Glossary

One Full product computed from one baseline hydrography dataset, on its native grid. Two are out, both of them FullBasin: on MERIT Hydro at 90 m, global; on HydroSHEDS v2.0 at 30 m, North and South America so far. Both use the same words and the same table layout.

See MERIT Hydro, HydroSHEDS v2.0.

Each released product is one Zenodo record with a DOI; a new version of the data goes into the same record, and each version also gets a DOI of its own. Cite the dataset, the paper, and the baseline data separately. The site serves single files at data.fullhydro.org; the archive of record is Zenodo.

In the data: MERIT-FullBasin 10.5281/zenodo.20344113.

See licence, region package.

The method that divides a continent before anything is computed per pixel: every land pixel is assigned to the outlet it drains to, whole basins are gathered into basin groups, and the groups — or the pieces of a basin too large on its own — become regions within a capacity. It is step 0 of FullBasin, the page Basin and region mask.

See basin, basin group, region, capacity.

The product that answers, at every land pixel, which basin it belongs to, where it drains to, what drains into it and what that upstream land is like: the basin and region masks, the upstream index and sixteen upstream attributes. On MERIT Hydro it is complete, released on Zenodo and in discussion at Earth System Science Data; on HydroSHEDS v2.0 the masks are out for North and South America and the per-pixel layers follow.

In the data: 10.5281/zenodo.20344113; the paper 10.5194/essd-2026-402.

See basin mask, upstream index (dfs_in, dfs_out), upstream attributes, FlowDivide.

Hydrography for every land pixel. FullHydro starts from a data provider's flow-direction grid and computes, at every pixel, where it drains to, what drains into it and what that upstream land is like. The results are released as rasters, tables and vectors, organised as datasets, one per baseline grid.

See dataset, FullBasin, FullRiver.

The river product planned to follow FullBasin on each grid: the channels themselves, where each one runs, how the reaches connect, and what flows through them. Not yet released.

See FullBasin, channel network.

The baseline hydrography of the 30 m dataset (Lehner et al., 2022), on a 1 arc-second grid, distributed in 10° × 10° tiles. FullHydro computes it one continent at a time; North America and South America are out. Its flow directions are not redistributed.

See tile, global grid (continental grid).

Stated per layer. MERIT-FullBasin: the upstream index under ODbL 1.0, every other layer CC BY 4.0. MERIT Hydro: CC BY-NC 4.0 / ODbL, from its own page. HydroSHEDS v2.0: from its own page.

See upstream index (dfs_in, dfs_out), DOI.

The baseline hydrography of the 90 m dataset (Yamazaki et al., 2019), built from MERIT DEM (Yamazaki et al., 2017) on a 3 arc-second grid. Its layers are not redistributed here: register on the MERIT Hydro page and cite Yamazaki et al. (2019).

In the data: dir, elv, upa, upg, wth, hnd.

See baseline variables (MERIT Hydro), tile, licence.

The tightest rectangle around an object's own pixels — a basin, a piece, a region — not rounded to degrees. Given in rows and columns and, for basins, also in degrees.

In the data: basin_row_min … basin_col_max, bbox_minlon … bbox_maxlat (MERIT bbox_geo_minlon …); region table bbox_row_min ….

See window (bounding rectangle), row and column.

The flow directions of a whole dataset assembled into one grid, never held in memory as a whole. MERIT Hydro: one global grid, 432,000 columns × 174,000 rows, 1,200 pixels per degree. HydroSHEDS v2.0: one grid per continent, North America 424,800 × 277,200, South America 219,600 × 259,200, 3,600 pixels per degree. Rows and columns count from its top-left corner.

See row and column, window (bounding rectangle).

The grid the baseline data comes on and everything is computed on: 3 arc-seconds for MERIT Hydro, 1 arc-second for HydroSHEDS v2. Nothing is resampled before computing. The vectors and map tiles are simplified for drawing, but every number they carry is the native-grid number.

See pixel (cell, grid cell), arc-second (″), vectors.

A pixel is one square of the native grid: 1 arc-second for HydroSHEDS v2.0, 3 arc-seconds for MERIT Hydro; every number in the tables is computed on it. A cell is one square of a coarser grid: the simplified grids the maps and vectors are drawn from, and the whole-degree cells the windows are aligned to. Grid cell is the neutral word used only in column names, because one table layout serves both resolutions.

In the data: basin_grid_cell_count, region_grid_cell_count, window_grid_cell_count count native pixels; coarse_grid_cell_count counts cells of the simplified grid; the MERIT tables still say basin_cell_count.

