The Geographic Automata Add-In for ArcGIS Pro

Extending ArcGIS Pro with the .NET SDK to support spatiotemporal modelling of dynamic geospatial systems

Overview

This project introduces and demonstrates the Geographic Automata add-in developed for ArcGIS Pro. The add-in was created to address the lack of generalized, flexible, explainable, and integrated geographic automata modelling tools in modern GIS software environments. Through combinations of ArcGIS Pro tools and new add-in functionality, beginners and advanced users can create simple to complex models of various dynamic geospatial systems in educational, research, and decision-making settings.


Background

Geographic automata (GA) formulation adapted from Torrens & Benenson's work [1].

Geographic Automata Systems provide a framework for modelling dynamic geospatial systems from the bottom-up [1] and continues to be a prominent research area in Geographic Information Science after 30 years [2]. The paradigm is characterized by cellular automata and agent-based models. Such approaches are useful for representing, examining, and explaining patterns and behaviours emerging from local interactions.

Key elements of a cellular automata (CA) model.

Of this paradigm, Cellular Automata (CA) models are widely used due to their theoretical simplicity, transparency, flexibility of behaviour specification, and minimal data requirements [3, 4].

The key elements of CA models include a grid of cells, cell states, neighbourhood functions, transition rules, and time [5, 6]. At each iteration or timestep of model execution, transition rules enact changes across a study area grid based on the neighbouring cell conditions of each cell.

How are CA Models Implemented?

Implementing geographic CA models typically involve programming, software tools with graphical user interfaces (GUIs), or some combination of both. Although standalone or integrated GUI-based CA modelling tools exist [7, 8], solutions are often specialized to certain phenomena or embellished with immutable sub-models. Consequently, even simple CA modelling projects have previously required patchworks of disconnected tools, with procedures severed from theoretical terminology.

Example of possible workflows for implementing geographic CA models.

Problem

Currently, there is a lack of integrated, general-purpose geographic automata modelling tools capable of supporting students, researchers, and decision-makers.

Integrated, GUI-based geographic automata modelling tools are needed to support users seeking alternatives to programming, learning about the modelling paradigm, or searching for an efficient means of implementing simple to complex models within a modern GIS software environment.


Project Objectives

  1. To introduce the Geographic Automata add-in, and
  2. To demonstrate its integration with ArcGIS Pro functionality for implementing a modern urban growth modelling procedure.

Application Example

Integrating ArcGIS Pro geoprocessing tools and the Geographic Automata add-in to model urban growth in the City of Chilliwack, British Columbia.

Study Area Description

This application example focusses on modelling urban growth in the  City of Chilliwack , located in the Fraser Valley region of southwest British Columbia. The municipality is home to approximately  101,700 people  as of 2022. The city is known for its sprawling agricultural landscape enveloped by the soaring Coast and Cascade Mountain ranges.  

Methodology

This example uses the Basic CA and Advanced CA tools to demonstrate an end-to-end modelling workflow in ArcGIS Pro 3.2.1.

Model A (Basic CA) and Model B (Advanced CA) are defined with simple experimental parameters. Both models include a novel integration of the   AutoML workflow  and CA. The procedure is inspired by previous studies using ML algorithms to create urban development “probability” or “suitability” maps to guide CA model behaviours [11, 14–16].

Overview of the methodology steps used to implement the application example.

Experimental Results

Calibration and Validation

Selected metrics from the Model Evaluation Report files for calibration and validation stages.

From the Model Evaluation Reports, Model A and Model B showed similar calibration measures. The performance of Model A produced by the simple Basic CA mechanisms was better than expected. The agreement between real and simulated urban developments for 2010 from Model A and Model B were nearly identical. The capacity of each model to project new urban developments for the validation period is where model performance starts to deviate.

Change Forecasting Performance

The number of cells counted as hits (correctly simulated urban developments), misses (real urban developments not simulated by the model), and false alarms (simulated urban developments to incorrect locations) for calibration and validation stages.

In addition to summary measures like accuracy, Kappa, or FOM, it is important to examine the quality of a model's forecasted changes. Based on FOM components including hits, misses, and "false alarms," Model B implemented with the Advanced CA workflow forecasted urban developments to more realistic locations in the validation period. Model B also simulated outcomes produced fewer "false alarms" (i.e., incorrectly simulated new developments) in calibration and validation periods.

Visual assessment of real versus simulated development forecasts for 2020 in three subareas.

Discussion

The Basic CA (Model A) simulations demonstrate the value of the rapid modelling tool for quickly parameterizing and investigating outcomes. With three built-in rule types, FOM values were competitive with the slightly more complex model. Nevertheless, it is expected that more rigorous parameter tuning of the more flexible behaviours possible through the Advanced CA (Model B) workflow will further improve calibration.

