USGS National Hydrography

Contents

root

Details

Visibility: Public
Owner: Carl Boettiger
Created: 25 Jul 2026
Last Updated: 25 Jul 2026

Analytics

28 days

Downloads

1,053

Data served

460.5 GB

Countries

28-day downloads

17 Jul 2026

13 Aug 2026

Analytics

README

USGS National Hydrography Dataset (NHD) — Streams (national)

Cloud-native distribution of the USGS National Hydrography Dataset, High Resolution stream network (NHDFlowline), nationwide (50 states + DC + PR/VI + Pacific territories as bundled in the USGS national seamless GDB), plus NHDPlus High Resolution flowlines with the network value-added attributes. Three layers:

Dataset	What it is	Features
`streams-by-order`	All NHD flowlines, with Strahler stream order attached from the value-added attribute table (`NHDFlowlineVAA`). ⚠️ Order is usable on only 11.1% of flowlines and its coverage is geographic — read this first.	~30.2 M lines
`perennial-streams`	NHD flowlines filtered to perennial streams/rivers (`FCODE = 46006`). Needs no stream-order attribute, so it is complete nationwide — use this for flow-permanence questions.	subset of the above
`nhdplus-hr/flowline`	NHDPlus HR flowlines + network VAA (order, drainage area, slope, routing). ✅ The source to use for stream order — 100.00% coverage of in-network non-coastline length. National (all 266 published units). Units are ingested whole, so any regional statistic must be masked to your area of interest first (read this).	25.58 M lines

Producer: U.S. Geological Survey, National Hydrography Program (High Resolution).
License: public-domain (US federal work).
Source: NHD_H_National_GDB.zip (national seamless High-Resolution GDB).
Geometry: vector lines (MultiLineString), OGC:CRS84 (lon, lat). Reprojected from the source NAD83 (EPSG:4269).
H3 hex: each dataset is indexed to H3 resolution 8. Line features are hexed to H3 res 8 by buffering each segment by the H3 cell circumradius before polyfill.

Each dataset ships three assets: GeoParquet (<name>.parquet), PMTiles (<name>.pmtiles), and hive-partitioned H3 hex (<name>/hex/h0=*/data_0.parquet).

DuckDB — query the GeoParquet

DuckDB — H3 hex (aggregation / spatial join)

Hex aggregation caveat. One hex row = one (flowline, h8) pair. Per-feature columns (lengthkm, shape_length) are repeated on every cell a flowline covers — never SUM them on hex data. Network length ≈ SUM(lengthkm) only after SELECT DISTINCT permanent_identifier, lengthkm. The H3 index columns (h8, h0) are the only fields safe to aggregate directly.

MapLibre GL JS — PMTiles

The source-layer is the last segment of the dataset name (streams-by-order or perennial-streams), NOT the GDB layer name.

Stream order — sparse and geographic (read before filtering on it)

streams-by-order carries streamorder (Strahler stream order, integer) joined from the NHD value-added attribute table NHDFlowlineVAA on permanent_identifier. Higher = larger downstream river; headwater channels are order 1. streamlevel (the downstream-mainstem hierarchy, counted down from the terminal outlet) is included as well.

⚠️ USGS populates streamorder for only a minority of NHD subbasins, and the gap is GEOGRAPHIC, not topological. Only 3,355,035 of 30,221,659 flowlines (11.1%) carry a usable order, while innetwork = 1 — the connected drainage network — covers 29,184,336 flowlines (96.6%). So this is not "order is only computed for networked reaches": most networked reaches have no order. It reflects NHD's per-subbasin refresh history:

15 of the 22 HUC2 regions are at exactly 0.0% usable coverage. The best-covered region is HUC2 17 at 66.1%, then 02 (28.2%), 11 (18.2%), 18 (12.4%), 01 (11.1%), 16 (4.0%), 04 (3.3%). No region is complete.
Coverage varies within a single HUC4: California's 1801 (Klamath) is 73.4%, the adjacent 1802 (Sacramento) is 0.7%.
streamorder and streamlevel coverage are uncorrelated — USGS populated one attribute or the other per subbasin (HUC2 17 = 66.1% order / 6.4% level; HUC2 14 = 0.0% / 83.6%). They are also not interchangeable: where both exist they agree on only 4.3% of rows (correlation −0.116).

⛔ Consequence: streamorder cannot select a stream class over any area larger than a verified subbasin. The predicate looks like an attribute filter and behaves as a geographic one — in California it selects 94,505 km of 1,083,964 km of in-network flowline (8.7%), and 92,393 km of that (97.8%) is HUC4 1801 alone. A "California streams by order" total computed this way is a Klamath-basin total. Where order is computed it is not selective by feature type (in 1801 it covers 87.3% of perennial, 86.1% of intermittent and 65.6% of ephemeral length), so the populated subset is not "the real streams" either. Always run the per-HUC2 coverage query above first.

streamorder = 0 is a source sentinel meaning "not computed", not an order (1,234,262 rows). The valid Strahler range is 1–10; a few thousand records carry out-of-range values up to 191. Filter streamorder > 0, never IS NOT NULL — IS NOT NULL admits the sentinel and certifies regions with nothing usable (HUC2 12 looks 96.7% covered and is 0.0% usable; HUC2 13 goes 17.5% → 0.0%).

For flow permanence, use perennial-streams (fcode = 46006) — it needs no VAA attribute and is complete nationwide.

Usable stream order lives in nhdplus-hr/flowline (below) — NHDPlus High Resolution's NHDPlusFlowlineVAA computes the network attributes properly, and it is published here for the 13 hydrologic units that intersect California with 100.00% coverage; the national fan-out is data-workflows#205. Full measurement of this gap: data-workflows#518.

FCODE (feature code) reference

perennial-streams is the subset where fcode = 46006. For reference, related stream FCODEs in the source:

FCODE	Meaning
46000	Stream/River (general, unspecified hydrographic category)
46003	Stream/River — Intermittent
46006	Stream/River — Perennial
46007	Stream/River — Ephemeral
46100	Submerged Stream

Source: USGS NHD NHDFCode domain table.

NHDPlus High Resolution — the source to use for STREAM ORDER (`nhdplus-hr/flowline`)

A third dataset in this bucket, added because base NHD-H cannot answer stream-order questions (above). USGS NHDPlus HR computes the network value-added attributes properly: across all 266 source units nationwide — 238 at HU4 granularity plus 28 HU8 units in Alaska — streamorde > 0 covers 100.00% of in-network, non-coastline flowline length in every single unit, against base NHD's 11.4% nationally with 15 of 22 HUC2 regions at exactly 0.0%.

	`streams-by-order` (base NHD-H)	`nhdplus-hr/flowline` (NHDPlus HR)
extent	national, incl. all of Alaska + territories	national — all 266 published units (238 HU4 across HUC2 01-18, 20-22 + 28 HU8 in HUC2 19). Alaska coverage is partial (a subset of region 19's HU8s), so base NHD-H remains the complete Alaska source
features	30,221,659	25,583,812
usable stream order	11.1% of flowlines; 15 of 22 HUC2 at 0.0%	100.00% of in-network non-coastline length in all 266 units
use it for	extent, flow permanence (`fcode`), complete Alaska	stream order, drainage area, slope, network navigation

Units are whole — mask to your area of interest

Source units are hydrologic units, and each is ingested whole; none is clipped to any state line. The collection is now national, so this applies to every region you might ask about, and an unmasked query returns a units total, never a state total.

California is the worked example, from the 13 HU4 units that intersect the state: 30.3% of their 1,175,296 km of in-network flowline lies outside California — in Nevada, Utah, Oregon and Arizona. Three units are mostly out of state:

HU4	unit	in-network km	of which in CA	% outside CA
1606	Great Salt Lake / Escalante	224,178	11,966	94.7
1605	Central Nevada Desert Basins	50,618	10,621	79.0
1503	Lower Colorado	104,386	42,566	59.2
1808	North Lahontan	13,855	11,543	16.7
1801	Klamath	126,408	111,251	12.0
1807	Southern California Coastal	79,437	70,660	11.0
1809	Northern Mojave–Mono Lake	111,214	100,348	9.8

Ingesting whole units is correct — a hydrologic unit is the natural unit of a stream network — but it means a regional statistic has to be masked to that region first, on the hex asset, by joining h8 and h0 against a boundary hex layer. Without the mask, out-of-region length is counted as local: California order 1–2 (headwater) length is 924,634 km across those 13 units and 641,393 km inside the state — a 44% overstatement. The same trap applies to every state now that coverage is national, and it is larger for states whose units extend well beyond them.

