Hong Kong Road Gradient Report

Summary

This report uses the Transport Department's official Road Network (2nd Generation) centrelines with the Lands Department 5 m DTM, splitting every road into 100 m segments and computing longitudinal gradients, counted per travel direction (two-way roads once in each direction). All work is in Hong Kong 1980 Grid (EPSG:2326).

Last updated: 2026-09-14 · Road data revision cut-off: 2026-08-27 (Transport Department)

1. Data

Tunnel handling (important)

The DTM only represents the ground surface, so no valid elevation exists for road inside a tunnel (sampling returns the mountain above it, producing bogus gradients). All in-tunnel segments are therefore removed. Only 277 features in the official data are named "…Tunnel" (e.g. Aberdeen Tunnel, Cross-Harbour Tunnel); Tai Lam Tunnel, Eagle's Nest Tunnel, Lung Shan Tunnel, Tseung Kwan O–Lam Tin Tunnel and Nam Wan Tunnel are stored under the connecting road's name, so a name filter cannot find them.

Solution: extract all 3,808 road tunnel geometries from OpenStreetMap (tunnel=yes, railway tunnels excluded), reproject to EPSG:2326, buffer by 15 m, and drop any segment whose midpoint falls inside the mask — 252 km removed, matching OSM's 255.9 km of tunnelled road (a 30 m buffer would have wrongly removed ~125 km of surface roads beside tunnels).

After this step, zero segments remain inside tunnels.

2. Method

2.1 Segmentation and sampling

Each centreline is resampled into 100 m segments (51,498 in total). Elevations at both ends are read from the DTM by bilinear interpolation; gradient = Δelevation ÷ horizontal length × 100%. The long baseline greatly reduces the effect of the DTM's ±5 m vertical accuracy.

To avoid dropping short links, any road or connector shorter than 100 m is kept as a single segment (minimum 10 m), and each road's tail is retained. Total line length is therefore 3,451 km — about 93% of the official network (~3,887 km); strict truncation would cut coverage to roughly 58%.

2.2 Directional counting (two-way counted twice)

• Two-way (TRAVEL_DIRECTION = 1, 998 km): each segment yields both +g and −g, and its length counts twice.
• One-way (= 3, 2,453 km): counted once, sign along the digitised direction (= direction of travel).

Line length 3,451 km becomes 4,448 km directional. The distribution is symmetric (p05 = -10.13%, p95 = 10.07%), confirming the treatment.

2.3 Smoothing and binning

DTM sampled after 25 m NaN-aware box smoothing. Bins are 1% wide, centred on whole percentages (0 = −0.5% to +0.5%); values beyond ±10% are pooled into the two overflow bins.

2.4 Tooling

No QGIS package exists for this ARM64 host, so the GDAL Python stack is used (pyogrio, rasterio, scipy, pyproj) — the same GDAL algorithms QGIS uses internally.

3. Results

3.1 Territory-wide gradient map

Red = uphill, blue = downhill. Two-way roads are coloured by their uphill (positive) gradient; one-way roads keep the sign of the legal direction of travel. Basemap: hillshade of the Lands Department 5 m DTM; water and coastline from OpenStreetMap.

Download high-resolution version (3000 × 2100 PNG)

3.2 Absolute steepness version (colour scale 0–10%)

Download high-resolution version (3000 × 2100 PNG)

3.3 Signed gradient distribution (100 m segments)

3.4 Full bin table

Negative = downhill, positive = uphill; x labels are bin centres in whole percent.

3.5 Cross-check: official vs OpenStreetMap (both 100 m)

Two independent road datasets give close results (medians differ by 0.14 percentage points), indicating the findings are robust.

3.6 Exceedance

3.7 Steepest roads (official names)

4. By road class (OSM supplement — the official data has no class field)

5. Reference: road speed limit map

For reference, the map below shows the statutory signed speed limits from the Transport Department's Road Network (2nd Generation) SPEED_LIMIT layer:

• Red 30 km/h, orange 70, green 80, blue 100, purple 110 (signed sections)
• Grey = the rest of the network — Hong Kong's statutory default speed limit is 50 km/h for roads without signs, so the layer only covers 860 km (4,338 sections) of signed road

Download high-resolution version (3000 × 2100 PNG)

Speed limits reflect the 2026-08-27 revision and include in-tunnel sections (speed limits are statutory and apply inside tunnels too — unlike the gradient analysis, which must exclude them). Limits can change under temporary traffic arrangements.

6. Limitations

• Roads and connectors shorter than 100 m are kept whole, so a minority of segments are shorter than 100 m (minimum 10 m); total length 3,451 km is about 93% of the official network.
• DTM vertical accuracy ±5 m: much less influential at a 100 m baseline, though bridges and retaining walls can still produce outliers.
• All in-tunnel segments excluded (252 km, verified against an OSM tunnel mask); bridge decks and elevated roads remain included, because the DTM records them as terrain.
• Gradients are average slopes over 100 m and do not capture very short steep pitches; travel directions reflect the 2026-08-27 revision, so temporary traffic arrangements may not be captured.

7. Reproducibility

• grade_gov.py 100 — official FGDB → 100 m segmentation → DTM sampling → directional gradients
• plots_100.py — binning, histogram, official vs OSM comparison
• make_map.py — gradient maps (hillshade, scale bar, urban detail inset)
• make_speed_map.py — speed limit map (SPEED_LIMIT layer)
• Data: segments_gov_100m.csv, summary_gov_100m.json, summary_compare_100m.json
• Segment-level GeoJSON/CSV (by street or district) available on request

Questions, or want a specific street or district analysed? Contact us.