A watershed as a shared,
testable world.
Terrain, water, vegetation and resources are represented as changing entities. Their influence rules, interventions and observations remain connected as the landscape evolves.
The world is being calculated through the Gongjing core.
One evolving system, kept scientifically inspectable.
A domain calculation becomes more useful to people and scientific AI when its state, assumptions, interventions and evidence remain connected.
Shared representation
Entities, states, resources, influences and constraints use a common structure that a person or software agent can inspect.
Controlled intervention
Change one permitted rule while preserving the starting world, then compare the resulting trajectory.
Traceable observations
Every displayed measure is tied to a model state, scale, parameter set and provenance record.
From a watershed model to a shared scientific world
The watershed is one replaceable domain module inside the Gongjing architecture.A hydrology calculation alone produces model states. Gongjing organizes how those states can be described, observed, changed and compared. Its shared core provides the same commands for any compatible domain module. Here, a watershed adapter supplies simplified rules for flow, terrain change, vegetation and resources.
Each update follows a feedback loop: terrain and vegetation redirect water; water changes terrain and vegetation; the next calculation uses the changed world. Gongjing retains the parameters and observations needed to inspect that loop.
Same starting world + one changed influence rule = a controlled computational comparison. This is the central scientific workflow demonstrated here.
| Worldspace concept | Meaning in this case study |
|---|---|
| Entities and states | Terrain surface, routed water, vegetation and stored resources, each with a state that changes over time. |
| Influences | Terrain routes water; water lowers terrain and affects vegetation; vegetation resists flow and stabilizes terrain. |
| Resources | Locally available water and stored resources can support vegetation, depending on flow conditions. |
| Agency | A person now chooses interventions through the interface. A scientific AI could use the same defined commands, but autonomous research is not claimed here. |
| Constraints | Water enters across the grid, moves to downstream cells and leaves through an open lower boundary. State values stay within declared limits. |
| Observations and scale | Cell fields are summarized into flow-path candidates and watershed-level measures without discarding the underlying state. |
A reproducible case: does erosion reorganize flow?
Use landscape number 0. Compare Full feedback with No erosion at step 280. Terrain and vegetation begin in the same state; the erosion rule is the controlled difference.
| Condition | Average ground lowering | Flow in busiest 5% of cells |
|---|---|---|
| Full feedback | 0.103 model units | 47.9% |
| No erosion | 0.000 model units | 29.9% |
In this model comparison, erosion changes the terrain and produces more concentrated flow. Plants can still redirect water when erosion is off. Gongjing makes the comparison repeatable by keeping the starting state, settings, intervention and observations together.
The flow measure adds routed flow across cells below the first eight rows. It describes the distribution of calculated flow within the grid, not separate volumes leaving a real watershed.
Open Gongjing worldspace parametersInspect the current world, expose selected influence rules, and manage its experiment record.
Inspect the Gongjing architecture and live calls
Shared core → watershed module → observation packet. The core uses six domain-independent commands. The view draws only the returned observation fields, while domain calculations remain in the watershed module.
| Command | Purpose |
|---|---|
| initialize | Create a world from a landscape number and parameter set. |
| step | Calculate its next state. |
| observe | Return fields, measurements, scale and provenance. |
| describe | Expose entities, influences, constraints and assumptions. |
| intervene | Change an allowed parameter, restart or restore a state. |
| coarseGrain | Summarize connected cell-level patterns at a larger scale. |
Waiting for observations.
Scientific scope and implementation checks
This is an architecture proof of concept with a functioning landscape-evolution module. It is not presented as a calibrated watershed forecast. Its scientific purpose is to demonstrate how an evolving domain model can be made inspectable, intervenable, comparable and reproducible through Gongjing.
The default landscape and numerical rules come from the original watershed showcase benchmark. The live JavaScript calculation has been checked against its Python reference through 280 steps. All displayed quantities use relative model units.
Water is routed to three positions in the next grid row, with uniform input and an open lower edge. The module is a downstream routing approximation, not a full two-dimensional fluid solver. It does not represent a measured watershed, storm timing, water storage or a sediment transport budget. Ground lowering is capped, and smoothing acts on the lowering field. Candidate paths are cross-scale descriptions of model fields, not mapped real rivers.
Modular landscape modeling has a broader scientific background, including Landlab v2.0. This demonstration uses its own compact watershed module and does not run Landlab.