The same holds for every future tranche: a set of hydrologic units is never a state extent.

These two are complementary — do not substitute one for the other. Base NHD-H is the denser, more recently edited network: over the 13 California units it holds 289,426 km of ephemeral channel that this NHDPlus HR vintage does not have (plus 24,336 km intermittent, 7,257 km perennial), because NHD has since been densified with ephemeral washes. Conversely, for the 2,017,710 features shared by both with the same fcode, NHDPlus HR lengthkm runs +1.9% (ephemeral), +4.8% (intermittent), +5.5% (perennial) longer. That length divergence is a per-VPU editing-vintage difference, not a different length algorithm — where the underlying NHD has not been re-edited the two agree to the millimetre — so compare per unit using the vpu_vintage column rather than applying a blanket offset. fcode itself is ~96% stable (81,782 of 2,017,710 shared features reclassify, in both directions): there is no redefinition of ephemeral or perennial.

Sourced from the per-HU4 VPU/Current/GDB downloads, not the National Release 2 aggregate, which USGS documents as defective (region 06 falls back to Beta data with a disconnected network; a GridCode bug affects VPUIDs 0903/1007/1015/1021/1022/1025; 0415 is a pre-Beta prototype). Each unit's source filename and vintage is recorded in the STAC collection description.

streamorde is NULL, never 0, wherever order is not computed (off-network reaches, coastline, canals/pipelines): the source's non-positive sentinels — including -9 on divergent paths — are normalised to NULL at build time, so the published range is exactly 1-10. Audit coverage with > 0 anyway; it is the correct predicate for the source and for streams-by-order.

MapLibre source-layer for nhdplus-hr/flowline.pmtiles is flowline (zoom 0-10). Tiles drop features on dense tiles by design — never compute quantities from them; use the GeoParquet or hex.

Per-unit provenance (each source filename, its publication vintage, and that unit's measured coverage) is published as nhdplus-hr/flowline/units-manifest.csv — 266 rows.

README

USGS National Hydrography Dataset (NHD) — Streams (national)

Dataset	What it is	Features
`streams-by-order`	All NHD flowlines, with Strahler stream order attached from the value-added attribute table (`NHDFlowlineVAA`). ⚠️ Order is usable on only 11.1% of flowlines and its coverage is geographic — read this first.	~30.2 M lines
`perennial-streams`	NHD flowlines filtered to perennial streams/rivers (`FCODE = 46006`). Needs no stream-order attribute, so it is complete nationwide — use this for flow-permanence questions.	subset of the above
`nhdplus-hr/flowline`	NHDPlus HR flowlines + network VAA (order, drainage area, slope, routing). ✅ The source to use for stream order — 100.00% coverage of in-network non-coastline length. National (all 266 published units). Units are ingested whole, so any regional statistic must be masked to your area of interest first (read this).	25.58 M lines

Producer: U.S. Geological Survey, National Hydrography Program (High Resolution).
License: public-domain (US federal work).
Source: NHD_H_National_GDB.zip (national seamless High-Resolution GDB).
Geometry: vector lines (MultiLineString), OGC:CRS84 (lon, lat). Reprojected from the source NAD83 (EPSG:4269).
H3 hex: each dataset is indexed to H3 resolution 8. Line features are hexed to H3 res 8 by buffering each segment by the H3 cell circumradius before polyfill.

Each dataset ships three assets: GeoParquet (<name>.parquet), PMTiles (<name>.pmtiles), and hive-partitioned H3 hex (<name>/hex/h0=*/data_0.parquet).