See native resolution, pixel count.

The position of a pixel on the grid: the row counts downward from the top edge, the column rightward from the left edge. The tables give outlets and bounding boxes in rows and columns so that a pixel can be read from the rasters directly; the MERIT basin table gives both the global position and the position inside the region's window. Every rectangle is left closed and right open: row_max and col_max are one past the last pixel, so nrow = row_max − row_min and a loop runs from row_min up to but not including row_max.

In the data: outlet_row, outlet_col, basin_row_min … basin_col_max; MERIT global_outlet_row, local_outlet_row ….

See global grid (continental grid), bounding box.

The producer's unit of distribution: HydroSHEDS v2.0 comes in 10° × 10° tiles, MERIT Hydro in 5° × 5° tiles packed in 30° packages. A tile edge cuts flow paths wherever the terrain takes them, which is why the work is organised by basins and regions; the tile grids are drawn on the maps because they say which files to fetch. Not the map tiles the site draws from.

See region, map tiles.

The smallest rectangle on whole-degree lines that holds a region or a basin group, read from the continental file in one go. Its edges are multiples of the pixels per degree, so a region is cut straight out of the global rasters with integer arithmetic and no resampling. window_grid_cell_count counts the whole window; fill_percent is the share the region actually covers.

In the data: row_min, row_max, col_min, col_max, nrow, ncol, window_grid_cell_count, fill_percent; group tables window_nrow, window_minlon ….

See region, bounding box, capacity.

Everything upstream of a pixel, the pixel itself included. A basin is the catchment of an outlet; at any other pixel the catchment is a part of a basin. FullBasin stores sixteen attributes of its catchment at every pixel with at least 10 km² upstream.

See basin, drainage divide, upstream attributes.

The pixels with at least a chosen upstream drainage area, drawn as rivers; the example basin draws those of at least 1 km². FullRiver will describe it reach by reach.

See FullRiver, upstream drainage area.

The boundary of a catchment: the line separating the pixels that drain through a given pixel from those that do not. The shape attributes are read off one walk around it.

See catchment, upstream attributes.

The direction each pixel drains to, one of its eight neighbours, as distributed by the data producer. Everything on this site is computed on top of it, continent-wide. The flow directions themselves are not redistributed; the upstream index encodes them losslessly.

In the data: MERIT Hydro dir, UInt8, nodata 247.

See upstream index (dfs_in, dfs_out), flow path.

The chain of pixels the water passes from a pixel to its outlet, following the flow directions. The page Where does a pixel drain to? follows it neighbour to neighbour.

See flow direction (D8), longest flow path.

A pixel that no other pixel drains into — where every flow path starts. The longest flow path of a basin runs from its most distant headwater to its outlet.

See flow path, longest flow path.

An outlet that is not on the coast: the lowest pixel of a closed depression, where the flow directions end. Its basin is endorheic. MERIT Hydro has 9,716 of them; the Caspian basin, global_basin_id 3, is the largest.

In the data: outlet_dir 255.

See outlet.

The longest flow path inside a basin, from its most distant headwater to its outlet. The 30 m map draws it, with arrows, for a cut basin, so that the way the water passes from region to region can be followed.

See cut basin, headwater.

The pixel a basin drains to: a river mouth on the coast or an inland sink. Every terminal pixel of the flow directions is an outlet — 24,587,290 on MERIT Hydro, most of them single coastal pixels — and each outlet has its own drainage area; only those of at least 1 km² are called basins on this site. The tables on this site keep the basins of at least 1 km²; the full MERIT table on Zenodo has every outlet. On the maps an outlet is a red downward triangle.

In the data: outlet_row, outlet_col, outlet_lon, outlet_lat; outlet_dir 0 at the coast, 255 in an inland sink.

See basin, inland sink.

Where two flow paths meet is a confluence; the smaller of the two joining streams is the tributary, and its last pixel before the junction is the tributary outlet. FlowDivide cuts a basin too large for one region at tributary outlets, so that every piece keeps a single outlet.

See piece, inlet (exit pixel).

Pixel p is upstream of pixel q when the flow path from p passes through q; q is then downstream of p. Everything upstream of a pixel is its catchment. With the upstream index the test is one comparison of two numbers.

See catchment, upstream index (dfs_in, dfs_out).

The whole drainage area of one outlet, at the coast or in an inland sink. Basins are numbered by area from 1, so basin 1 is the largest of whatever the numbering covers, global or continental — the Amazon globally and in South America, the Mississippi in North America — and the row of a basin in the basin table is its id.

