there IS vertical motion.
We asked a different question: "What if we don't wait for vertical motion at all?"
The Geometry That Changes Everything
Think about the basic geometry of thermal flying. Your wing is twelve meters wide. A typical thermal is anywhere from 300 to 1,000 meters wide. You're one percent the size of what you're trying to detect. Maybe less.
This means something crucial and completely unavoidable: you can't enter a thermal all at once. It's geometrically impossible. One wingtip has to cross the boundary first.
At typical cruise speed, around 45 kilometers per hour, it takes about one second for your entire wingspan to cross into a thermal. During that second, one side of your wing is in rising air experiencing higher pressure, and the other side is still in normal air.
That's an asymmetric condition. And asymmetry creates a measurable difference.
Traditional varios ignore this completely. They're waiting for your entire aircraft to start moving vertically. But for that crucial first second, there's already a signal available if you know where to look.
We look at the pressure difference between your left and right wing cells.
The Sensor Challenge
Okay, so we want to measure pressure inside your wing cells. How sensitive do the sensors need to be?
A moderate thermal, say three meters per second, creates a dynamic pressure change of about five Pascals when one side of your wing enters and the other side hasn't yet. That's the signal we're trying to detect.
Five Pascals. For context, that's about the pressure of a grain of rice resting on your fingertip. It's tiny.
For a long time, pressure sensors couldn't reliably measure changes that small. They were either too noisy, meaning random fluctuations would overwhelm the signal, or too slow to respond, or too big and heavy and power-hungry to put in a flying instrument.
Then MEMS technology caught up. MEMS stands for Micro-Electro-Mechanical Systems. These are tiny pressure sensors, literally two millimeters square, that can detect pressure changes as small as 0.4 Pascals. That's about one-tenth the signal we're looking for.
The signal-to-noise ratio is good. The sensors are fast, responding in milliseconds. They're tiny and lightweight. And they draw almost no power.
This is the same technology that's in your smartphone's barometer, refined and optimized for our specific application.
Where We Put Them
We spent months in simulation and wind tunnel testing figuring out the optimal sensor placement. Put them in the wrong spot and you're measuring noise instead of signal.
The sweet spot turned out to be 65 percent of the way from the center of the wing toward the tips. This is where asymmetry is strongest during thermal entry but you're far enough from the wingtip vortices that you're not getting overwhelmed by turbulent air.
We place them at about 30 percent chord from the leading edge, in the upper surface cells. This is near the pressure recovery region where internal cell pressure is most responsive to external airflow changes.
Four sensors total. Two on the left side, two on the right side, with slight fore-aft offset for validation. If both sensors on one side agree and both sensors on the other side agree, we have high confidence. If there's disagreement, we know to be more cautious.
The whole sensor package weighs about 20 grams and fits inside the cell without affecting flight characteristics at all.
What We Actually Measure
Every hundredth of a second, we're reading pressure from all four sensors plus data from an inertial measurement unit that tracks acceleration and rotation.
The pressure readings tell us what the air is doing. The IMU tells us what the wing is doing. Together, they paint a picture of what's happening.
We're not just looking at absolute pressure. We're looking at pressure asymmetry between left and right. We're looking at the rate of change. We're looking at variance, which spikes at thermal boundaries where turbulence increases. We're correlating pressure changes with wing movements to distinguish real thermals from gusts and pilot inputs.
It's a multi-feature detection system. No single sensor or single calculation determines whether a thermal is detected. We're looking at the whole pattern.
When your right wing enters a thermal, here's what we see: right-side pressure drops by several Pascals while left-side pressure stays constant. Asymmetry signal jumps. Variance increases as you cross the turbulent boundary. The wing tips slightly right as that side experiences more lift. The gyroscope picks up the rotation.
All of these signals arrive within a tenth of a second of crossing the thermal boundary. Long before you feel anything. Way before your vario knows anything is happening.
From Signal to Beep
Once we have the raw sensor data, we need to turn it into something useful. This is where the algorithm comes in.
We calculate dozens of features from the raw data. Pressure asymmetry and its rate of change. Variance ratios. Gradient calculations. Turbulence energy from the accelerometer. Rotation patterns from the gyroscope.
Then we weight these features based on what we learned from thousands of hours of real flight data. Asymmetry is the strongest signal and gets the most weight. Variance is also very reliable. Turbulence and rotation help confirm but aren't decisive on their own.
The algorithm outputs a thermal confidence score between zero and one. Above 0.65, we're confident there's a thermal and we alert. Below 0.45, we're confident there isn't. In between is a hysteresis zone where we maintain whatever state we were in before.
This prevents oscillation. You don't want beeping on, beeping off, beeping on as you hover around the threshold. Once we've alerted, we need a clear signal that you've left the thermal before we go quiet again.
The whole process, from sensor reading to alert decision, takes about one-tenth of a second. Compare that to the two to three seconds your traditional vario needs.
Dealing with the Real World
Lab conditions are easy. Everything is controlled and predictable. Real flying is chaos.
You're dealing with turbulence, gusts, wind shear, pilot inputs, wing movements, temperature changes, and atmospheric noise. The algorithm has to distinguish a real thermal from all of that.
This is where machine learning helped enormously. We flew test pilots with prototype systems for hundreds of hours, recording everything. Pressure, IMU, GPS, and most importantly, pilot annotations. Every time a pilot encountered a thermal, they marked it in the app. Every time they hit turbulence that wasn't a thermal, they marked that too.
This gave us thousands of labeled examples. Real thermal entries. Real false positives. Real marginal cases. We used this data to train the detection algorithm to recognize the patterns that matter and ignore the patterns that don't.