DuckDB — query the GeoParquet

DuckDB — H3 hex (aggregation / spatial join)

Hex aggregation caveat. One hex row = one (flowline, h8) pair. Per-feature columns (lengthkm, shape_length) are repeated on every cell a flowline covers — never SUM them on hex data. Network length ≈ SUM(lengthkm) only after SELECT DISTINCT permanent_identifier, lengthkm. The H3 index columns (h8, h0) are the only fields safe to aggregate directly.

MapLibre GL JS — PMTiles

The source-layer is the last segment of the dataset name (streams-by-order or perennial-streams), NOT the GDB layer name.

Stream order — sparse and geographic (read before filtering on it)

15 of the 22 HUC2 regions are at exactly 0.0% usable coverage. The best-covered region is HUC2 17 at 66.1%, then 02 (28.2%), 11 (18.2%), 18 (12.4%), 01 (11.1%), 16 (4.0%), 04 (3.3%). No region is complete.
Coverage varies within a single HUC4: California's 1801 (Klamath) is 73.4%, the adjacent 1802 (Sacramento) is 0.7%.
streamorder and streamlevel coverage are uncorrelated — USGS populated one attribute or the other per subbasin (HUC2 17 = 66.1% order / 6.4% level; HUC2 14 = 0.0% / 83.6%). They are also not interchangeable: where both exist they agree on only 4.3% of rows (correlation −0.116).

For flow permanence, use perennial-streams (fcode = 46006) — it needs no VAA attribute and is complete nationwide.

FCODE (feature code) reference

perennial-streams is the subset where fcode = 46006. For reference, related stream FCODEs in the source:

FCODE	Meaning
46000	Stream/River (general, unspecified hydrographic category)
46003	Stream/River — Intermittent
46006	Stream/River — Perennial
46007	Stream/River — Ephemeral
46100	Submerged Stream

Source: USGS NHD NHDFCode domain table.

NHDPlus High Resolution — the source to use for STREAM ORDER (`nhdplus-hr/flowline`)

	`streams-by-order` (base NHD-H)	`nhdplus-hr/flowline` (NHDPlus HR)
extent	national, incl. all of Alaska + territories	national — all 266 published units (238 HU4 across HUC2 01-18, 20-22 + 28 HU8 in HUC2 19). Alaska coverage is partial (a subset of region 19's HU8s), so base NHD-H remains the complete Alaska source
features	30,221,659	25,583,812
usable stream order	11.1% of flowlines; 15 of 22 HUC2 at 0.0%	100.00% of in-network non-coastline length in all 266 units
use it for	extent, flow permanence (`fcode`), complete Alaska	stream order, drainage area, slope, network navigation

Units are whole — mask to your area of interest

HU4	unit	in-network km	of which in CA	% outside CA
1606	Great Salt Lake / Escalante	224,178	11,966	94.7
1605	Central Nevada Desert Basins	50,618	10,621	79.0
1503	Lower Colorado	104,386	42,566	59.2
1808	North Lahontan	13,855	11,543	16.7
1801	Klamath	126,408	111,251	12.0
1807	Southern California Coastal	79,437	70,660	11.0
1809	Northern Mojave–Mono Lake	111,214	100,348	9.8

The same holds for every future tranche: a set of hydrologic units is never a state extent.

streamorde is NULL, never 0, wherever order is not computed (off-network reaches, coastline, canals/pipelines): the source's non-positive sentinels — including -9 on divergent paths — are normalised to NULL at build time, so the published range is exactly 1-10. Audit coverage with > 0 anyway; it is the correct predicate for the source and for streams-by-order.

Per-unit provenance (each source filename, its publication vintage, and that unit's measured coverage) is published as nhdplus-hr/flowline/units-manifest.csv — 266 rows.

