Unsupported Browser

We've detected an older browser version that will not give you the best experience while using The Atlas. Please consider revisitng this site after downloading one of the alternatives below.

Forecasting Ambulance Needs for the City of San Diego

San Diego, CA, USA
City of San Diego

Government Champion

Performance and Analytics Team


Zero upfront cost to local government

Project Status

In Progress/Under Construction since 2021


Project Type

At a Glance

The City of San Diego's Performance and Analytics team used time series data from San Diego’s Emergency Medical Services to predict the number of 911 calls and expected ambulance needs for 2019. Predictions were estimated using Prophet, an additive regression model forecasting tool developed by Facebook research.

Problem Addressed

Ambulance providers in the City of San Diego are subject to specific emergency response times. The City requires providers to meet response time standards 90% of the time. However, as of 2016, five out of the eight medical response zones failed to meet the established compliance rate.

The City is divided in emergency response areas, and each of them has a different volume of calls as well as its own trends. The City is taking advantage of past years’ available data to estimate the optimal number of ambulances for each area and improve compliance rates across the City. The City can achieve this outcome by reducing both over staffing and under staffing in different days and dispatch locations.

The City of San Diego's performance and analytics team used/is using time series data in an additive regression model forecasting tool to address this/these challenge(s).

Solution(s) Used

Data scientist, Zaira Razu-Aznar and data champion, Anne Jensen, both from the City of San Diego developed the use of data to improve the city's EMS service delivery.

They developed a forecasting model that accounts for time and seasonal variations on the volume of calls received by dispatchers, the location of the incidents, and the average time an ambulance spends on task (from the moment it is assigned to respond to an incident to the time it is back to the station). To make a more efficient use of resources and be able to better staff and plan ambulance allocations, we applied the forecasting model to three separate (but closely related) time series from 01-01-2015 to 12-01-2018:

- Daily incidents: total volume of ambulance calls per day

- Response times: from “ambulance assigned to incident” to “ambulance on-scene” per call and per day (totals).

- Time on task: from “ambulance assigned to incident” to “ambulance cleared from hospital” per call and per day (totals)


  1. Increased staffing. For 2019 and beyond, the Fire Department will include the predictions in the RFP process as a factor used to inform providers who want to make a bid to the City.
  2. Compliance with response times will improve. The Fire Department now has more information (call volumes and time on tasks) to hold contractors accountable.
  3. Increased transparency and data driven processes. Predictions results will inform contractors proposals.

Something Unique

A unique feature of this project, besides the predictions themselves, is that we consider not only the volume of calls but also the ambulance time on task. This allows us to account for factors outside of the ambulance response itself, including travel times and hospital wait times.

Who Should Consider?

City Administrations and Fire Departments with available time series data on emergency calls and dispatch times willing to make data-driven decisions in emergency response planning.