Detection accuracy in normal conditions
The current system has about 91 percent accuracy in normal conditions. Strong thermals over three meters per second are detected correctly 97 percent of the time. Moderate thermals between two and three meters per second are detected about 92 percent of the time. Weak thermals between one and a half and two meters per second are detected about 75 percent of the time.
The remaining errors are mostly false negatives in very weak or very broken thermals, and occasional false positives in strong turbulence. We're continuously improving this with more data and algorithm refinement.
The Advance Warning You Actually Get
So how much warning do you actually get in practice?
We've measured this extensively. The answer depends on thermal strength and your flight speed, but typically it's one and a half to two and a half seconds before your traditional vario alerts.
For a weak thermal around one and a half meters per second, you get about one to one and a half seconds warning. For moderate thermals around three meters per second, it's closer to two seconds. For strong thermals over five meters per second, it can be two and a half seconds or more.
Advance warning before traditional vario alerts
Why the variation? Stronger thermals create larger pressure signals that we detect more confidently and quickly. Weaker thermals create smaller signals that take slightly longer to confirm. But even in the worst case, weak thermals, you're still getting a full second of advance warning.
At cruise speed, one second is 12 to 15 meters. Two seconds is 25 to 30 meters. That's the difference between centering the core and overshooting it.
Directional Information: The Game Changer
Here's what really changes the game though: because we're measuring left versus right pressure, we know which direction the thermal is.
If your right wing hits first, the core is to your right. If your left wing hits first, the core is to your left. It's that simple.
Or we can show it visually on screen with a simple indicator showing asymmetry direction and magnitude.
This transforms the scratching experience. You're not searching for the core anymore. You know where it is the moment you touch the thermal boundary. Turn that direction, chase the stronger signal, and you're centered.
What About Battery Life and Weight?
The whole system runs on a 500 milliamp-hour battery that weighs about 15 grams. With the sensors, microcontroller, buzzer, and optional display, total weight is about 45 grams. That's lighter than most varios and it includes the battery.
Runtime is 10 to 12 hours of continuous operation. That's a full day of flying with margin. Charge it overnight via USB-C and you're ready for the next day.
The system draws so little power because modern MEMS sensors are incredibly efficient and the microcontroller spends most of its time in low-power modes, only waking up fully when needed.
The Beta Program Reality Check
We're currently testing with 50 pilots across different sites and conditions. Alpine flying with strong thermals and complex terrain. Flatland soaring with weak thermals and long XC distances. Coastal sites with sea breeze and convergence.
The feedback has been revealing. The technology works. The physics is sound. The advance warning is real and measurable.
But we're also finding edge cases we didn't anticipate. Certain wind shear conditions create false positives. Very broken thermals are harder to confirm. Some wing types generate more internal pressure variance than others just from normal flying.
Each of these discoveries leads to algorithm refinement. The system gets smarter with every flight, every data upload, every pilot annotation.
We're also discovering that pilot psychology matters more than we expected. Some pilots love the early warning and use it aggressively. Others find it distracting and prefer more conservative alerts. The system needs to adapt to different flying styles.
That's why we're building in adjustable sensitivity and alert modes. Aggressive mode for when you're low and desperate. Conservative mode for when you just want confirmation of obvious thermals. Customizable to match your flying style and risk tolerance.
What's Next
The current system works. Pilots are using it successfully in real flights. But there's room for improvement.
We're working on GPS integration for thermal mapping. Imagine every thermal you encounter being logged with location, strength, and time. Over months and seasons, that builds a database of where thermals form at different sites under different conditions. Predictive thermal detection based on historical patterns.
We're exploring pilot-to-pilot networking. When one pilot finds a thermal, nearby pilots get an alert. Not just "someone found lift" but actual location and strength data. Collaborative thermal hunting.
We're refining the machine learning models with neural networks that might squeeze out another few percentage points of accuracy. The question is whether the improvement justifies the extra computational cost and power draw.
And we're always working on making the hardware smaller, lighter, cheaper, and more reliable.
The Engineering Philosophy
Here's what guides our development: we don't add complexity for its own sake. Every feature has to provide real value to pilots in real conditions.
Spatial detection provides measurable advance warning that translates directly to better thermal centering and more altitude gained. That's valuable. So we build it.
Fancy graphical displays that look cool but don't actually help you fly better? We skip those. Unless there's a clear benefit, we keep it simple.
Why It Matters
At the end of the day, this isn't about the technology. It's about expanding your flying envelope.
It's about being able to work weak conditions that used to frustrate you. It's about making it home on marginal days instead of landing out. It's about detecting turbulence before it hits you instead of reacting after the fact.
The physics is on our side. The geometry of thermal entry creates signals that can be detected earlier than anyone has been detecting them before. Modern sensors are good enough to measure those signals. Modern processing power is good enough to interpret them in real-time.
The engineering challenge was figuring out how to put all of that together into something small, light, reliable, and actually useful in real flying.
We think we've done that. But ultimately, pilots will decide whether this changes flying for them or not.
The technology is here. The question is what you do with it.
Want to help refine this technology? We're looking for experienced pilots to join the ParaBaro beta programme. You'll get early access to the hardware, direct input into algorithm development, and the chance to shape the future of thermal detection.
Just want to stay updated? Create an account and we'll let you know when ParaBaro is ready for wider release.
This is part of our blog series on thermal physics and detection technology. Catch up on the earlier posts: The Invisible Giants You're Flying Through, Your Vario Is Lying To You, and The Lost Art of Scratching.