Patch Generating Land Use Simulation

93 concepts68 papers1 proposal

Traditional cellular automata (CA) models simulate land-use change by evaluating individual pixels. They look at a single cell, calculate its suitability based on local drivers, and decide whether it transitions to a new state (such as from cropland to urban built-up land). While computationally straightforward, this pixel-by-pixel approach suffers from a major flaw: it produces highly fragmented, "salt-and-pepper" patterns that do not reflect how real-world landscapes evolve. In reality, urban expansion and ecological shifts occur in cohesive, multi-cell patches—such as industrial parks, residential subdivisions, or reforested zones.

To bridge this gap, modern predictive land-use frameworks decouple macro-level demand from micro-level spatial allocation. By combining machine learning rule-mining with patch-generation strategies, models like the Future Land Use Simulation (FLUS) and the Patch-generating Land Use Simulation (PLUS) capture both the complex socio-environmental drivers of change and the realistic, organic growth of geographic features.

The Two-Core Architecture: Demand vs. Allocation

Modern patch-based simulation models operate via a two-step decoupled architecture. First, a top-down macro-model determines how much land of each type is needed. Second, a bottom-up spatial model determines where and how that land expands.

As explained by Tiantian Huang et al., this framework integrates top-down system dynamics (which project total land demand based on population, GDP, and policy drivers) with bottom-up cellular automata. Once the total pixel budget for each land-use type is established, the spatial allocation engine takes over, driven by two distinct cores: a rule-mining engine and a patch-growth engine.

+-----------------------------------------------------------------+
|                      Top-Down Macro Demand                      |
|       (System Dynamics / Markov Chains project total area)      |
+-----------------------------------------------------------------+
                                |
                                v
+-----------------------------------------------------------------+
|                 Bottom-Up Spatial Allocation                    |
|                                                                 |
|   1. Rule Mining (LEAS / ANN)   --> Calculates suitability maps |
|   2. Patch Growth (CARS / CA)   --> Spreads cohesive land patches|
+-----------------------------------------------------------------+

1. Mining Transition Rules with Machine Learning

Before a model can grow patches, it must understand the underlying geography of development. Older CA models relied on logistic regression to calculate suitability maps, which failed to capture non-linear interactions between driving factors like topography, climate, and infrastructure proximity.

Modern frameworks solve this by using advanced machine learning algorithms to analyze historical land-use transitions:

  • Artificial Neural Networks (ANNs): In the FLUS model, introduced by Xiaoping Liu et al. and utilized by Xiaozheng Du et al., ANNs ingest multi-temporal spatial datasets to calculate the complex probability that any given cell will convert to a specific land type.
  • Random Forests and LEAS: The PLUS model introduces a Land Expansion Analysis Strategy (LEAS). As detailed by Yuheng Fu et al. and Xuemin He et al., LEAS uses a random forest algorithm to sample cells that changed between two historical periods. By analyzing variables like elevation, slope, precipitation, and distance to highways, the algorithm determines the exact operational contribution of each driver to land expansion, outputting a highly accurate growth probability map for each land class.

2. Cohesive Growth via Clustered Autonomous Random Seeds

Knowing where land is likely to change is not enough; the model must simulate how it changes. To prevent fragmented pixelation, frameworks like PLUS implement a patch-generating strategy known as the Clustered Autonomous Random Seeds (CARS) algorithm.

Instead of changing pixels independently, the CARS mechanism mimics organic growth through a seeding and competition process:

  1. Seed Planting: If a land-use type (e.g., urban) is under-allocated relative to macro-demand, the model randomly selects starting points ("seeds") on the probability map. The selection is heavily weighted toward cells with high suitability scores.
  2. Neighborhood Evaluation: Once a seed is planted, the model evaluates its immediate neighborhood using a convolution matrix. The probability of an adjacent cell converting depends not only on its own suitability but also on the number of neighboring cells belonging to the expanding land type. This creates a powerful clustering effect.
  3. Multi-Class Competition and Inertia: Real landscapes feature intense competition; urban expansion might try to swallow cropland, while ecological policies try to preserve it. To resolve this, Xiaozheng Du et al. highlight the use of a self-adaptive inertia coefficient. If a specific land type is falling short of its target demand, its inertia increases, making it harder for competing land types to overwrite it.

As Wangping Li et al. demonstrate, this patch-generating strategy allows cellular automata to accurately simulate the spatial evolution of land-use patterns at the patch level, generating realistic, contiguous shapes rather than isolated pixels.

Simulating "What-If" Policy Scenarios

Because these models separate rule mining from spatial allocation, planners can easily manipulate the simulation environment to test hypothetical policies. This is achieved by introducing spatial constraints, transition matrices, and scenario-specific demand weights.

For instance, Fangting Xie et al. and Xuemin He et al. demonstrate how models can be tuned to evaluate divergent futures:

  • Natural Development Scenario: Historical trends continue without intervention.
  • Cropland Protection Scenario: The transition matrix is modified to strictly forbid the conversion of high-yield agricultural land to urban areas.
  • Ecological Priority Scenario: Sensitive zones are designated as absolute constraints where no development can occur.

To make these ecological scenarios even more robust, researchers have begun coupling patch simulations directly with environmental networks. Wenbin Nie et al. developed the EN-PLUS model, which extracts regional ecological corridors and nodes to serve as dynamic spatial constraints. By embedding these ecological networks directly into the patch-generation phase, the simulation actively steers urban growth patches away from critical biodiversity pathways, providing a proactive tool for sustainable regional planning.

Key Foundational Work

A Spatial Modeling Framework to Evaluate Domestic Biofuel-Induced Potential Land Use Changes and Emissions (2014) — contributed 2 concept elements to this direction.
History & citationVersion 1 · 2026-10-04

Current · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_patch_generating_land_use_simulation_1f108295,
  title = {patch-generating land use simulation},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/patch-generating-land-use-simulation/rdr_1f10829526334b61a3d4b600a4939030},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-20},
}

This exact version (version 1, frozen text):

@misc{lacuna_patch_generating_land_use_simulation_1f108295_v1,
  title = {patch-generating land use simulation},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/patch-generating-land-use-simulation/rdr_1f10829526334b61a3d4b600a4939030/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-20},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake