# Aquifer regionalization: hand-drawn condition evidence affects model properties

**Yes: the coarse KGZ search polygons were used to position the boundary between two aquifer-property zones. Their influence is not confined to the display overlay.** The cross-border expansion also assigns a different uniform property recipe outside the inherited footprint. These two dependencies should remain separate in interpretation.

## Source chain (read)

- `scripts/build_gwfd_screening_zones.py:98–113`: `wedge_out_envelope` unions `GW_UPWELL_01` and `GW_UPWELL_03`, applies a 1.5 km morphological opening, takes southern positions in 2 km bins, and smooths them. This is a geometric transformation of condition evidence, not an independently mapped aquitard contact.
- `scripts/build_gwfd_screening_zones.py:132–138`: candidate groundwater polygon `GWB_EVIDENCE_01` supplies the outer footprint; the smoothed envelope divides Zone 1 from its northern remainder.
- `scripts/build_gwfd_screening_zones.py:150–172`: the Chu line, a 2.5 km buffer and fixed easting cuts split out Zone 3 and leave Zone 2. Thus not every zone edge comes from the hand-drawn polygons, but all three zones are part of that screening construction.
- `docs/records/model-domain-review-2026-09-23/lowland_mf6_candidate/run.py:71–101`: rasterizes the three polygons and keeps sufficiently large components as `data['old']`.
- `docs/records/model-domain-review-2026-09-23/lowland_mf6_candidate/run.py:224–243`: applies the three inherited Kh, Kv, thickness and storage recipes only inside `old`; expanded cells keep the uniform defaults. This is an inherited-footprint switch, not an independent cross-border geological delineation.
- `scripts/report_natural_seepage_candidate.py:15–18`: current optional map overlay reads the raw MCHS flood-risk GeoPackage. User clarification now explicitly limits it to rough reconnaissance guidance.

## Recomputed identity and current inputs (measured)

Command: `PYTHONPATH=diagnostics/coupled-improvement-2026-09-14/scratch/runtime .venv/bin/python docs/records/domain-seepage-investigation-2026-09-28/outlet_evidence/probe_aquifer_regionalization.py`.

The raw MCHS overlay and release condition-evidence layer have union symmetric-difference area **0.0 m²**. Recomputing the envelope from the release polygons gives **0.0 m** separation from the stored envelope. Reconstructed Z1, Z2 and Z3 each have **zero** symmetric-difference area from the stored polygons. This confirms the actual geometry chain, rather than inferring it from file names.

Recomputed `old` contains **6286 cells**. Property recipes reproduce native DIS/NPF inputs: maximum thickness difference **1.14e-13 m** (floating-point rounding), Kh/Kv differences zero. Input hashes are in `aquifer_regionalization_receipt.json`; the cell assignments are in `aquifer_zone_cells.csv`.

| Country | Expanded uniform default | Zone 1 | Zone 2 | Zone 3 |
|---|---:|---:|---:|---:|
| KAZ | 10219 | 0 | 0 | 0 |
| KGZ | 184 | 2520 | 3197 | 569 |

The source evidence attributes describe manual digitisation from MChS atlas page 75, pending boundary review and possible shallow groundwater; they explicitly state that the polygons do not independently prove artesian head or upwelling at every location (measured source attributes retained in the receipt). These are not surveyed hydraulic-property contacts. The stored zone status is `screening_not_adopted`; that metadata label does not undo their measured consumption by the native model.

## What is supported, and what is not

**Read literature synthesis, not independently reverified primary literature in this probe:** `docs/phase2/chu_groundwater_fd_aquifer_conceptualization_2026-08-11.md:66–99` supports a broad fan-recharge / artesian-transit / river-terrace conceptual sequence. Lines 179–202 provide zonal transmissivity priors and literature ranges. `docs/phase2/chu_groundwater_fd_gw4_spike_stage1_2026-08-11.md:19–31` records the simplifying layer recipes. These support hypotheses about contrasting materials and structure, not the precision of the implemented boundary, not a property jump at the country boundary, and not site-specific values at every cell.

**Inherited modelling choices:** uniform within-zone properties, the representation of a multilayer aquifer with two layers, the exact smoothed polygon boundary, river-corridor width and easting rules, and the uniform expanded-domain recipe. The regional literature cannot be used to certify these detailed choices without spatial support.

## Practical consequence

A common surface-seepage rule alone cannot demonstrate whole-domain physical consistency while these inherited property assignments remain. The candidate may preserve them to isolate the drainage experiment, but must disclose them as shared structural uncertainty. Do not extend the KGZ polygon-derived zones into KAZ, move the search polygons to fit heads, or smooth properties merely to remove the visible seam.

The next geological qualification should locate independent lithology/stratigraphy, borehole screens/logs, cross-sections and aquifer-test support on both sides; place property transitions using those data and compare plausible uncertain geometries. Continue to distinguish shallow groundwater from confined pressure. Uncertain boundaries should be tested, not treated as observed contacts.

A related historical scoring dependency is read in `scripts/gwfd_daily_coupled_loop.py:343–363`: it evaluates shallow/artesian occurrence against these polygons and their buffer. This probe does not establish whether that historical scorer controls the current experiment. It should not be reused as independent observational validation after the user's clarification; otherwise the same weak spatial evidence influences both model construction and its apparent fit.

No native model, source geometry or parameters were changed. This record identifies an input-meaning limitation that affects numbers; it does not claim that all literature-supported regional contrasts are false or that the correct replacement has been established.