Computation Engines
The STAF web applications gather watershed metrics for you: impervious cover, wetland extent, road density, dam storage, riparian vegetation and the like. Two engines produce them. With a spreadsheet calculator you enter these values yourself, from these sources or your own. This page says what each engine is, which app uses which, what each cannot produce, and how to read the labels in a report.
The two engines
StreamCat lookup engine. EPA StreamCat publishes precomputed landscape summaries for every reach of the NHDPlus V2 network (about 1:100,000 scale). The lookup engine snaps a point to its V2 reach, delineates that reach’s basin with the USGS NLDI service, and reads the published values for the reach’s COMID. It is fast, reproducible, and citable by data vintage. It also carries the EPA modeled integrity indices that only exist per V2 reach. Its limit is coverage: roughly nine of every ten stream miles in the full-resolution NHD are not on the V2 network, and most of those are headwaters.
STAF site engine. The site engine works on the high-resolution NHD. Given a point on any stream, it snaps the point to the nearest HR reach and delineates that reach’s watershed, the HR reach watershed: the NHDPlus HR catchments upstream of the reach, aggregated and checked against the reach’s published drainage area. The reach, not the point, is the outlet, so a point partway up a reach gets the reach’s whole upstream area. It then computes the watershed metrics from source data: NLCD land cover on the watershed and a 100 m riparian buffer, TIGERweb roads, National Inventory of Dams records inside the polygon, SSURGO soil erodibility, and EROM flow. It usually takes well under a minute, up to about five minutes on a large basin, and refuses, with a reason, when a basin exceeds its budget. Every value it produces carries its source, vintage and the engine version.
| StreamCat lookup engine | STAF site engine | |
|---|---|---|
| Works on | NHDPlus V2 reaches only | Any NHD stream in the conterminous United States |
| Watershed | The V2 reach’s published basin | The HR reach watershed of the reach the point snaps to |
| Speed | Seconds | Usually under a minute, up to about five on a large basin |
| Cannot produce | Anything off the V2 network | The EPA modeled integrity indices, NRSA field observations, precipitation and temperature normals |
| Produces differently | Runoff (EROM-derived rather than the StreamCat water-balance grid) and the riparian buffer (built on the high-resolution stream network) |
Which app uses which
No app asks the user to pick a method. Each applies one fixed policy, and each report says which engine produced every value.
EASI, SFARI, and DEEP display one solid blue stream network by default. In the Layers menu, StreamCat coverage is initially off. Enabling it distinguishes StreamCat reaches from other streams and reveals the selected data-source reach and its downstream connector. This display setting leaves the assessment point, reach, watershed, and calculation policy unchanged.
Each app resolves a StreamCat source reach before allowing new analysis. The selected point remains visible while the status above Delineate shows lookup progress with a small spinning circle while work is active, including retry pauses. The circle stays stationary for reduced-motion preferences. Temporary routing failures retry up to three times, after pauses of 5, 10, and 15 seconds; if unresolved, Retry StreamCat lookup retries the same snapped point. A no-match result or invalid service response also blocks analysis with an explanation. A resolved source means a valid COMID; individual metrics are retrieved later and can still be unavailable. Imported results remain readable while missing source information is resolved. Coverage toggles do not trigger lookup or affect readiness.
| Tier | App | Policy |
|---|---|---|
| Screening | EASI | Within 150 ft of an NHDPlus V2 reach, the StreamCat lookup engine supplies watershed metrics in seconds. Elsewhere the STAF site engine calculates the HR reach watershed. Reach-keyed metrics, including the modeled benthic probability and the integrity fallbacks, can use the nearest StreamCat reach downstream; their source details include the routed distance and drainage-area ratio. If the engine cannot compute the watershed, the watershed metrics are unavailable with guidance, never a stand-in. |
| Rapid | SFARI | Every click snaps to the high-resolution NHD. The site engine computes the HR reach watershed and every watershed value at every site. The StreamCat lookup engine supplies, by COMID, the EPA modeled indices that exist only per V2 reach (flow permanence, dewatered segments, barriers, nutrients) and stands in, labeled, for a watershed value the engine could not compute. On a stream outside V2 the COMID is the nearest StreamCat reach downstream, named with the routed distance and the drainage-area ratio. Direct services (gages, water quality, wetlands, dams) answer the rest. The assessor keeps every score. |
| Detailed | DEEP | Every click snaps to the high-resolution NHD. The site engine computes the HR reach watershed at every site. Auto-pulled values follow the assessment metric by metric: a curve whose own predictorSource records engine predictors takes the site-engine value (every landscape metric, base-flow index and road-stream crossings included since engine 0.3.0), and every other curve takes the lookup-engine value by COMID that it was fitted on, so a mixed bundle scores both. On a stream outside V2 the COMID is the nearest StreamCat reach downstream, named with the routed distance and the drainage-area ratio. A value from the other engine is shown as reference and not scored. |
| Detailed (builder) | StreamCurves | The predictor source is the one choice in the program: the StreamCat lookup engine by default, or the site engine, recorded in every build’s provenance. An engine-sourced build also recomputes the scored landscape metrics with an engine analog (impervious, crop, wetland, road density, dam density, base-flow index, road-stream crossings) over the HR reach watershed at every training site, and stamps those curves as engine-fitted. Surface water storage is one combined wetland metric, woody plus herbaceous, the same sum EASI scores. The reference screen always runs on the lookup engine, keyed on each NRSA site’s own archive COMID, so pool membership is decided on the reach the crew sampled, and the candidate panel is wadeable-only. |
Reading the labels
- EASI reports name the engine on watershed rows and mark downstream desktop evidence with a dagger, explained below the metrics table. Detailed source information retains the source reach, distance, and drainage-area ratio; routine provenance does not add a report-top banner. CSV and GeoJSON exports retain engine and reach provenance.
- SFARI marks every pulled value Desktop on the Assessment page and names its data in the value’s (i); a value from the nearest StreamCat reach downstream says so under it. The main report and assessment PDF mark affected desktop evidence with a dagger and a short explanation below the metrics table. The report tooltips, the desktop metrics PDF and the report CSV retain the detailed source descriptions, engine included.
- DEEP marks an auto-pulled value Desktop beside its input, and shows an advisory when a value is shown as reference only. Get Forms names the data behind each value, and the report CSV keeps the engine and basis.
- StreamCurves records the predictor source in the run manifest, the published bundle and the science support report.
Equivalence study
Whether HR reach delineation changes screening scores enough to make the site engine the standard on covered streams was tested, not assumed. The study ran EASI at 30 NRSA sites in two pilot regions with StreamCat inputs and with site-engine inputs and compared the watershed-metric ratings, the condition class, and the DEEP curve indices against a rule fixed before the run. Outcome (2026-09-02): the engines are not interchangeable. 30 of 30 sites ran on 2026-09-02: watershed-metric rating agreement 0.84 pooled (Eastern Corn Belt Plains 0.78, Northeastern Highlands 0.90) against the 0.90 bar, condition-class agreement 0.97, median DEEP index shift 0.013. The class and the index agree well; the rating disagreements come mostly from watershed extent, because an HR reach partway up a V2 reach, or on a tributary the V2 network does not carry, drains far less than the V2 reach’s whole basin. So EASI keeps the StreamCat lookup engine on covered streams and documents the approximation, DEEP keeps refusing engine values against StreamCat-fitted curves, and StreamCurves builds engine-predictor versions of the pilot assessments so DEEP can score engine values against curves fitted the same way.