The Process: A Thousand Judgements.

Bark & Beam pieces are among the most geometrically accurate physical representations of global population and economic distribution that can be rendered at this scale. They are computationally derived and materially executed — equal parts data artifact and crafted object.

  • It all begins with the selection of high quality datasets.

    • Global, gridded population datasets are used instead of country totals as country borders obscure how people actually occupy space.

    • Spatially distributed GDP (PPP) data is used rather than national GDP figures as economic activity is uneven within countries and must be spatially resolved.

    • Datasets span billions of individual data points at kilometer-scale resolution. This resolution preserves urban concentration, corridors, and gradients.

    • The Earth is reprojected using an equal-area map projection. Most world maps distort area, exaggerating high latitudes and minimizing the tropics.

    • Each grid cell represents the same physical land area. This allows population and economic values to be compared meaningfully across latitudes.

    • The projected world is divided into a fixed rectangular grid. This Discretization allows continuous data to be translated into physical units.

    • Each cell represents a defined geographic footprint. This means every block corresponds to a real place.

    • Cells below a land-area threshold are excluded. Oceans remain intentional negative space.

    • Population values are summed within each grid cell. Density is computed per physical area, not per country.

    • GDP values are aggregated within the same cells. Economic output is spatially uneven.

    • Tercile binning is applied across land cells only as relative differences matter more than absolute extremes.

    • GDP values are logarithmically scaled before binning as global economic output spans orders of magnitude.

    • Three different wood species encode population density. This allows for material variation to be perceived without explanation.

    • Relief height encodes economic output. Vertical difference conveys magnitude intuitively.

    • Each block is milled to a fixed footprint and assigned height. The dimensional accuracy preserves spatial relationships.

    • Blocks are assembled by hand onto a rigid backer. Alignment matters; no block is placed arbitrarily.

    • Final pieces are framed and mounted using a concealed cleat system. The work is meant to float, not hang like a poster.

Technical Specifications

This section details the computational pipeline used to generate Bark & Beam reliefs. It is intended for readers familiar with spatial data, cartographic projections, and quantitative aggregation.

Data Sources

1

Population data

  • Format: Global raster (GeoTIFF)

  • Native resolution: ~30 arc-seconds (≈1 km at the equator)

  • Approximate grid size: ~43,200 columns × 21,600 rows

  • Total raster cells: ~930 million

  • Year used: Single snapshot (2020) for the current reliefs

GDP data

  • Format: NetCDF (time series)

  • Native resolution: 5 arc-minute grid (≈10 km)

  • Approximate grid size: ~4,320 columns × 2,160 rows

  • Total raster cells: ~9.3 million per year

  • Year used: Single snapshot (2015) for the current reliefs

Spatial Reference System

2

All datasets are reprojected into Equal Earth (EPSG:8857) for analysis and gridding so grid cells remain spatially comparable at global scale.

All inputs are clipped to latitude ≥ −60° prior to gridding/aggregation to remove Antarctica from the dataset and the physical build

Land cell threshold = 10% land fraction to avoid spurious micro-islands while retaining meaningful coastal structure

Density uses land area, not total cell area, to avoid coastline dilution

GDP is log-transformed before tercile binning to prevent extreme concentration from flattening the relief

Rank-based terciles provide stable, balanced physical encoding under heavy-tailed global distributions

Grid Construction

3

Bark & Beam uses a fixed rectangular grid per version. Each cell becomes either:

  • Ocean cell (excluded from fabrication), or

  • Land cell (fabricated as a hardwood block)

Versions (grid resolution)

  • COURSE: 40 × 20 grid → 800 total cells

  • FIELD: 52 × 26 grid → 1,352 total cells

  • SURVEY: 80 × 40 grid → 3,200 total cells

Countries are intersected with grid cells to compute land area per cell

For each land cell polygon, population is aggregated as a zonal sum over the population raster. GDP is aggregated as a zonal sum over the GDP rasterized surface

Population density is computed as “people per km² of land within the cell”, not per total cell. This choice prevents coastal cells or fragmented land cells from being artificially diluted by ocean area

4

Zonal Aggregation

Normalization & Binning

5

The objective is a stable, interpretable three-level physical encoding rather than a continuous (and visually fragile) mapping

All binning is performed on land cells only (ocean cells remain zero-coded)

Land-cell population density values are converted into rank percentiles, then split into terciles (1/2/3). Rank-based binning is chosen for robustness to heavy-tailed distributions and to preserve balanced material representation across the world grid

Land-cell GDP totals are transformed via log10(GDP + 1), then converted into rank percentiles and split into terciles. The log transform prevents a small number of extreme GDP cells from collapsing the visual dynamic range

Material Encoding

6

Each land cell encodes two variables simultaneously:

Species (Population density tercile):

  • Bin 1 → Maple

  • Bin 2 → Cherry

  • Bin 3 → Walnut

Height (GDP tercile): Discrete and version-scaled:

  • COURSE: 1.00″ / 2.00″ / 3.00″

  • FIELD: 0.75″ / 1.50″ / 2.25″

  • SURVEY: 0.50″ / 1.00″ / 1.50″