DEV Community

Future Built AI
Future Built AI

Posted on

CityEngine, Site3D, Archshaper, GenAnywhere, Cityweft: A Quarry Park Pipeline

This is the build order I used to model a fourteen-acre quarry park, with the inputs, outputs, and checks for each platform. The site is an abandoned limestone quarry with a three-acre lake, fifty-five-foot cliffs, and a residential ridge on the north side. The program is a rim trail, switchbacks, terraces, an 1,800-square-foot pavilion, and a 58-space gravel lot.

Constraints before tools

Four numbers drove every decision: trail grade max 8 percent, road grade max 7.5 percent after the first pass hit 11, terrace risers at 18 inches, and no stormwater leaving the site untreated. I wrote those into a text file and kept it open for the whole project. Every model was checked against that file before it went to anyone else.

Step 1: CityEngine for repeated geometry

Input: the rim centerline from the survey, plus a CGA rule file for railing, boardwalk, steps, and terrace walls. Output: generated geometry along the curve with consistent spacing. Check: rule parameters matched the terrace spec, and the generated step count matched the hand calculation within one riser. The win here is regeneration. The alignment changed four times; the rule did not.

Step 2: Site3D for alignment and earthwork

Input: the access road from the county road to the lot, the trail profile, and the drainage outlets. Output: corridor surfaces, cross sections, earthwork volumes, and two culvert locations. Check: max grade, sight distance at the entrance, and the swale capacity for a 25-year storm. This is where the first design failed. An 11 percent run for 200 feet passed on screen and would have failed every winter. The rebuild added one curve and removed four stalls.

Step 3: Archshaper for the pavilion

Input: the massing block from the layout model, roof pitch, and a short material brief — quarried stone base, timber trusses, standing-seam roof. Output: building options with structure, openings, and roof geometry. Check: truss spans, door clearance, and solar orientation. The third option survived almost intact. The glazing changed from west-facing to south-facing after a simple check of afternoon sun angles.

Step 4: GenAnywhere as the fast sanity model

Input: a rectangle drawn on the map covering the quarry and two blocks of houses. Output: a textured massing model in roughly five minutes, with terrain, water plane, building blocks, and coarse vegetation. Check: the bowl profile against the survey spot elevations, and the water plane against the known pool level. It is not a design model and it is not survey accurate. It is the model I show first, because it is cheap enough to throw away.

Step 5: Cityweft for context and sightlines

Input: the park model plus the surrounding town. Output: streets, roofs, tree heights, and terrain in one scene. Check: views from the ridge, the approach from the south, and shadow behavior in December. This step changed the pavilion orientation by ninety degrees. The roof would have clipped the ridge line seen from four houses, and the fix cost nothing at this stage.

Handoffs that broke

• Scale. The GenAnywhere model came in at the wrong size until I tied it to the surveyed rim line.
• Units. The Site3D profile exported in feet while the terrace rule expected inches for risers. One regeneration produced a wall three times too tall.
• Versioning. Two trail alignments circulated for a day because I saved them under the same name.
• Terrain. Cityweft terrain did not match the survey at the north wall, which shifted a sightline check by six feet.

What I would automate next

I would script the grade check so it runs against every exported alignment instead of relying on me to open the profile. I would also bake the four constraint numbers into the rule file, so a bad parameter fails at generation rather than at review. The versioning mistake would go away with a pre-commit hook that stamps the terrain file hash into each model export.
None of this is complicated modeling. It is bookkeeping with geometry attached, and the platforms only help if the numbers between them stay honest. The one check I would never automate is the walk. I still put on boots and look at the rim from the road, because the model has never once told me how the wind feels at the overlook.

Top comments (0)