Abstract:
Emergency response planning is a vital aspect of urban resilience, requiring precise decisions on resource allocation and infrastructure placement. In our research, we explored how data-driven tools can optimize emergency response times using geospatial analysis and computational design methods. Focusing on Kadikoy district in Istanbul, we simulated ambulance routes from stations to emergencies and onwards to emergency rooms, using distance as a proxy for response time.
Our study included three scenarios: analyzing the current infrastructure, identifying the optimal location for a new ambulance station using generative optimization algorithms, and determining the best site for consolidating two existing stations into one. Using Grasshopper with tools like WallaceiX, Urbano, and Shortest Walk, alongside Python for data visualization, we generated color-coded maps, route visualizations, and statistical analyses to pinpoint areas of need. Results revealed that the eastern part of Kadikoy is underserved and would benefit most from additional infrastructure.
This work demonstrates the power of AI and computational tools in urban planning that we have learned during this Data-Driven Design course, used in our case to generate insights to improve emergency preparedness. Future efforts will focus on incorporating traffic data and scaling simulations to larger urban areas for real-world applications.
Background
Emergency response planning is an important part of urban planning. Planners must consider and decide how many resources to allocate to emergency preparedness, how many personnel to employ, and where to place emergency response infrastructure to ensure effective and rapid response times.
This planning requires input from many stakeholders and is typically done at the municipal level. Our research seeks to investigate how data driven planning tools can help stakeholders make more informed decisions about emergency response planning and preparedness.
In our investigation, we used geospatial data to analyse emergency response times in the Kadikoy district and surrounding neighborhoods in Istanbul, Turkiye. Specifically, we wanted to measure how far ambulances have to travel from their ambulance station, to a digitally simulated emergency, and then onwards to an emergency room, measured in meters. The total distance the ambulance must travel is then used as a proxy for emergency response time; thus, the further the ambulance must travel, the slower the response time, and vice versa.
To begin, we conducted an analysis of the base scenario which reflects the current situation. Then we performed an analysis of a hypothetical scenario wherein the city of Istanbul constructs a new ambulance station in the Kadıköy district, and we apply a multi-objective generative optimization algorithm to determine the best location for the new ambulance station. Finally, we performed a third analysis of a second hypothetical scenario wherein the city of Istanbul decides to consolidate the two existing ambulance stations into one station and use a similar algorithm to determine the optimal location for this consolidated station.
Data Sources:
- Street network: OSMNX Python library.
- Emergency room locations: Internet search.
- Ambulance station locations: Municipal records.
Tools:
- Grasshopper with Urbano, Shortest Walk, and Wallacei plugins.
- Python (Matplotlib) for data visualization.
At this point, it is important to note that incorporating traffic into our simulations was outside the scope of our research, and therefore we simplified our investigation to exclude traffic data.
Base Scenario
To begin, we conducted an analysis of the current base scenario. That is, we analyzed the response times using the current configuration of fire stations and emergency rooms in the Kadıköy district.
Methodology
To conduct our analysis, we plotted a grid of points in Kadıköy. The grid points are 500mx500m apart and are used to digitally represent the location of an emergency. Then for each ‘emergency’, we plotted the route from both ambulance stations to each emergency, and the route from each emergency to both emergency rooms, and then selected the shortest route to and from both for each emergency. This created the shortest path for each of the 105 digitally simulated emergencies.
Data Visualization
We color coded the grid of points to represent the total distance an ambulance must travel from the ambulance station, to the emergency, and then onwards to an emergency room at a hospital. Hospitals are represented by purple pins, while ambulance stations are represented by pink pins. The locations with the shortest distance travelled are colored green, while the locations with the longest distances to travel are red.
We also plotted the routes the ambulances must travel to reach their destinations. Where multiple routes overlap, we created thicker lines that are proportional to the number of simulated routes that overlap. For instance, if Street A has only one ambulance route on it, and Street B has 4 ambulance routes on it, then Street B will be shown as four times thicker than Street A.
Results and Analysis:
Based on the visualizations, we can come to several conclusions. First of all, the Eastern portion of the city is relatively underserved to first responders and would likely benefit from an additional ambulance or fire station located in the east end.
Second of all, we can see which routes are more frequently used by emergency responders. This data can be useful to city planners and administrators as it indicates which streets are more critical for first responders and more necessary to ensure that they are safe and clear to navigate. That is, these critical routes require a greater amount of maintenance to ensure that they are navigable, and cities should avoid disturbances on these streets, such as demonstrations, festivals, or parties.
We extracted the list of emergency response distances from grasshopper and produced a boxplot and histogram that shows the distribution of response distances from this base scenario. These charts will be more relevant in the next sections, but they still provide an overview of the statistical distribution.
Data Driven Experiments
To take our research further, we wanted to see how we could use data driven methodology and computational tools to inform stakeholders about urban planning decisions. To that end, we have imagined two hypothetical scenarios and constructed computational models to perform experiments that attempt to determine the optimal solution to each problem.
