From terrain.drs exists a series of terrain files. File 0 may be dirt, File 1 grass, and so on.

Each File has roughly a hundred animation frames that, when unpacked and lined up, create one large tile roughly ten times the area.

Programmatically, map data is stored in a 2D array of integers. At runtime, this 2D array is projected isometrically onto the screen like:

+--------+        /d\
|a b c d |       /c h\
|e f g h | -->  /b g l\
|i j k l | --> +a f k p+
|m n o p |      \e j o/
+--------+       \i n/

Given array values a -> p are all integers, and equal to 1, they will all render a mega grass tile.

The animation frame index to use, however, is dictated by:

int bound = sqrt(tile_count) // Roughly 100 for tile_count
int index = (x % bound) + ((y % bound) * bound);

Given the width and height of each tile is known, and a projection transformation of:

int xx = (y + x) * scale.width / 2;
int yy = (y - x) * scale.height / 2;

Where x and y are map coordinates, and xx and yy projected isometric coordinates, the final render with a mix of grass and dirt tiles may look something like:

Of course, this does not include tile blending.

Likewise, if a point is to be clicked on screen, the reverse projection calculations are to be applied to query the map.