1INSTALL spatial; LOAD spatial;
2
3-- Perennial streams in a bounding box (lon/lat). The geometry column is SHAPE.
4SELECT permanent_identifier, gnis_name, lengthkm, fcode
5FROM read_parquet('https://s3-west.nrp-nautilus.io/public-usgs-nhd/perennial-streams.parquet')
6WHERE ST_Intersects(SHAPE, ST_MakeEnvelope(-122.6, 37.7, -122.3, 37.9))
7LIMIT 20;
8
9-- Stream-order distribution (streams-by-order).
10-- NOTE: streamorder = 0 is a 'not computed' sentinel, not an order — filter > 0.
11SELECT streamorder, COUNT(*) AS n
12FROM read_parquet('https://s3-west.nrp-nautilus.io/public-usgs-nhd/streams-by-order.parquet')
13WHERE streamorder > 0
14GROUP BY streamorder ORDER BY streamorder;
15
16-- Usable order coverage by HUC2 region — run this before trusting any order-filtered
17-- total over an area. 15 of the 22 regions come back at exactly 0.0%.
18SELECT substr(reachcode, 1, 2) AS huc2, COUNT(*) AS n,
19       ROUND(100.0 * COUNT(*) FILTER (WHERE streamorder > 0) / COUNT(*), 1) AS pct_usable
20FROM read_parquet('https://s3-west.nrp-nautilus.io/public-usgs-nhd/streams-by-order.parquet')
21WHERE innetwork = 1
22GROUP BY 1 ORDER BY 1;

1INSTALL spatial; LOAD spatial;
2
3-- Perennial streams in a bounding box (lon/lat). The geometry column is SHAPE.
4SELECT permanent_identifier, gnis_name, lengthkm, fcode
5FROM read_parquet('https://s3-west.nrp-nautilus.io/public-usgs-nhd/perennial-streams.parquet')
6WHERE ST_Intersects(SHAPE, ST_MakeEnvelope(-122.6, 37.7, -122.3, 37.9))
7LIMIT 20;
8
9-- Stream-order distribution (streams-by-order).
10-- NOTE: streamorder = 0 is a 'not computed' sentinel, not an order — filter > 0.
11SELECT streamorder, COUNT(*) AS n
12FROM read_parquet('https://s3-west.nrp-nautilus.io/public-usgs-nhd/streams-by-order.parquet')
13WHERE streamorder > 0
14GROUP BY streamorder ORDER BY streamorder;
15
16-- Usable order coverage by HUC2 region — run this before trusting any order-filtered
17-- total over an area. 15 of the 22 regions come back at exactly 0.0%.
18SELECT substr(reachcode, 1, 2) AS huc2, COUNT(*) AS n,
19       ROUND(100.0 * COUNT(*) FILTER (WHERE streamorder > 0) / COUNT(*), 1) AS pct_usable
20FROM read_parquet('https://s3-west.nrp-nautilus.io/public-usgs-nhd/streams-by-order.parquet')
21WHERE innetwork = 1
22GROUP BY 1 ORDER BY 1;

1-- How many res-8 cells does the perennial network touch in each h0 region?
2SELECT h0, COUNT(DISTINCT h8) AS cells
3FROM read_parquet('https://s3-west.nrp-nautilus.io/public-usgs-nhd/perennial-streams/hex/h0=*/data_0.parquet')
4GROUP BY h0 ORDER BY cells DESC;

1-- How many res-8 cells does the perennial network touch in each h0 region?
2SELECT h0, COUNT(DISTINCT h8) AS cells
3FROM read_parquet('https://s3-west.nrp-nautilus.io/public-usgs-nhd/perennial-streams/hex/h0=*/data_0.parquet')
4GROUP BY h0 ORDER BY cells DESC;

1const map = new maplibregl.Map({
2  container: 'map',
3  style: 'https://demotiles.maplibre.org/style.json',
4  center: [-119, 37], zoom: 5
5});
6map.on('load', () => {
7  const p = new pmtiles.PMTiles('https://s3-west.nrp-nautilus.io/public-usgs-nhd/perennial-streams.pmtiles');
8  protocol.add(p);
9  map.addSource('streams', { type: 'vector', url: 'pmtiles://' + p.source.getKey() });
10  map.addLayer({
11    id: 'streams', type: 'line', source: 'streams',
12    'source-layer': 'perennial-streams',
13    paint: { 'line-color': '#2b8cbe', 'line-width': 0.6 }
14  });
15});

