Field Notes from the Terskey-Alatoo: A Gateway at 3,360 m, and Why Planning Comes Before Climbing

Climbing 700 metres on a 55-degree slope with solar panels, heavy batteries, and an IoT gateway strapped to our backs is not the usual picture of innovation. You wouldn't think of AI in this context at all. And yet this is exactly what it took to bring a climate-monitoring communication network to one of the most remote, highest-altitude places on Earth.

Our team recently completed the installation of a LoRaWAN gateway near the Kara-Batkak (3,360–4,800 m) and Aylama glaciers in the Terskey-Alatoo range. The gateway is now listening for affordable climate sensors deployed across high altitudes — places where ground-truth measurements are missing not only in Kyrgyzstan or across the Third Pole, but globally.

What the data layers don't show you

Most foundational climate AI models today fill the gaps with assumptions — CDF mapping, bias correction, reanalysis interpolation — because ground data from mountainous regions simply doesn't exist. The cryosphere — glaciers, snow, permafrost — remains under-observed, particularly in the world's high mountain ranges.

The Third Pole Regional Climate Center reports that in the entire Pan-Third Pole region, only 28 monitoring stations sit above 3,000 m out of roughly 700 in total. These are exactly the altitudes where the most dramatic changes are taking place.

And the stakes are not abstract. According to ICIMOD, even in a world that warms by just 1.5–2°C, Third Pole glaciers could lose 30–50% of their volume by the end of this century. Without precise, real-time, high-resolution data, AI models risk underestimating the hazards — permafrost instability, glacier detachments, glacial lake outburst floods, snow droughts. Every additional ground station shrinks that error bar.

The planning question we faced before the climb

From the trailhead, the question wasn't "will the gateway work?" — Semtech gateways at SF12 with a good antenna will hear sensors at extraordinary range when the path is clear. The question was: which ridge?

The local geometry of Terskey-Alatoo offers more than one plausible site. A saddle 800 metres east of where we eventually placed the gateway looked, on paper, like a shorter climb. But "on paper" can mean two very different things. A line drawn on a 2D map doesn't tell you what the radio sees.

This is the same problem we wrote about in the Andes: a Free-Space Path Loss calculation can return a perfectly comfortable link margin while ignoring a 40-metre ridge that obstructs the first Fresnel zone at the path midpoint. The link budget is right. The path isn't free space. The mountain wins.

So the workflow before the climb looks like this: drop the candidate gateway pin on the map. Drop the planned sensor positions on the moraine and at the glacier tongue. Run the terrain-aware profile — not FSPL, the full ITU-R P.1812 / Deygout-routed analysis on 30 m Copernicus GLO-30 elevation data. Look at the Fresnel zone, not just the line-of-sight. Read which diffraction method fired, and what extra dB it cost. Repeat for the alternative saddle. Then decide which ridge gets the equipment.

A second climb to relocate a gateway costs the project roughly half a field season. You only get one summer above 3,000 m. Planning the path before carrying it is not an optimisation. It's the difference between fifteen installations completed and seven.

Why precision matters — in RF and in climate, for the same reason

This is the part where the two halves of this story rhyme. A climate model that smooths over a 30–50% volume-loss tail because it has 28 high-altitude stations instead of 280 will produce a confident, wrong forecast. An RF link budget that smooths over a ridge because it uses FSPL instead of Deygout will produce a confident, wrong "go." Both failure modes look identical from outside: a number with a comfortable margin, derived from a model that didn't see what it needed to see.

Precision down to the third or fourth decimal matters for both. In climate science, it's the difference between catching a glacial lake outburst signal and missing it. In RF planning, it's the difference between a Fresnel clearance of +2 m and −3 m — between a gateway you mount once, and one you climb back up to move.

FresnelPath was built for exactly this class of decision. Pick the country, place the points, read the profile, see which obstacles the model selected, get the verdict before you load the pack.

The climb, honestly

Summertime is field season — the only window when these passes are reachable and the snowpack is thin enough to work in. For me, the climb was exhausting. Like everyone else on the team, I slipped through wet meadows, fell, got back up, and kept walking. At 3,000 m, looking down at the rivers carrying snowmelt into the irrigation systems of two countries, the view did what views do: it reminded me that innovation does not only happen in labs, conference rooms, or training runs. Sometimes it begins with sweat, altitude, and determination — carrying heavy equipment up a mountain so that the world can better understand what is at stake.

This was the second of fifteen installations planned under a research project funded by the Internet Society Foundation and implemented by the founder of FresnelPath, Aziz Soltobaev, as part of the Internet Society Kyrgyzstan Chapter projects. Thirteen more sites are ahead. Each of them is a saddle, a ridge, or a moraine where the difference between a path that closes and a path that doesn't will be decided on a screen weeks before anyone laces up boots.

For teams doing similar work

If you are deploying climate, hydrology, or hazard sensors in high-mountain terrain — Third Pole, Andes, Caucasus, Alaska Range, anywhere with vertical relief over 1,000 m — the workflow that worked for us:

1. Drop candidate gateway pins for every plausible ridge. Not just the one that looks easiest.
2. Drop the actual sensor positions. Glacier tongue, moraine, weather station mast — wherever the data has to come from.
3. Run the terrain-aware path profile for each pairing. Look at the Fresnel zone, not just LoS.
4. For the candidates that pass, look at which diffraction method fired. Multi-ridge paths in alpine terrain are almost always Deygout territory; that is the additional loss your spreadsheet didn't see.
5. For the gateway site, check the solar horizon-shade profile against the climbing season's irradiance. A perfect link at a site that loses three winter months to ridge shadow is still a failed deployment.
6. Only then schedule the climb.

The cryosphere is running ahead of our instruments. The least we can do is not waste a trip up.