In summary, this application example demonstrated the use of the Geographic Automata add-in to complete an end-to-end CA modelling workflow within ArcGIS Pro. Using the Geographic Automata add-in's general-purpose functions, a popular approach of integrating CA with ML was demonstrated. Beyond this example, the initial version of the add-in was successfully used in undergraduate and graduate student projects. The add-in was helpful to support beginners, for whom integrated modelling tools within a familiar GIS software environment are essential for creating positive, hands-on learning experiences with this modelling framework. For advanced users, the add-in can be used to rapidly implement experiments, quickly integrate CA with popular analysis techniques, and efficiently implement baseline models for comparison with more advanced routines. Decision-makers can also benefit from the explainable, transparent model mechanisms and reporting functionality for examining possible scenarios or policies. Therefore, the enmeshment of the Geographic Automata add-in within the ArcGIS Pro software environment can support students, educators, researchers, and decision-makers representing a variety of dynamic geospatial systems.


Future Directions

Since its development, the inaugural version of the Geographic Automata add-in was successfully deployed in research and educational settings. Planned upgrades include implementing “drag-and-drop” interactivity to match the behaviour of ArcGIS Pro’s built-in tools, improving UI elements, supporting dynamic variables, expanding model evaluation techniques, and adding generalized agent-based modelling functionality.


Significance

The Geographic Automata add-in was developed with the  ArcGIS Pro SDK for .NET  to address the scarcity of integrated CA modelling tools available for students, researchers, and decision-makers. The add-in contributes to general-purpose, flexible, explainable, and approachable tools for modelling an array of dynamic geospatial systems. In combination with ArcGIS Pro tooling, the add-in supports users in exploring system behaviours, generating forecasts, examining “what-if” scenarios, and backing policy-making procedures.


Acknowledgements

I would like to thank my supervisor, Dr. Suzana Dragicevic. Her advice and support throughout developing the Geographic Automata add-in were invaluable. This work was fully supported by the NSERC Discovery Grant awarded to Dr. Dragicevic.

Software Information

The Geographic Automata add-in was developed by  Alysha van Duynhoven . Download links and an official citation are forthcoming. The add-in will be freely available and distributed as an .esriAddInX file. If you have any questions about the software or would like to discuss potential use cases, please contact  alyshav@sfu.ca .

Data Sources

All images and videos in this StoryMap were created by Alysha van Duynhoven. The photo from Mount Cheam was taken by Alysha van Duynhoven on August 2nd 2016. Open datasets used in this project are available at sources included in the table below:

Datasets

Source

Accessed

Agriculture and Agri-Food Canada (AAFC) Land Use Time Series (GeoTIFFs for 2000, 2010, and 2020)

November 7th 2023

Municipal Boundary, Parks, Roads, Schools, Water Feature (Polygons), Zoning

March 7th 2024

Railways (Polyline)

March 7th 2024

Conservation Areas (Polygon)

March 7 2024

ASTER Global Digital Elevation Model V003

March 7th 2024

References

[1] P. M. Torrens and I. Benenson, “Geographic Automata Systems,” Int. J. Geogr. Inf. Sci., vol. 19, no. 4, pp. 385–412, Apr. 2005, doi: 10.1080/13658810512331325139.

[2] X. Wu, W. Dong, L. Wu, and Y. Liu, “Research themes of geographical information science during 1991–2020: a retrospective bibliometric analysis,” Int. J. Geogr. Inf. Sci., vol. 0, no. 0, pp. 1–33, 2022, doi: 10.1080/13658816.2022.2119476.

[3] I. Santé, A. M. García, D. Miranda, and R. Crecente, “Cellular automata models for the simulation of real-world urban processes: A review and analysis,” Landsc. Urban Plan., vol. 96, no. 2, pp. 108–122, May 2010, doi: 10.1016/j.landurbplan.2010.03.001.

[4] J. Cheng and I. Masser, “Understanding spatial and temporal processes of urban growth: Cellular automata modelling,” Environ. Plan. B Plan. Des., vol. 31, no. 2, pp. 167–194, 2004, doi: 10.1068/b2975.

[5] M. Batty, H. Couclelis, and M. Eichen, “Urban systems as cellular automata,” Environ. Plan. B Plan. Des., vol. 24, no. 2, pp. 159–164, 1997.

[6] R. M. Itami, “Simulating spatial dynamics: cellular automata theory,” Landsc. Urban Plan., vol. 30, no. 1–2, pp. 27–47, 1994, doi: 10.1016/0169-2046(94)90065-5.