Scenario 1: Where to place an additional ambulance station?
Our first scenario imagines that Istanbul allocates funds to construct a third ambulance station in the Kadikoy district, and planners want to ensure that the additional ambulance station optimally reduces emergency response times in that district. Our task is to determine where to place the additional ambulance station to optimally reduce ambulance response distances.
Methodology
To determine the optimal location for the third ambulance station, we chose to use an evolutionary multi-objective generative optimization algorithm. Specifically, we used the WallaceiX plugin for Grasshopper.
Wallacei generates and evaluates multiple simulations based on input parameters to identify optimal solutions measured by fitness values. For this case study, the algorithm tests various ambulance station locations, assesses their response times, and iteratively refines the solutions through successive simulations until a defined endpoint is reached.
To this end, we constructed a variable position point in Grasshopper that represents the location of the new ambulance station, and appended this point to the list of ambulance stations. The latitude and longitude coordinates were used as the input parameters. We created scripts in Grasshopper to find the mean, median, maximum, and standard deviation of response times, and set these results as the fitness values.
We ran 225 simulations of possible locations for each simulation we calculated the mean, median, maximum, and standard deviation of response times. Based on the results, the algorithm produced 34 pareto optimal sites for the new ambulance station.
Data Visualization
The 34 optimal sites are shown in this map in green. As we expected based on a priori examination of the base scenario map, Kadikoy would be best served by an additional ambulance station in the eastern part of the district.
We plotted one of the sites identified as optimal as a new ambulance station. After, we ran a similar analysis to the base scenario to determine regions of the city with shorter and longer response distances and to identify critical routes. The visualization is shown below.
Next, we compared the response distances between the base scenario and scenario 1 using boxplots and histograms.
Results and Analysis:
While the top quantiles do not change much, the boxplots and histograms show that adding an ambulance station in the Eastern part of the district reduces travel distances in the bottom quantiles, as this previously underserved area now benefits from improved coverage.
Scenario 2: What if there were only one ambulance station?
This scenario imagines that the city of Istanbul decides to consolidate the two ambulance stations in Kadikoy into one location and planners want to ensure that the consolidated ambulance station increases emergency response times by as little as possible. Our task is to determine where to place the consolidated ambulance station to minimize the increase in ambulance response distances.
Methodology
For that we used much of the same methodology as the first scenario. We constructed a variable point to represent the consolidated ambulance station and input the latitude and longitude coordinates as the genes into the MOGOA, and used the mean, median, maximum, and standard deviation of response distances as fitness values.
Data Visualizations
The evolutionary algorithm identified 20 pareto optimal sites, visualized in red. The hospitals again are represented by the purple pins. We observed that the 20 optimal locations could be categorized into 3 clusters and 2 outlier points, depicted below.
Cluster 1 in red
Cluster 2 in orange
Cluster 3 in yellow
Outlier 1 in light blue
Outlier 2 in dark blue
For the three clusters, we created a cluster centroid, using the average of the X-coordinates and Y-coordinates. We kept the two outlier points as they were. These new points are shown on the map below.
We then performed the distance travelled analysis described in the base case scenario above for each centroid and outlier, which we visualized in the boxplots and histograms below.
Results and Analysis:
The charts show that the consolidated station will have longer distances to travel than when there are the two ambulance stations in the base case scenario, which is to be expected since the Kadikoy district is only being served by one ambulance station.
The charts indicate that the 5 locations that were identified (the 3 centroids and 2 outliers) are shown to have very similar response distance statistics. There are tradeoffs between the points, however. For instance, cluster 1 and outlier 1 have a lower median value but larger bottom quantile response times, while cluster 3 has a higher median value but lower bottom quantile response times. The ultimate decision would lie with stakeholders to determine what tradeoffs they want to make to best serve their city.
Further Research Directions & Takeaways
We identified two promising directions to expand our research further:
The first path is to improve the efficiency of the computational models. The shortest walk calculation is heavy, and running a large number of simulations for the optimization algorithm is computationally taxing. This limits the number of simulations and complexity of simulations that can be performed on a standard computer. Further work is necessary to see if there are any effective strategies to reduce computational load to scale the methodology beyond one district of Istanbul without losing the high quality of data produced.
The second direction is to incorporate traffic data into the analysis. The incorporation of traffic information was outside the scope of this research because the dynamic and variable nature of vehicular traffic introduced a level of complexity that your authors were unable to handle effectively. However, traffic would be a critical factor in real world response times and therefore would necessarily be incorporated into a more robust real world application.
Key Takeaways
- Data-driven tools can significantly aid emergency response planning.
- Optimal resource allocation improves efficiency and reduces response times.
- Visualization helps stakeholders identify underserved areas and critical routes.
- While data can be used to better inform stakeholders, AI ultimately cannot replace human decision makers to assess tradeoffs.
- Future work can expand scope and improve model accuracy.
