# Seepage process investigation

Measured by `scripts/investigate_natural_seepage_processes.py` from retained daily native cell series, original RCHA budget records and the pinned DEM. No model parameters were changed or new model simulation performed.

## Findings

Across the changed cells, 70.9% of original April above-ground cell-days have upward flow across the lower-to-upper layer interface. Under the weakest candidate outlet this falls to 21.8%. These are cell-day fractions, not area observed flooded and not flow-volume fractions. The extra high heads cannot all be labelled upwelling: raising the shallow head can suppress or reverse the vertical gradient. Original simultaneous recharge versus positive upward exchange is in `native_recharge_vs_vertical_exchange.csv`. That comparison excludes lateral inflows and storage release and is NOT a complete source attribution or particle tracing.

Using a sloping head surface reconstructed from neighboring final heads changes total full-DEM R1000 potential discharge by -4.3% in April and -33.1% in September, relative to applying one constant cell head. This is a fixed-head sensitivity, not a new MODFLOW run or a proven correction. It demonstrates whether the assumed subcell water table matters; local head gradients, terrain errors and ponding are unresolved. The comparison to native ten-band discharge separately exposes elevation grouping error. Group sums refer only to cells with usable neighboring heads.

The separate [recharge source attribution](recharge_sources.md) traces the prescribed recharge to its source arrays and checks exact agreement with the native budgets. It distinguishes soil-bottom drainage from channel seepage and direct canal leakage.

## Treatment of hand-drawn polygons

Owner clarification in this chat: polygons are crude hand-drawn indications of where to look, assessed only on the KGZ side. They are not confirmed upwelling extents, observed inundation boundaries, or calibration targets. No absence inference is valid beyond their lines, including within KGZ; KAZ is unassessed by this particular source. They are excluded from parameter fitting, scoring, eligibility and the computations above. The earlier inside/outside table is retained solely as historical descriptive output, explicitly not validation.

## Implication

Do not select outlet resistance to reproduce the red-cell pattern or the hand-drawn outlines. Separate source pressure from outlet resistance and subcell geometry. Further physical qualification needs shallow and deeper heads at common locations, recharge evidence, outlet elevations/capacity and connected surface routes. Neither a uniform shallow head within each cell nor the reconstructed slope has been independently confirmed here.