Sixteen Metres a Year: Why FresnelPath Exists

Sixteen metres a year.

That is how much the Adygene Glacier — in the right arm of the Ala-Archa Gorge, 3,600 m above Bishkek — shrank in area every year between 1977 and 2015. Kyrgyz and German colleagues pulled the figure out of satellite imagery and climate models. If you only follow area, ignoring thickness and volume, the Adygene Glacier is gone by 2100. No snow visible from Bishkek in summer. The social and climatic bill comes due with it.

That number is the headline. The thing I keep thinking about is what sits underneath it.

Models are built on assumptions. Every one of them.

Every climate model — and every AI model dressed up on top of one — is built on incomplete data and a stack of assumptions. Parameterisations, distribution mapping, occasionally a bold guess. The uncertainty is real. It is also, in my experience as a programmer, almost always underestimated by the people writing the code. Each number in those databases has a provenance, and the provenance is rarely a clean API call.

In the case of Adygene, the provenance is a person.

The provenance is a person

A researcher hikes 10–15 km one way, 1,500 m of vertical gain, on steep slopes that turn lethal in early spring and late autumn — risking a fall into a cliff to measure melt by hand, pull data off a meteorological station onto a USB stick, and carry it back down to the institute. In cold months the climb is harder, so the expedition happens at most once a month.

The temporal resolution of “monthly glacier readings” is not a modelling choice. It is a function of how often a human body can survive the trip.

There are second-order problems too. Batteries underperform at altitude. Some days there is not enough sun for the sensors. Geostationary satellite communication exists — and on the budget of a Central Asian research institute, it doesn't. In practice, the back-haul is a flash drive in a backpack.

This is the gap FresnelPath was built around

LoRaWAN — 868 MHz in EU868 / RU864 regions, sub-dollar-per-month operating cost, chirp-spread modulation that decodes below the noise floor — can move that data off the mountain without a satellite link and without a monthly expedition. If the path closes.

And in alpine terrain, the path almost never closes the way a spreadsheet says it does.

FSPL is comfortable. The ridge at km 7 is not. A 60% Fresnel zone clearance check, ITU-R P.526 diffraction over the actual GLO-30 DEM, P.1812-6 with real clutter and time-percentage parameters — those are the differences between “the math worked on paper” and “the gateway hears the sensor.”

Run that analysis before you commit hardware, before you commit a 1,500-metre climb. See where the obstruction is. See whether moving the gateway 200 m up the ridge clears it, or whether you need a relay. See whether the panel orientation you planned actually survives winter horizon-shade at that latitude. None of that requires a flight. It requires elevation data, the right propagation model, and an honest verdict.

Adygene is the smallest version of a very large problem

The reason I keep coming back to Adygene is that it is the smallest version of a problem that is everywhere: glaciers, forest fires, avalanches, livestock corridors, snowpack stations, off-grid clinics, anti-poaching telemetry, water-quality sensors in mountain reservoirs. The instrumentation is cheap now. The field labour is not. The satcom is not.

The bottleneck is whether somebody planned the link honestly before a researcher carried a gateway up 1,500 m to find out it doesn't work.

That is the boring engineering problem we are trying to make routine. So that the next sixteen-metres-a-year measurement does not depend on a person willing to risk the cliff.

If you operate in this kind of terrain — or fund people who do — that is what FresnelPath is for.