SWAT+ drainage area audit (TxtInOut vs NHDPlus HR) | SWATGenX

This is QA step 3: with the network delineated and each gage assigned to a channel, we verify the assigned channel carries the right drainage area. The area is reconstructed from executable SWAT+ chandeg.con files (what the model actually routes), compared with original NHDPlus HR TotDASqKm, USGS NWIS site drainage area, and the WBD upstream HU12 sum (model context) at each gaged channel. Because a wrong-channel pick masquerades as an area error, gage assignment comes first; read the ratios here on calibration-ready stations so a true SWAT+-vs-NHD area-definition difference is not confused with a mis-assigned gage. Across 98 stations, 97 have both SWAT+ and NHD values; 78 fall within a 0.5–2.0 SWAT/NHD ratio. Agreement is reported as |SWAT+ − NHD| / NHD × 100 at each gage. Large SWAT–NWIS gaps are treated as gage-assignment diagnostics, not proof of NHD-to-SWAT conversion failure.

Models

8

Stations

98

Matched SWAT–NHD

97

Within 0.5–2.0×

78

• SWAT+ column: TxtInOut/chandeg.con column area (hectares ÷ 100), row matched by GIS channel id (v3 NHD-first/SWAT-second assignment for Peace 03100101; stations.shp channel elsewhere)
• NHD column: Original NHDPlus HR HU4 geodatabase flowline TotDASqKm within 500 m of gage (domain-clipped)
• NWIS column: meta_{VPUID}.csv nwis_drain_area_km2 (USGS NWIS site-service drain_area_va / contrib)
• |Δ| vs NHD = |SWAT+ − NHD| / NHD × 100 (same information as SWAT/NHD ratio, expressed as percent difference from NHD).
• GIS channel in stations.shp matches gis_id in chandeg.con; SWAT lcha is the routing index for channel_sd output (they differ when noted).

Portfolio tables above compare executable chandeg.con drainage area (what SWAT+ runs) with original NHDPlus HR TotDASqKm at each gaged channel — not post-processed national pickles. For the large Peace River HUC-8 model we traced where differences enter the pipeline (phases 1–3), using data-backed scrutiny similar to the calibration runtime benchmark experiment log.

Frozen audit chain (Peace mainstem): Original NHD and streams.pkl reach TotDASqKm match; post-processed polygon upstream area is ~+2–3% vs NHD; QSWAT AreaC and chandeg.con are ~+15–17% on mainstem. The offset first appears at the QSWAT/SWAT+ channel-area stage.

Pipeline stages (Peace 10-gage panel)

Stage D uses national watersheds.pkl clipped to Peace HUC-12s — not necessarily identical to per-project QSWAT Watershed/Shapes when those files are missing. Stage H (live rivs1.AreaC / project SQLite) is pending on this workspace; when shapes are restored, re-run the internal 3b script to close the fork: H ≈ F ≫ NHD → QSWAT channel area; H ≈ NHD and F ≫ H → SWAT+ text export.

Each item below was tested with scripts and CSV traces under publication/analysis/qa/ (internal reproducibility). Status reflects weight of evidence on Peace mainstem gages unless noted as a separate outlier class.

Peace River HUC-8: 76 matched SWAT–NHD; median SWAT/NHD ratio 1.1178; 65 within 0.5–2.0×. Values in km² unless noted.

Peace River rows in the portfolio tables above use v3 NHD-first / SWAT-second GIS channel ids when looking up chandeg.con area (same method used for current streamflow extraction), not the legacy da_distance channel in older stations.shp snapshots. Per-gage assignment class, mapping method, and calibration funnel are on the USGS station → SWAT+ assignment page.

Across 8 evaluation models, drainage-area agreement is generally good (78 of 97 matched station pairs within 0.5–2.0× SWAT/NHD). Peace River mainstem gages show a systematic ~10–17% excess in SWAT+ cumulative area vs original NHD VAA that is not explained by comparing to local reach area, using merged NHD artifacts, NHD VAA inconsistency, orphan merge, lake omission, pre-inflated TotDASqKm in streams.pkl, or post-processed polygon accounting alone.

The leading interpretation is a delineation-definition difference between NHDPlus HR hydrosequence cumulative drainage and QSWAT/TauDEM-derived channel contributing area — not necessarily a SWATGenX preprocessing error. That protects the NHDPlus HR method used in model generation while flagging channel-area semantics for calibration interpretation. Assignment outliers and missing chandeg rows remain a separate review class.

These stations are flagged for follow-up (reservoir/auxiliary objects, tributary vs mainstem assignment). They are excluded from the headline SWAT/NHD success metric unless both areas exist.

Oklawaha (FL) (03080102) — 0308/huc12/030801020804

Stations

3

SWAT+ chandeg area

3

NHD TotDASqKm

3

USGS site (NWIS)

0

Matched SWAT–NHD

3

Median |Δ| vs NHD

98.2%

Median SWAT/NHD ratio

0.0176

Within 0.5–2.0× (SWAT/NHD)

0 / 3

Missing chandeg

0

Assignment outliers

3