In the data: bsn, bat; basin_grid_cell_count (MERIT basin_cell_count), basin_area_km2.

See outlet, basin id, basin mask, basin table.

One or more whole basins gathered under one HydroBASINS Level-01, Level-02 or Level-03 unit. A group has no capacity: it need not share an outlet, need not be connected, and can be any size. A group that fits the capacity becomes one region as it is; one that does not is divided along whole-degree lines, or its basin too large on its own is cut into pieces. Islands are gathered into their own groups.

In the data: gat_<continent>_30m_l2.csv, gat_…_l3.csv, grp_<continent>_l2.parquet, grp_…_l3.parquet; group_id, group_kind (1 one basin, 2 several basins, 3 islands); group_l2, group_l3 in the basin table.

See HydroBASINS unit, region, window (bounding rectangle), basin group table.

The number of a basin, equal to its row in the basin table: on MERIT Hydro basin_id = table_index + 1, so a pixel of the mask leads to its record by an array offset. MERIT has one global numbering and, inside each region, a local one. HydroSHEDS v2.0 is numbered per continent (continental_basin_id), and global_basin_id numbers the basins of North America and South America together, by area.

In the data: MERIT global_basin_id, region_id, local_basin_id; HydroSHEDS v2.0 continental_basin_id, global_basin_id.

See basin, basin table, basin mask.

The largest whole-degree rectangle a region may have, in square degrees. It bounds the rectangle, not the land inside it, so that a region is read as one window and addressed with a 32-bit index: 160 square degrees is 2,073,600,000 pixels at 1 arc-second, just under 2³¹. At 30 m the capacities of 160, 80 and 40 square degrees suit workstations with roughly 64, 32 and 16 GB of memory; MERIT Hydro v1.1 uses one capacity of 1400 square degrees.

See partition, window (bounding rectangle), square degree.

A basin that exceeds the capacity on its own and is cut into pieces at tributary outlets — the four largest basins of South America at 160 square degrees. Such a basin has no single region: its region column is 0, shown as cut on the map, and the region table names it in cut_basin_id. In the basin groups, and on MERIT Hydro, no basin is cut.

In the data: needs_cut, cut_basin_id.

See piece, capacity, tributary (confluence).

Basin 8595 of MERIT-FullBasin, 729 km², in Fujian, China: the basin on which the dataset pages show what a browser cannot draw for the whole world — the flow directions, the upstream index and the sixteen attributes at pixel level.

See FullBasin, upstream index (dfs_in, dfs_out).

A unit of HydroBASINS (Lehner and Grill, 2013) at its first, second or third level (Level-01, Level-02, Level-03), whose division and codes were defined manually. Level-01 is the nine large units. The Arctic (8) is a basin group of its own in both datasets; at 30 m it lies inside the North America extent of HydroSHEDS v2. It is used only to gather whole basins that lie together into a basin group; nothing is computed on it. The region number carries the Level-03 code in front: 62211 is the eleventh region of unit 622, the Amazon.

In the data: level2_code, level3_code in the group tables; level1, level2, level3 in the region tables; MERIT HydroSHEDS_L1 … _L3.

See basin group, region.

At a cut, the exit pixel is the pixel whose downstream neighbour lies in another part — a piece has at most one, its outlet — and the inlet is the pixel in the downstream part that receives that water. The piece table gives, for each piece, the inlet of its parent piece that its outlet flows into.

In the data: parent_inlet_row, parent_inlet_col (−1 for the piece holding the basin outlet); crossing_length_m, the distance from the outlet pixel to that inlet.

See piece, region order (topological level).

Basins that do not sit on a continental landmass, gathered by proximity into a group of their own. MERIT Hydro v1.1 numbers them apart from the 65 continental regions: 23 island groups, 20000 + continent × 100 + n (20101–20901), and two antimeridian groups, 10001–10002, stored as they are in the UInt16 region mask. (Version 1.0 had 29 island groups, 20001–20029, stored in a UInt8 mask as 211–239; its two antimeridian groups, 10001–10002, were stored as 111–112.) HydroSHEDS v2.0 marks them with group_kind 3.

See region, basin group, region mask.

One division of a continent into regions. The 30 m dataset shows three groupings and three partitions: the groupings gather whole basins into basin groups along the HydroBASINS Level-01, Level-02 and Level-03 units, with no limit on size; the partitions, at capacities of 160, 80 and 40 square degrees, start from the Level-03 groups and cut where a region exceeds the capacity. MERIT Hydro v1.1 has one partition of 90 regions; version 1.0 on Zenodo has 96.