1const map = new maplibregl.Map({
2  container: 'map',
3  style: 'https://demotiles.maplibre.org/style.json',
4  center: [-119, 37], zoom: 5
5});
6map.on('load', () => {
7  const p = new pmtiles.PMTiles('https://s3-west.nrp-nautilus.io/public-usgs-nhd/perennial-streams.pmtiles');
8  protocol.add(p);
9  map.addSource('streams', { type: 'vector', url: 'pmtiles://' + p.source.getKey() });
10  map.addLayer({
11    id: 'streams', type: 'line', source: 'streams',
12    'source-layer': 'perennial-streams',
13    paint: { 'line-color': '#2b8cbe', 'line-width': 0.6 }
14  });
15});

1-- Stream length by Strahler order, nationwide (the query base NHD cannot answer).
2-- This is a NATIONAL total. For any state or region, mask first — see below.
3SELECT streamorde, ROUND(SUM(lengthkm)) AS km
4FROM read_parquet('https://s3-west.nrp-nautilus.io/public-usgs-nhd/nhdplus-hr/flowline.parquet')
5WHERE innetwork = 1 AND streamorde > 0
6GROUP BY 1 ORDER BY 1;

1-- Stream length by Strahler order, nationwide (the query base NHD cannot answer).
2-- This is a NATIONAL total. For any state or region, mask first — see below.
3SELECT streamorde, ROUND(SUM(lengthkm)) AS km
4FROM read_parquet('https://s3-west.nrp-nautilus.io/public-usgs-nhd/nhdplus-hr/flowline.parquet')
5WHERE innetwork = 1 AND streamorde > 0
6GROUP BY 1 ORDER BY 1;

1-- California only: keep the flowline hex cells that fall inside the state
2WITH ca AS (SELECT DISTINCT h8, h0 FROM read_parquet('s3://public-ca30x30/ecoregion/hex/h0=*/data_0.parquet'))
3SELECT ROUND(SUM(len)) AS km FROM (
4  SELECT s._cng_fid, ANY_VALUE(s.lengthkm) AS len
5  FROM read_parquet('s3://public-usgs-nhd/nhdplus-hr/flowline/hex/h0=*/data_0.parquet') s
6  JOIN ca USING (h8, h0)
7  WHERE s.innetwork = 1 AND s.streamorde IN (1,2)
8  GROUP BY s._cng_fid);

1-- California only: keep the flowline hex cells that fall inside the state
2WITH ca AS (SELECT DISTINCT h8, h0 FROM read_parquet('s3://public-ca30x30/ecoregion/hex/h0=*/data_0.parquet'))
3SELECT ROUND(SUM(len)) AS km FROM (
4  SELECT s._cng_fid, ANY_VALUE(s.lengthkm) AS len
5  FROM read_parquet('s3://public-usgs-nhd/nhdplus-hr/flowline/hex/h0=*/data_0.parquet') s
6  JOIN ca USING (h8, h0)
7  WHERE s.innetwork = 1 AND s.streamorde IN (1,2)
8  GROUP BY s._cng_fid);

USGS National Hydrography

USGS National Hydrography

USGS National Hydrography Dataset (NHD) — Streams (national)

DuckDB — query the GeoParquet

DuckDB — H3 hex (aggregation / spatial join)

MapLibre GL JS — PMTiles

Stream order — sparse and geographic (read before filtering on it)

FCODE (feature code) reference

NHDPlus High Resolution — the source to use for STREAM ORDER (nhdplus-hr/flowline)

Units are whole — mask to your area of interest

USGS National Hydrography Dataset (NHD) — Streams (national)

DuckDB — query the GeoParquet

DuckDB — H3 hex (aggregation / spatial join)

MapLibre GL JS — PMTiles

Stream order — sparse and geographic (read before filtering on it)

FCODE (feature code) reference

NHDPlus High Resolution — the source to use for STREAM ORDER (nhdplus-hr/flowline)

Units are whole — mask to your area of interest

NHDPlus High Resolution — the source to use for STREAM ORDER (`nhdplus-hr/flowline`)

NHDPlus High Resolution — the source to use for STREAM ORDER (`nhdplus-hr/flowline`)