SWAT+ delineation QA
SWAT+ drainage area audit (TxtInOut vs NHDPlus HR)
The final QA step before calibration: we read the drainage area SWAT+ actually routes from executable TxtInOut files (chandeg.con) and compare it to the original NHDPlus HR and NWIS values at every gaged channel — every pair shown, outliers investigated instead of averaged away.
QA step 3 of 3 (delineation → gage assignment → this audit). Numbers are computed live from the committed per-gage catalog; the Peace deep-dive below documents every hypothesis we tested.
- 98 gages · 8 models
- 78 of 97 matched pairs within 0.5–2.0
- Median SWAT/NHD ratio 1.10
- Read from executable chandeg.con
This is the third and final step of the SWATGenX streamflow-QA pipeline. The delineation builds the network, gage assignment ties each USGS gage to a channel, and this audit verifies the last link: whether that assigned channel carries the correct drainage area in the executable model. For every gage with an assignment, we read chandeg.con area (hectares → km²) for the matching GIS channel, compare it to the original NHDPlus HR flowline TotDASqKm at the gage, and list NWIS site metadata where available — across all eight publication models, not just the Peace River benchmark.
Read it in order: assignment quality comes first (a wrong-channel pick will look like an area error), so the drainage-area ratios here should be interpreted on the calibration-ready gages identified on the gage-assignment page. That keeps a true SWAT+-vs-NHD area difference from being confused with a mis-assigned gage.
The headline metric is SWAT+ versus NHD on matched reaches: portfolio-wide, 78 of 97 matched station pairs fall within a 0.5–2.0 ratio, with a median SWAT/NHD ratio of 1.10. The population deliberately includes every station — canals and lake outlets too — so the out-of-band pairs are dominated by assignment-review classes rather than conversion errors: on Peace River, 52 of the 54 assignment-clean gages are in-band (median 1.10). Large SWAT+–NWIS gaps are read as tributary-versus-mainstem assignment diagnostics unless SWAT+–NHD also diverges.
Below the portfolio tables, a Peace River deep-dive documents every hypothesis we tested — wrong metric, merged NHD artifacts, orphan reaches, lakes, the polygon stage, and QSWAT/TauDEM-vs-VAA definition differences — in an experiment-log format, so genuine agreement, systematic offsets, and assignment outliers are reported separately rather than averaged together.
FAQ
How accurate are SWAT+ drainage areas compared to NHDPlus HR?
Close, at the executable-model level: across eight evaluation models, 78 of 97 gage–channel pairs with both values fall within a factor-of-two band (0.5–2.0 SWAT/NHD ratio), with a portfolio median ratio of 1.10. The out-of-band pairs concentrate in assignment-review stations (canals, lake outlets) rather than area-conversion errors — on the Peace River model, 52 of the 54 assignment-clean gages are in-band (median 1.10). Peace mainstem reaches carry a documented systematic +10–17% SWAT-over-NHD offset that traces to a QSWAT/TauDEM-vs-NHD-VAA delineation-definition difference, not a preprocessing bug.
Why compare chandeg.con instead of shapefiles?
Shapefiles and SQLite exports document preprocessing products, but SWAT+ executes from TxtInOut. The chandeg.con area column is what the solver uses for channel contributing drainage area at each reach.
Why can SWAT+ and NWIS drainage areas disagree when SWAT+ and NHD agree?
NWIS drainage area describes the gage watershed metadata, while the assigned GIS channel may be a tributary reach. SWAT+ versus NHD is the direct check that NHDPlus HR drainage accounting survived into executable model files.
What does “missing chandeg row” mean?
The station is linked to a GIS channel id that does not appear in chandeg.con—often reservoir outlets or auxiliary routing objects. Those rows are flagged for assignment review, not hidden as conversion errors.
Why can Peace River mainstem gages show ~10–17% higher SWAT+ than NHD TotDASqKm?
A phased trace on a 10-gage Peace panel shows the offset first appears at QSWAT channel AreaC and chandeg.con, not at original NHD VAA, cleaned-flowline reach TotDASqKm, or post-processed watershed polygons (~+2–3% there). The leading interpretation is a delineation-definition difference between NHDPlus HR cumulative drainage and QSWAT/TauDEM contributing area—not necessarily a preprocessing bug.
Did SWATGenX inflate NHD drainage area before QSWAT?
No on the tested reaches: original zip TotDASqKm and cleaned-flowline reach TotDASqKm match within 0.1% (e.g. 02294650: 856.95 km² both). the SWAT+ stream layer does not export a drainage-area column for QSWAT to copy. Live rivs1 verification is pending where project shapes are not on disk.
Next steps
Last updated 2026-07-13.