In the data: group_l2, group_l3, region_160deg2, region_80deg2, region_40deg2 in the basin table, 0 where the basin is cut over several regions.

See basin group, capacity, region.

A connected part of one basin cut at tributary outlets, with one outlet and possibly several inlets; not a complete subtree. The pieces of a cut basin become regions. The piece table has one row per piece with its outlet, its parent piece (the piece it drains into) and the step between the two.

In the data: pat; piece_id, parent_piece_id (0 for the piece holding the basin outlet), outlet_row, outlet_col, parent_inlet_row, parent_inlet_col, piece_grid_cell_count, piece_area_km2, downstream_depth, crossing_length_m.

See cut basin, inlet (exit pixel), region, piece table.

What is computed at once: a set of whole basins, or the pieces of one cut basin, within the capacity, read together as one window of the continental grid. The region mask gives every land pixel its region; the region table has one row per region. MERIT Hydro v1.1 has 90 regions and no basin is cut; at 30 m the region number is the Level-03 code followed by a serial number.

In the data: rgn, rat; region_id, kind, basin_count, piece_count, cut_basin_id, region_grid_cell_count, topological_level.

See partition, window (bounding rectangle), region order (topological level), region mask.

The order in which regions are computed: a region after every region upstream of it, so that every value crossing a cut has arrived before it is needed. topological_level counts from the region holding the basin outlet (0) upstream, so the highest level is computed first and level 0 last; for pieces, downstream_depth counts the same way.

See region, piece.

The layers of MERIT Hydro that everything here is computed from, none of them redistributed: dir the flow direction; elv the hydrologically adjusted elevation, m; upa the upstream drainage area, km²; upg the number of upstream pixels; wth the river channel width, m; hnd the height above the nearest drainage, m.

In the data: dir UInt8 nodata 247, upg UInt32 nodata 0, upa Float32 nodata −9999.

See MERIT Hydro, flow direction (D8), upstream drainage area.

Sixteen numbers describing the catchment of a pixel, stored at every pixel with an upstream drainage area of at least 10 km²: eight terrain statistics over the catchment's pixels and eight shape measures from a walk around its outer boundary. Each is one raster per region, <variable>_region*_90m.tif, Float32, nodata −9999, CC BY 4.0. Table 3 of the MERIT-FullBasin paper.

See catchment, drainage divide, upstream drainage area.

The area of the catchment: the areas of its pixels summed, the pixel itself included, each pixel with the area it has at its latitude. MERIT Hydro's upa carries it at every pixel; the basin table gives it at every outlet as basin_area_km2. Unit km².

In the data: MERIT Hydro upa; basin_area_km2, piece_area_km2, area_km2.

See catchment, square kilometre (km²), baseline variables (MERIT Hydro).

Two numbers at every land pixel: the positions at which a depth-first walk of the flow network enters and leaves it. Pixel p is upstream of pixel q when dfs_in(q) ≤ dfs_in(p) ≤ dfs_out(q) — one interval test, the same at a headwater and at the mouth of the Amazon, so the catchment and its divide follow from two stored numbers. The walk is per region; a region holds whole basins, so no upstream relation crosses one. The index encodes the flow-direction raster losslessly and therefore carries MERIT Hydro's terms.

In the data: dfs_in_region*_90m.tif, dfs_out_region*_90m.tif, Int32, nodata −1, ODbL 1.0.

See upstream and downstream, catchment, licence.

One 3,600th of a degree; the grids are stated in it. On the ground one arc-second of latitude is about 31 m everywhere; one arc-second of longitude is about 31 m at the equator and shrinks with the cosine of the latitude, so a pixel is a square only in degrees.

See native resolution, pixel (cell, grid cell).

Two uses. Slope is an angle in degrees, 0 flat, 90 vertical. Longitude and latitude are decimal degrees on WGS 84, negative west of Greenwich and south of the equator.

In the data: mean_slope_upstream, std_slope_upstream; outlet_lon, outlet_lat, bbox_minlon …, label_lon, label_lat.

See upstream attributes, arc-second (″).

The shape ratios — convexity, hypsometric integral, elongation ratio, Gravelius compactness, circularity ratio, lemniscate ratio — and every id, index and level: basin and region ids, the upstream index, topological level, downstream depth, color_id.

See upstream attributes.

