\( \newcommand\D{\mathrm{d}} \newcommand\E{\mathrm{e}} \newcommand\I{\mathrm{i}} \newcommand\bigOh{\mathcal{O}} \newcommand{\cat}[1]{\mathbf{#1}} \newcommand\curl{\vec{\nabla}\times} \newcommand{\CC}{\mathbb{C}} \newcommand{\NN}{\mathbb{N}} \newcommand{\QQ}{\mathbb{Q}} \newcommand{\RR}{\mathbb{R}} \newcommand{\ZZ}{\mathbb{Z}} \)
UP | HOME

Geography

Table of Contents

1. Introduction

Suppose we want to model a planet's atmospheric dynamics, or we want to simulate a fictitious world (like Dwarf Fortress). First, we must encode the geography of the planet on a computer. How can we do it?

Fans of the Civilization game series may think of a hex grid. This is one approach, but I am lazy and want something which can be shown in the terminal.

Thus we use a more coarse approximation using square tiles on a cube. We know the circumference of the Earth is approximately 40,000 kilometers (from this historic definition of the kilometer). This gives an approximate surface area of 509.295818e6 square kilometers. We can construct a cube with approximately the same area, its side length \(\ell\) then satisfies

\begin{equation} 6\ell^{2}=509.295818\times 10^{6}~\mathrm{km}^{2}. \end{equation}
(let ((radius (/ 20e3 pi)))
  (sqrt (/ (* 4 pi (expt radius 2))
           6)))
9213.177319235614

This gives \(\ell\approx 9213\) kilometers. More generally, for a planet with radius \(R\) units, we would have \(\ell\approx R\sqrt{2}\).

One face will contain the North pole, its opposing face will contain the South pole. The remaining four faces contain the equator.

1.1. Tiling Each Face

We then subdivide each face into square Region tiles, and subdivide each region into Local Area tiles. (I borrow this terminology from Dwarf Fortress.)

Suppose we have 8-by-8 region tiles for a face (for a total of 64 regions per face, and 384 regions on the planet). This would require each region having an area of \((\ell/8)^{2}\) — for Earth, this is 1,326,291 square kilometers (slightly larger than the combined areas of California, Nevada, Arizona, and Utah).

(defun sq (x) (* x x))

(let* ((radius (/ 20e3 pi))
       (side-length (sqrt (/ (* 4 pi (expt radius 2))
                             6))))
  (list :1-byte (sq (/ side-length 8))
        :2-byte (sq (/ side-length 16))
        :4-byte (sq (/ side-length 32))
        :8-byte (sq (/ side-length 64))))
- :1-byte
- 1326291.1924324615
- :2-byte
- 331572.7981081154
- :4-byte
- 82893.19952702885
- :8-byte
- 20723.29988175721

We see for 16-by-16 regions on a face, each region is approximately 330,000 square kilometers (about the size of New Mexico [314,160 km2]); for 32-by-32 regions on a face, each region is about 83,000 square kilometers (about the size of Maine [80,005 km2]); for 64-by-64 regions on a face, each region is approximately 20,000 square kilometers (slightly larger than the land area of New Jersey [19,047 square km]).

For context, many war games — like Gary Grigsby's War in the East 2 — use 10 nautical miles for each tile length (about 18.5 kilometers). If we try to make each local area tile have side length 18.5 kilometers, then we find one face of the "planet cube" consists of 498-by-498 local area tiles. We can observe \(512=2^{9}\approx 498\) and thus if we choose to work with 512-by-512 local area tiles per face, then 1 local area tile is approximately 18-by-18 kilometers [323.8 square km] (roughly twice the area of Washington, D.C. [158 km2]; or, if you prefer, a little smaller than the city of Detroit [351 square kilometers]).

We can compare this to various crude discretizations used in climate models, where the horizontal resolution is roughly 100 kilometers (and vertical resolution is 95 levels). The horizontal resolution changes with latitude, but it gives us a sense of size (about 5 times larger than our local area tile's edge length).

Do we really need all these local area tiles? It depends on what degree of error and accuracy we wish. For a region tile which is entirely Ocean, arguably, we do not need local area tiles. It depends on what we want to accomplish with our encoding.

Observe then that the planet would require \(6\times 2^{18}\approx 1.5\times 10^{6}\) local area tiles. If we have \(n\) bytes of data per tile (for, e.g., average temperature, precipitation, terrain, or whatever), then we need something on the order of \(2n\) Megabytes of memory to store it.

Last Updated 2022-11-17 Thu 18:47.