By applying state-of-the-art data science approaches to both Queensland Fire & Emergency Services (QFES) data (internal, public, or otherwise), the project forecasted the likelihood of major hazards (e.g. cyclones, fire) which drive the service demands for QFES. Using these likelihoods, the project generated 1,000 statewide 10-year service demand scenarios. Each service demand scenario was assessed regarding proposed capital and operational investment plans. With this innovation, data rather than tradition now drives investment decisions.
Historically Queensland Fire and Emergency Services (QFES) have used the opinion and experience of their senior executive in consultation with key stakeholders to determine where and when to make capital and operational investments in its evolving emergency response capability. In this digital age there are now decades of available data which can be used to forecast and model the likely future demands upon Queensland's Fire and Emergency Services in the coming years. This valuable input has not yet been deployed to QFES's long term strategic planning.
In response, QFES worked with Queensland's Department of Housing & Public Works to develop a joint emergency service demand forecast based on the objective data. Data was sourced from both department's holdings, Queensland government Open Data, and other 3rd party data sources to inform sophisticated machine learning forecasts of hazard probabilities (e.g. flood, cyclone, fire, road crash, rescue etc) and evolving exposures (e.g. people, assets) over the coming 10 years. This stream, the first of 3, in this project used self-organising maps, Bayesian algorithms and regression techniques to generate the daily probabilities in each location over the coming 10 years.
Based on these probabilities, a data science team (comprising the 2nd of the 3 streams) generated 1000 potential emergency service demand scenarios using stochastic randomisation processes. Each service demand scenario generated 10 years of statewide service demands including incident type, location, severity and duration. Service demands included structure fires, bushfires, swift water rescues, entrapment rescues, road crashes, false alarms, animal rescues, requests for assistance and many other categories of QFES emergency responses. The 1000 10 year scenarios included some scenarios where seemingly everything goes wrong at once and others where there seems little to do in between relatively easy demands on QFES capabilty. Most scenarios are somewhere in between these two extremes.
The third and final stream used Python and in-memory SQLite to (i) generate an artificially intelligent demand response dispatch engine and (ii) simulate the servicing of the 10 years of service demands (including travel time and response escalation). With access to all of the resources available to QFES across the state, combined with the future planned changes to capability (either new, enhanced or decommissioned), the AI dispatches the closest available resource to each of the service demands, tracking which service locations exhaust their resources and when. Across the entire 10 year scenario, a count of local service location and network-wide resource exhaustions is maintained as well as the proportion of times that response time targets were achieved (e.g. x% of responses achieved within 14 minutes). This was repeated for each of the 1000 scenarios to identify which locations are the most frequently unable to service their local service demands. Differing investment plans are also assessed to identify which investments provide the greatest benefit to the community in terms of reduced local resource exhaustion and increased likelihood of adequate response times.
Once all 1000 scenarios are simulated and assessed against each of the candidate investment plans, an assessment is made regarding which of the investment plans is most value for money for QFES and the Queensland community. The project has left QFES with the capability to continue assessing further investment options against the generated scenarios and also the ability to re-generate the scenario with differing assumptions regarding the impact of investment in PPRR (prevention, preparation, response & recovery) and the impacts of changing environmental factors like climate change. A range of geospatial animations have been produced for the more interesting snippets of scenarios (e.g. where network-wide resource exhaustions occurred) and to demonstrate which incident types are varying their duration, frequency and/or severity by season, by location and over the 10 year period.
The results are being used to determine which locations throughout the state are most likely to produce the biggest improvement in emergency response per dollar invested. The timing of these investments is also being assessed to ensure the most beneficial investments are being made at the earliest time feasible.
The model itself will undergo continual improvement. Some of the assumptions of the model were based on scant sets of data, whilst other simplifications were taken to save on computational overhead. As more data and computational resource is made available, the model can evolve to become a closer and closer reflection of reality. Further, post-hoc assessment of how the scenarios reflected subsequent actual events will allow further refinements to the model to further improve its accuracy and utility.
What Makes Your Project Innovative?