Lengths along the ground: catchment perimeter, basin length, basin width.

In the data: perimeter_catchment, length_basin, width_basin.

See upstream attributes.

Elevations, relief, river width, and the length of a crossing between two pieces.

In the data: mean_elev_upstream … relief_basin, elv, wth, hnd, crossing_length_m.

See upstream attributes.

fill_percent: the share of a window that its region or group actually covers.

See window (bounding rectangle).

A number of squares of a grid, no unit. The columns are named <what>_grid_cell_count whether they count native pixels or cells of a simplified grid.

In the data: basin_grid_cell_count, region_grid_cell_count, window_grid_cell_count, coarse_grid_cell_count; MERIT basin_cell_count, region_cell_count, upg.

See pixel (cell, grid cell).

The area of a whole-degree rectangle counted in degrees of longitude times degrees of latitude, whatever its size on the ground: 160 square degrees is 16° × 10° or 20° × 8°, 2,073,600,000 pixels at 1 arc-second. Capacities are stated in it because the memory a window takes depends on its pixel count, not on its area on the ground.

In the data: directories 160deg2, 80deg2, 40deg2.

See capacity, window (bounding rectangle).

Areas of a basin, a piece, a region, or the catchment of a pixel: the sum of the areas of its pixels, each pixel with the area it has at its latitude, never measured from the simplified outlines.

In the data: basin_area_km2, piece_area_km2, area_km2, upa.

See upstream drainage area, vectors.

The region table's counterpart for a grouping: one row per basin group with its unit code, the number of basins it holds, its window and a label point.

In the data: gat_<continent>_30m_l2.csv, gat_…_l3.csv; group_id, group_kind, level2_code / level3_code, basin_count, land_grid_cell_count.

See basin group, region table.

The raster in which every land pixel carries the id of its basin. One per dataset at the native resolution; MERIT also has one per region, and a remapped one numbered with the local basin id.

In the data: bsn_global_90m.tif, bsn_region*_90m.tif, bsn_region*_90m_remapped.tif; bsn_<continent>_30m.tif; UInt32, nodata 0.

See basin, basin id.

One row per basin in id order, with its outlet, pixel count, area, bounding box and — at 30 m — its group in each grouping and its region in each partition. The full MERIT table on Zenodo has one row for every outlet, 24,587,290 in all; the site serves the 340,991 basins of at least 1 km², all 27 columns. The 30 m tables keep the basins of at least 1 km².

In the data: bat_global_90m.csv (Zenodo), bat_global_90m_gt1km2.csv, bat_region*_90m.csv; bat_<continent>_30m_ge1km2.csv.gz.

See basin, basin id, CSV.

A colour number carried by every polygon of the basin, basin group and region files: 1 to 8, chosen so that no two neighbouring polygons share a number, and −9999 (0 in the 90 m files) where a row has no polygon. Any map colours the polygons by it without choosing colours itself; which hue a number gets is the map's choice, the 90 m GeoPackages carry the palette as their default style. The numbers are given per file, so the same basin may carry another number in another file.

In the data: color_id in the bsn_, grp_ and rgn_ vector files and in the map tiles; the palette: 1 orange, 2 sky blue, 3 bluish green, 4 yellow, 5 blue, 6 vermillion, 7 reddish purple (Okabe and Ito), 8 purple, with three reserve entries.

See vectors.

The rasters are compressed inside the GeoTIFF: DEFLATE in the MERIT-FullBasin v1 archive, ZSTD in v2. The per-region zips are stored uncompressed, since the GeoTIFFs inside are already compressed; the bytes sit in the zip unchanged, so GDAL can read one window out of a zip on Zenodo without downloading it. Read the compressed size in the file tables, not the raw one.

See GeoTIFF, region package, reading one window (HTTP).

The format of every table: plain text, one row per object, a header row of column names. The 30 m basin tables are gzip-compressed (.csv.gz).

See basin table, region table.

Each raster has one type and one NoData value: UInt16 for the MERIT v1.1 region mask (UInt8 in v1.0; NoData 0), UInt32 for the basin masks (0), Int32 for the upstream index (−1), Float32 for the sixteen attributes (−9999). MERIT Hydro's dir uses 247.

See GeoTIFF.

A prefix says what the file is: bsn basin mask or polygons, bnd basin boundaries, rgn region mask or outlines, grp basin group outlines, bat basin table, rat region table, gat basin group table, pat piece table. Then the scope (global, region*, a continent), the resolution (90m, 30m; absent on vectors), and where needed the filter (gt1km2, ge1km2) or the partition (l2, l3, 160deg2). The wildcard * is the region id written with five digits, region_00064.