[7] X. Liu et al., “A future land use simulation model (FLUS) for simulating multiple land use scenarios by coupling human and natural effects,” Landsc. Urban Plan., vol. 168, no. July 2016, pp. 94–116, 2017, doi: 10.1016/j.landurbplan.2017.09.019.

[8] D. Stevens, S. Dragicevic, and K. Rothley, “iCity: A GIS-CA modelling tool for urban planning and decision making,” Environ. Model. Softw., vol. 22, no. 6, pp. 761–773, 2007, doi: 10.1016/j.envsoft.2006.02.004.

[9] S. Yassemi, S. Dragićević, and M. Schmidt, “Design and implementation of an integrated GIS-based cellular automata model to characterize forest fire behaviour,” Ecol. Modell., vol. 210, no. 1–2, pp. 71–84, Jan. 2008, doi: 10.1016/j.ecolmodel.2007.07.020.

[10] T. Anderson and S. Dragicevic, “A Geosimulation Approach for Data Scarce Environments: Modeling Dynamics of Forest Insect Infestation across Different Landscapes,” ISPRS Int. J. Geo-Information, vol. 5, no. 2, p. 9, Feb. 2016, doi: 10.3390/ijgi5020009.

[11] D. Gounaridis, I. Chorianopoulos, E. Symeonakis, and S. Koukoulas, “A Random Forest-Cellular Automata modelling approach to explore future land use/cover change in Attica (Greece), under different socio-economic realities and scales,” Sci. Total Environ., vol. 646, pp. 320–335, Jan. 2019, doi: 10.1016/j.scitotenv.2018.07.302.

[12] H. Couclelis, “From cellular automata to urban models: new principles for model development and implementation,” Environ. Plan. B Plan. Des., vol. 24, no. 2, pp. 165–174, 1997, doi: 10.1068/b240165.

[13] R. G. Pontius et al., “Comparing the input, output, and validation maps for several models of land change,” Ann. Reg. Sci., vol. 42, no. 1, pp. 11–37, Mar. 2008, doi: 10.1007/s00168-007-0138-2.

[14] H. Shafizadeh-Moghadam, A. Asghari, A. Tayyebi, and M. Taleai, “Coupling machine learning, tree-based and statistical models with cellular automata to simulate urban growth,” Comput. Environ. Urban Syst., vol. 64, pp. 297–308, Jul. 2017, doi: 10.1016/j.compenvurbsys.2017.04.002.

[15] H. Liu, R. Homma, Q. Liu, and C. Fang, “Multi-scenario prediction of intra-urban land use change using a cellular automata-random forest model,” ISPRS Int. J. Geo-Information, vol. 10, no. 8, Aug. 2021, doi: 10.3390/ijgi10080503.

[16] A. Rienow, A. Mustafa, L. Krelaus, and C. Lindner, “Modeling urban regions: Comparing random forest and support vector machines for cellular automata,” Trans. GIS, vol. 25, no. 3, pp. 1625–1645, 2021, doi: 10.1111/tgis.12756.

[17] L. Wang, J. Yang, S. Wu, L. Hu, Y. Ge, and Z. Du, “Enhancing mineral prospectivity mapping with geospatial artificial intelligence: A geographically neural network-weighted logistic regression approach,” Int. J. Appl. Earth Obs. Geoinf., vol. 128, no. March, p. 103746, 2024, doi: 10.1016/j.jag.2024.103746.

[18] N. Mahdizadeh Gharakhanlou and L. Perez, “Flood susceptible prediction through the use of geospatial variables and machine learning methods,” J. Hydrol., vol. 617, no. PC, p. 129121, 2023, doi: 10.1016/j.jhydrol.2023.129121.

[19] M. Glockmann, Y. Li, T. Lakes, J. P. Kropp, and D. Rybski, “Quantitative evidence for leapfrogging in urban growth,” Environ. Plan. B Urban Anal. City Sci., vol. 49, no. 1, pp. 352–367, 2022, doi: 10.1177/2399808321998713.

[20] X. Liu, L. Ma, X. Li, B. Ai, S. Li, and Z. He, “Simulating urban growth by integrating landscape expansion index (LEI) and cellular automata,” Int. J. Geogr. Inf. Sci., vol. 28, no. 1, pp. 148–163, 2014, doi: 10.1080/13658816.2013.831097.

PhD Candidate, Simon Fraser University

Example of possible workflows for implementing geographic CA models.

Overview of the methodology steps used to implement the application example.

Selected metrics from the Model Evaluation Report files for calibration and validation stages.

The number of cells counted as hits (correctly simulated urban developments), misses (real urban developments not simulated by the model), and false alarms (simulated urban developments to incorrect locations) for calibration and validation stages.

Visual assessment of real versus simulated development forecasts for 2020 in three subareas.