Previous efforts to ascertain the best timing and location for new emergency services capability has been based upon the opinion of experienced executive informed by the feedback of key stakeholders. Although well meaning and professionally conducted, the final decisions lack the hard objective evidence to demonstrate that the decisions are indeed the most beneficial options. This can lead to accusations of political interference and boondoggling, even when none has occurred. The data and evidence basis upon which the project develops its forecasts and investment plan assessments eliminates any charge of subjectivity. This project has achieved this in an ostensibly uncertain environment where service demands are driven so strongly by random events like storms, earthquakes and bushfires. QFES believes this approach will be invaluable in many similar environments where random events hold such significant sway over the effectiveness of capital and operational investment decisions.
What is the current status of your innovation?
The innovation has delivered its initial phase of developing the three streams of work to an initial level of inter-operable capability. These initial investment plan assessments have been completed and are for the first time being used to infuse objective data into the strategic investment plans of the department. However the model requires significant enhancement and real-world testing to demonstrate reliable robustness against the variability of reality. Further the ubiquity of the approach is untested. Can this approach be used by Ambulance, Police, or Defence for instance? Can this be used in other countries and jurisdictions? The coming period will demonstrate the reliability and applicability of this innovative approach to emergency agency strategic planning.
Collaborations & Partnerships
The project involved the collaboration of several universities, Queensland's Fire & Emergency Services (QFES), Dept of Housing & Public Works (DHPW), and data science companies: DeepConnect & Data Analytics Consulting. Adelaide University and QFES provided expertise in emergency event simulation & data, QUT provided a PhD intern, DHPW provided geospatial and python coding expertise. DeepConnect and Data Analytics Consulting provided data science expertise, toolsets & project management.
Users, Stakeholders & Beneficiaries
The key users will be Strategy and Planning personnel within the Qld Fire & Emergency Services department. Stakeholders include the personnel of QFES, the central agencies who manage the funding of QFES, and most importantly the people of Queensland who will obtain a better prepared, more capable statewide emergency service capability for the money invested.
Results, Outcomes & Impacts
The project has provided the first data-based assessment of proposed emergency investment proposals ever available. This feedback has enabled the development of data-driven changes to proposed investment plans which will lead to better "bang-for-each-buck" spent on emergency services in the coming few years. It is expected that the ability to balance Prevention, Preparation, Response, Recovery (PPRR) investments against investments in emergency response will also be facilitated. The model will be further improved and tested against real world outcomes further increasing the trust in its results and further increasing its utility for strategic investment planning.
Challenges and Failures
One of the most difficult issues was the paucity of data available. Sometimes this data was strong in some areas but missing in others leading to patchy overall coverage, whilst other datasets were simply insufficiently granular to be of value. In many cases the project was forced to make assumptions and test those estimates with Subject Matter Experts for "reasonableness". Further data improvements will come as legacy systems are being replaced by more modern, comprehensive information systems and as more data becomes available that may have been beyond the budget of the initial project.
Conditions for Success
All the primary stakeholders were open, motivated, appreciative of other's skills and experience, and simply very interested in being part of the project. No hurdle was considered to be a showstopper, simply a barrier needing to be worked around (or under or over). The final client (QFES Futures) was very supportive and included members of her team on the project as well as securing funding for the external members of the team. A clear and repeatedly communicated vision for the project enabled all project team members to know how they fit within the project and helped convey the purpose of the project clearly to all stakeholders.
This approach to assessing investment options in an uncertain environment is applicable to many use cases. Not just Ambulance and Police, but any service that is dependent on uncontrollable random events for its workload. The approach should also be applicable in any jurisdiction around the globe who provide such services. In Australia alone, we estimate there are over 40 agencies that could deploy this approach, collectively with budgets in the A$10Bs of dollars.
A key lesson is simply "Aim high and don't let anything become a roadblock". By compromising on the vision, rather than abandoning it, we were able to achieve far beyond what new stakeholders expected was possible. This momentum has allowed the project to attract further resources, which is bringing yet more progress into the realm of the achievable.
- Identifying or Discovering Problems or Opportunities - learning where and how an innovative response is needed
- Generating Ideas or Designing Solutions - finding and filtering ideas to respond to the problem or opportunity
- Developing Proposals - turning ideas into business cases that can be assessed and acted on
- Implementation - making the innovation happen
8 November 2017