Team course project (2 people)

Dunkirk heatwave sensitivity

Université de Technologie de Compiègne · UR06 Geomatics (autumn 2023)

An ArcGIS study that decomposes 'why cities heat up' into three continuous fields — building heat, vegetation cooling and water cooling — combines them with two-pass kernel density and multi-criteria weighting, and produces a heatwave sensitivity index for Dunkirk. The building effect is about 3.3× vegetation and 55× water.

Background

This was the UR06 “Geomatics” course project at UTC (A23), done in a team of two. The brief required using only raw geographic layers (BD TOPO, ADMIN EXPRESS, Enedis, CORINE) — no pre-built heat-island or LCZ layers — and building the whole vector-to-raster chain in ArcGIS, ending with a heatwave sensitivity map of Dunkirk and a report of about 20 pages.

The original report is lost. The documents in the repository were reconstructed from the .mxd, the shapefiles, raster statistics and the conversations of the time; every element is tagged [C] confirmed / [D] inferred / [to be completed].

Data

  • BD TOPO (IGN): 43,912 buildings, 1,131 vegetated areas, 626 water bodies, 254 + 307 hydrographic segments, tidal flats and the land/sea boundary;
  • ADMIN EXPRESS: municipal boundaries and population;
  • Enedis / open data: electricity consumption for 42 IRIS units, joined and converted into consumption per unit area;
  • CORINE Land Cover 1990 / 2000 / 2006 / 2012: diachronic analysis, artificialisation rising from 74.2% to 78%.

Method

  1. Project everything to RGF93 / Lambert 93 (EPSG:2154, metres), otherwise areas and distances are meaningless;
  2. Union + Feature To Point: each building or green area contributes once, weighted by its own area;
  3. Two-pass kernel density (small then large radius): a single kernel produces a bell shape and cannot express both “strong at contact” and “weak at distance”; combining two passes gives a gradient that decays with distance;
  4. Mask to the municipal boundary;
  5. Normalise, invert the limiting criteria, weighted sum (multi-criteria analysis, MCA):
S = 0.50 · B* + 0.33 · (1 - V*) + 0.17 · (1 - E*)

Buildings are aggravating, vegetation and water are limiting, so the latter two must be inverted; otherwise more vegetation would increase the computed sensitivity — the classic sign error in multi-criteria analysis.

Results

Statistics read directly from the .aux.xml delivered with the rasters:

RasterCellMaxMean
Building heat effect35.5 m0.06190.00754
Vegetation cooling33.5 m0.01900.00438
Water cooling44.4 m0.00110.00015

The maximum building effect is 3.3× vegetation and 55× water. Dunkirk is a territory dominated by buildings, with fragmented vegetation and localised water: 78% of the surface is artificialised, 39% of it industrial and transport land.

The most sensitive areas are the dense city centre, the port-industrial front and large facilities; the least sensitive are the large basins, dunes and tidal flats, and the agricultural fringe to the south. The report’s key point is not “where it is hot” but where compensation is missing: places with strong building effect and weak vegetation accumulate both penalties, and those are the priorities for public action.

Limitations

  • The original report is lost and docs/ is a reconstruction; the weights (0.50 / 0.33 / 0.17) and kernel parameters (search radius, kernel function) were not preserved and are now defensible reconstructions, not original values;
  • the three rasters have different cell sizes (35.5 / 33.5 / 44.4 m), so combining them triggers implicit resampling — a real flaw of the original work, discussed critically in the report;
  • the figures were re-rendered from the rasters, not exported from the original ArcMap layout;
  • there is no field validation, and the index is static: it ignores population exposure and behaviour.

Report (FR/ZH), technical appendix, ArcMap project and re-rendering script: https://github.com/yicongl1/Dunkerque-Heatwave-Sensitivity

Technical notes

01 Splitting 'why cities heat up' into three continuous fields: kernel density and multi-criteria weighting RGF93 projection, Union + Feature To Point, two-pass kernel density, clipping, and why vegetation and water must be inverted. 02 After the report is lost: rebuilding reproducible research from project files Reconstructing the method from the .mxd, shapefiles and raster statistics, marking confidence with [C]/[D], and which parameters are gone for good.

Related skills

  • ArcGIS
  • Academic writing
  • Git