See basin mask, basin table, vectors, region package.

The format of the 90 m vectors: one file holding the polygons or lines with their attributes and, for the basin, group and region files, the colour palette as the default style. Opened directly by QGIS, GDAL and GeoPandas.

See vectors, GeoParquet, color_id.

The format of the 30 m vectors: a Parquet file whose last column holds the geometry as WKB, longitude and latitude on WGS 84, the rows in id order, zstd-compressed. Opened by QGIS 3.28 and later, GDAL 3.5 and later with the Parquet driver, GeoPandas (read_parquet), DuckDB and R. A few times smaller than the same rows as a GeoPackage.

See vectors, color_id, GeoPackage.

Every raster is a GeoTIFF on the native grid, co-registered with the baseline data so that the layers line up pixel for pixel. The MERIT-FullBasin v1 archive stores them DEFLATE-compressed, one strip per row; the copies on this site, and the v2 archive, are Cloud-Optimized GeoTIFFs — tiled, with overviews — so that one window can be read over HTTP.

See reading one window (HTTP), compression, data type (NoData).

The vector tiles the site's maps draw from: the simplified basins, groups and regions of a continent in one PMTiles file, read by range requests, carrying the same ids and color_id as the vector files. Not the producer's distribution tiles.

In the data: bsn_rgn_<continent>.pmtiles.

See vectors, tile.

One row per piece of a cut basin in a partition: the region it lies in, its outlet pixel, the inlet pixel of its parent piece, and the step between them. The region order follows from it.

In the data: pat_<continent>_30m_<partition>.csv.

See piece, region order (topological level).

Every file on this site and on Zenodo is served with HTTP range requests, so GDAL, rasterio and DuckDB can read a window straight off the network without downloading the file: open https://data.fullhydro.org/merit-hydro-90m/flowdivide/v1.1/bsn_global_90m.tif with rasterio and sample a coordinate. For Zenodo set GDAL_HTTP_USERAGENT; without a user agent the request comes back 403.

See GeoTIFF, compression.

The raster in which every land pixel carries the id of its region, one per partition. MERIT v1.1: one global raster of 90 regions, UInt16. HydroSHEDS v2.0: one raster per continent for each of the three capacities; the basin groups are given by their tables and outlines instead.

In the data: rgn_global_90m.tif, UInt8, nodata 0; rgn_<continent>_30m_<partition>.tif.

See region, partition, island group.

On Zenodo, one zip per region holding that region's rasters and its basin table — the unit of download for the per-pixel layers. The published Zenodo record is MERIT-FullBasin v1.0, with 96 regions; version 1.1 has 90.

In the data: region_00064.zip.

See region, DOI, compression.

One row per region of a partition: its window, its pixel count, the number of basins and pieces it holds, its kind (whole basins, or the pieces of one cut basin), its topological level, and a point inside it where its number goes on a map. The MERIT v1.1 table has 90 rows; a second copy adds the MERIT Hydro tiles each region needs.

In the data: rat_global_90m.csv, rat_global_90m_with_tiles.csv; rat_<continent>_30m_<partition>.csv; label_lon, label_lat.

See region, window (bounding rectangle), basin group table.

The basin polygons, their boundaries as lines, and the group and region outlines, simplified for clipping and for the map: GeoParquet for the 30 m dataset and for the MERIT v1.1 region and group outlines, GeoPackage for the MERIT basin polygons. The outlines are simplified; the numbers are not — every attribute is computed on the native grid. One row per basin, group or region, the same rows as the table, so that the row of an id is the same in the table and in every file; a basin too small to hold a cell keeps its row with no geometry. A file carries the id, the size and a location for the object — the outlet of a basin, a label point for a group or region — and the columns of the simplification (coarse_grid_cell_count, cells_per_degree, the cell rectangle, color_id); everything else is in the table. Where a basin sits enclosed inside another it shows as a hole in the polygon; the boundary file has no holes. On MERIT Hydro, 36 of the 340,991 basins are too small to survive the simplification and have no polygon.

In the data: bsn_global_gt1km2.gpkg, bnd_global_gt1km2.gpkg, rgn_global.gpkg; bsn_<continent>_ge1km2.parquet, bnd_…, grp_<continent>_l2.parquet, grp_…_l3.parquet, rgn_<continent>_<partition>.parquet.

See color_id, map tiles, GeoParquet, GeoPackage.