Vol. I · Field Data — Golf
Published Q3 · MMXXVI
Vol. I · Field Data — Golf
Published Q3 · MMXXVI
Two golf sites, anonymised to region level. On one, three years of imagery separate structural drift from a single-season episode — two different repair bills. On the other, the irrigation system poured more than twice the water the turf asked for — and the data can prove it.
What this page is
Modelled figures are labelled modelled and show their arithmetic.
We compared the same 24.0 hectares of playing surfaces across three seasons — every 10-metre pixel of 2021 against its 2024 counterpart, using foundation-model representations rather than a single vegetation index. The course's 76 mapped playing zones were graded by how far their signature moved; zones smaller than 20 satellite pixels were excluded from conclusions, leaving 18 statistically reliable zones.
The result splits cleanly by construction, not by maintenance: fairways — native clay push-up construction — moved a mean 0.200 on the change index, while USGA-specification sand-built greens (0.173) and tees (0.175) held their signature. The engineered sand profiles are doing their job; the clay fairways are structurally drifting. The operator's own reported symptom — dry in drought, waterlogged after rain — is the ground-level signature of exactly this soil behaviour.
The second finding is the more valuable one. A hole flagged as stressed in this season's thermal imagery does not appear among the top three-year changes. Its problem began in 2026 — an episode with a findable cause, likely irrigation coverage — not accumulated degradation calling for reconstruction. One data layer, two very different repair bills.
Case A · verified numbers
Change index = distance between foundation-model representations of the same pixel in two years, public 10-metre satellite imagery. Zones under 20 pixels excluded; edge pixels of narrow fairways partially capture rough.
A trend and an episode look identical on the ground in any single season. Only a multi-year layer can tell them apart — and they carry different price tags.
On a 4.5-hectare instrumented section of a Dubai course, we compared two satellite-derived series over eleven weeks of late spring: the cumulative irrigation water that actually reached the root zone (inferred from soil-moisture response), and the volume the turf required over the same period. The gap is the over-irrigation — not a modelling assumption, but the difference between two measured series.
Applied exceeded advised by 115% — an excess of 10,563 cubic metres on 4.5 hectares between April and mid-June. The arithmetic in dirhams is shown in the panel, labelled modelled: at a reference AED 2.50 per cubic metre it is AED 26,407 for one section in one part of one season. The same intensity scaled to a typical 35-hectare irrigated course models to roughly AED 206,000 per season.
Nothing was broken on this course. Sprinklers ran, turf looked acceptable, bills were paid. That is precisely the point: over-irrigation at this scale is invisible to visual inspection and invisible on a monthly invoice. It only exists as a number when applied and required are measured separately — and it is the number a superintendent can act on line by line.
Case B · numbers & arithmetic
Water volumes are satellite soil-moisture-response estimates, not flow-meter readings. Dirham figures are modelled against a reference rate — actual unit cost depends on source (potable / TSE / well); see our tariff explainer. Linear scaling assumes similar irrigation practice across the course.
The distance between learned representations of the same ground pixel in two different years, computed with a geospatial foundation model over public satellite imagery at 10-metre resolution. A higher value means the surface's spectral-temporal signature moved further between seasons — structural change in the turf system, independent of any single vegetation index. Site median here: 0.197 between 2021 and 2024.
Construction. Greens and tees are built to USGA specification on engineered sand; the fairways are native clay push-up. The sand held its signature (0.173–0.175); the clay drifted (0.200). The operator's own symptom — dry in drought, waterlogged after rain — is classic clay soil-water behaviour.
A trend accumulates across seasons; an episode is a single-season stress with a local, findable cause. The hole stressed in 2026 thermal imagery is absent from the three-year top list — so the fix is diagnosing this season's cause, not budgeting reconstruction. The distinction only exists if a multi-year layer exists.
Satellite soil-moisture response: cumulative water that actually entered the root zone versus the volume the turf required over the same period. On the 4.5-hectare section, applied exceeded advised by 115% — 10,563 m³ of excess in eleven weeks. An estimate of water reaching the turf, not a flow-meter reading.
Modelled, arithmetic shown: 10,563 m³ × AED 2.50/m³ = AED 26,407 for one section over eleven weeks; scaled linearly to a 35-hectare course, ≈ AED 206,000 per season. Actual unit cost depends on the water source — potable, TSE or well.
Client identities and exact locations are withheld by policy; locations are given to region level only. The numbers, method and limitations are published in full so the analysis can be judged on its own. Reference conversations are available on request.
Begin
A sixty-day pilot on a single property — no equipment cost. Outcome report at day sixty: measured water trajectory, intervention windows, salinity status, and the consumption delta relative to the prior schedule. Then decide the next step on evidence, not on description.