Urban growth prediction is the linchpin of informed city planning and strategic investment decisions. It holds the key to shaping urban landscapes, optimizing resources, and driving economic prosperity. However, the complexity of urban dynamics often leaves decision-makers grappling with uncertainty about where and when urban development will occur.
In this article, we delve into the fascinating world of urban growth prediction, exploring how technology and data science are reshaping the urban landscape. Join us on a journey through the synergy of ArcGIS and R, two powerful tools that, when combined, unlock the potential to foresee urban growth with unprecedented accuracy.
Project Vision and Scope
Our journey begins with a vision—a vision to harness the power of data to predict urban growth. A dedicated team comprising Jian Lange and Witold Fraczek from Esri, along with the expertise of Professor Carsten Lange from California State Polytechnic University, Pomona, embarked on a remarkable project.
Their mission was clear: Develop a machine learning model to identify locations with a higher probability of urban development. To set the stage, they chose The Research Triangle in North Carolina as their canvas, encompassing prestigious institutions like North Carolina State University, Duke University, and the University of North Carolina at Chapel Hill. This region, spanning 3,744 square miles, represented a microcosm of the United States’ land use patterns.
Bridging Analytical Preferences
Before they could unravel the secrets hidden in the data, the team faced a significant challenge—diverging analytical preferences. Esri’s team favored the ArcGIS Pro, while Professor Lange was an advocate for R, a statistical computing language. These different approaches risked data fragmentation and collaboration hurdles.
The solution? Enter the R-ArcGIS Bridge—an Esri R package that seamlessly integrates ArcGIS Pro and R. This bridge automated analytical processes, ensured data integrity, and married ArcGIS Pro’s spatial capabilities with R’s statistical prowess.
Workflow and Model Insights
The team first prepared and processed the land cover raster datasets within ArcGIS Pro. Subsequently, they transferred the data into R as an R data frame using the R-ArcGIS Bridge. In R, they employed the Random Forest method for modeling and training purposes. Finally, the predictions generated were reintegrated into ArcGIS Pro using the R-ArcGIS Bridge to create visual representations of the final results.
1. Data Preparation
The foundation of any predictive model is data. The team sourced land cover raster data from the National Land Cover Database (NLCD) for 2001 and 2016. ArcGIS Pro transformed these datasets into urban and non-urban categories. To enhance predictive variables, the team created seven new raster layers, incorporating factors like drive time to urban centers, freeway proximity, and environmental considerations.
These spatial datasets were then amalgamated into a single attribute table. The process was automated using ModelBuilder, resulting in a colossal dataset of 8.9 million records.
2. Seamless Data Transfer
The magic of the R-ArcGIS Bridge began with setting it up on a system equipped with R, RStudio, and ArcGIS Pro. The bridge linked RStudio with ArcGIS using arc.check_product(), connecting the feature class to an R variable using arc.open(). This step transformed the feature class attributes into an R data frame—DataAllFromPro.
3. Training and Modeling
Random Forest, a powerful machine learning algorithm, was the team’s weapon of choice for predicting urban development. The dataset was divided into training and test data, with the majority class imbalance addressed using Synthetic Minority Oversampling Technique (SMOTE) from the DMwR-package in R. This strategic balancing yielded an 89% accuracy in predicting urban development and 95% accuracy in predicting no urban development.
4. Returning to ArcGIS
With predictions in hand, it was time to return to ArcGIS Pro for visualization. Using arc.write() in RStudio, the data frame found its way back into the original geodatabase. A joined table and raster layer paved the way for insightful visualizations.
5. Visualizing Success
ArcGIS Pro’s robust cartographic capabilities breathed life into the predictions. The results mirrored expectations, with projected urbanization aligning closely with existing development and road networks. The team even discovered intriguing congruence between predicted urban areas and actual 2020 satellite imagery.
A large-scale view of the predicted probability of urban growth displayed over a satellite imagery basemap. Red and orange denote areas of high suitability and therefore high probability of urban growth, whereas green indicates areas that have a low probability of urban growth.
A large-scale view of the predicted probability of urban growth over the study area, displayed over a basemap of satellite imagery.
Enabling Collaborative Analytics
In retrospect, this project delivered on its promise—predicting urban growth in the Research Triangle area while harmonizing the strengths of ArcGIS and R. The R-ArcGIS Bridge played a pivotal role in this achievement, facilitating seamless data exchange and collaboration across domains. This prototype project exemplifies the potential of multidisciplinary teams, leveraging their strengths and preferred tools to tackle complex spatial data science challenges head-on. In a world driven by data, this synergy is the beacon guiding us toward a smarter, more informed urban future.
