Air Quality Forecasting - Stage 4

01 Update your plan for Stage 4 modernization

By Stage 4, evolve air quality forecasting into a mature operational service that supports multi-scale, multi-pollutant, and multi-sector applications. Increasingly, use forecasts to support policy implementation, event preparedness planning, and operational decision-making across the jurisdiction. Update your forecasting plan to reflect expanded technical capabilities, larger user bases, and integration with regional and national decision-support and planning systems.  Efforts should include defining governance and coordination structures and encouraging data sharing among government agencies and with neighboring countries to address transboundary pollution (e.g., regional haze/ozone, wildfire smoke, dust transport).  Identify advanced modeling priorities, required high-performance computing and cloud infrastructure, observational data streams, and communication system upgrades.  Define measurable milestones for modernization, covering, for example, forecast reliability, operational uptime, forecast skill, communication performance, and user engagement.  

Evaluate forecast performance routinely using a small set of standard metrics that are meaningful for both technical staff and decision-makers (e.g., see Section 5.6 of the EPA guidelines and Kang et al., 2007). For concentration forecasts, track bias, mean absolute error, root mean square error, and correlation or similar measures of agreement between forecasted and observed concentrations. For high-pollution events and alert thresholds, also track event-detection metrics such as probability of detection, false alarm ratio, critical success index, missed alert rate, and exceedance accuracy (see example from AIRWISE system). Review metrics by pollutant, location, season, forecast lead time, and event type (e.g., wildfire smoke, crop residue burning, desert dust, ozone episodes, stagnation events, or winter inversions). Document results and use them to identify recurring biases, improve forecast methods, and support transparent public reporting. 

02 Secure stable funding and long-term technical capacity

By Stage 4, secure stable institutional funding to retain technical capacity, including a long-term workforce, and plan for the future.  Transition funding from project-based implementation to one supported by government mandates and budgets. Maintain a permanent operational and technical staff and implement succession planning to preserve institutional knowledge and operational capacity (see the capacity-building example from South Asia).  Within your staffing strategies, include hiring and/or training technical specialists in atmospheric modeling, operational meteorology, software engineering, data science, information technology, and communications.  Establish training and certification programs to maintain technical competency and support workforce development (see example training from US EPA).  Infrastructure needs include cloud computing resources, high-performance computing, data storage and backup systems, cybersecurity protections, and software maintenance.  

03 Implement advanced forecast models and methods

Begin implementing advanced forecasting methodologies and operational modeling approaches. Examples ​​​​of such tools include chemical transport models with data assimilation (e.g., GEOS-CF, CAMS; see also Menut et al., 2018), ensemble forecasting systems (e.g., multi-model forecasts), coupled chemistry-meteorology models, and probabilistic forecasting methods (see Chapters 2 and 3 and the WMO Training Materials and Best Practices for Chemical Weather/Air Quality Forecasting for more information on advanced models).  Where data is incorporated into models (e.g., within data assimilation systems), increasingly incorporate observations from the air quality monitoring net, satellites, meteorological agencies, and fire detection systems.  Focus on approaches that improve the characterization of forecast uncertainty (e.g., ensemble forecasts).  Whenever possible, quantify uncertainty using probability metrics to help decision-makers and the public better understand forecast confidence.

Models should increasingly improve the representation of, for example, boundary-layer conditions, atmospheric transport, atmospheric processes that affect forecast performance, and chemistry-meteorology coupling.  Use source apportionment studies, as well as meteorological and air quality monitoring data, including VOC/NOx sensitivity regimes (which help determine whether high-ozone episodes are more strongly driven by changes in NOx emissions, VOC emissions, or long-range transport), aerosol chemistry, and boundary layer evolution, to evaluate and improve forecast modeling.  Where supported by the selected modeling system, develop diagnostic capabilities such as tagged tracers, source-sector or geographic tagging, zero-out or emissions-perturbation simulations, adjoint or sensitivity methods, and ensemble emissions experiments. These methods can help determine whether forecast errors are associated with particular source sectors, precursor emissions, geographic regions, boundary conditions, or atmospheric processes. They should normally be tested retrospectively before being incorporated into routine forecast operations. 

04 Expand monitoring and observational inputs

Increasingly, incorporate more real-time observational inputs (e.g., reference-grade monitoring stations and dense, properly calibrated low-cost sensor networks; satellite observations; indicators of traffic, fire, dust, and industrial activity; meteorological observations; where available, near-real-time aerosol chemical composition, particle size measurements, and source-factor estimates). Low-cost networks, which have been carefully calibrated and quality-controlled appropriately, can substantially improve the spatial coverage of observations.  Satellite data are particularly useful for characterizing transboundary flows (e.g., smoke, dust, and regional-scale pollution) (Holloway et al., 2021 and Vital Strategies, 2023).  An example of the benefits of integrating air-quality information from models and observations is available in Malings et al., 2024. In all cases, document new datasets, including quality assurance and control procedures, and assess their impact on the forecast system and skill before operationally incorporating them. 

05 Forecast transboundary and regional pollution

At stage 4, air quality forecasting covers transboundary air pollution, and supports its assessment and policy response. Many pollution events cross political boundaries and require coordinated approaches to forecasting and response (see examples of US-Canadian and Chinese transboundary pollution, US EPA/Environment Canada, 2015 and Mai et al., 2023).  Working with relevant policy-makers, establish coordination mechanisms (e.g., Memoranda of Understanding or MOUs) with neighboring and regional forecast centers and meteorological organizations. These mechanisms should include data sharing, model intercomparison exercises, operational coordination, and communication alignment, all of which support regional-scale preparedness planning, emergency response coordination, and public communication. The forecasting system should provide information on the likelihood and timing of transboundary pollution transport. Quantitative assessment of the contribution of pollution originating outside the jurisdiction should use an appropriate attribution method, such as regional source tagging, geographic emissions perturbation experiments, boundary-condition sensitivity analysis, trajectory analysis combined with observations, or coordinated multi-model studies.   

06 Expand decision support services 

At Stage 4, air quality forecasting should support a broader range of operational, policy, and decision-support applications.  For example, forecasts can inform and support episodic emission controls, emergency response measures, school advisories, transportation planning, and health-sector preparedness.  Tailor data products to the needs of specific sectors and decision-makers (e.g., transport agencies, health ministries, emergency managers, aviation authorities, and industrial regulators).  Forecasting and data products should support planned policy actions and longer-term planning activities (e.g., by using atmospheric models with policy-scenario emissions) (see Decision Support Guidance Stage 4). Explore examples of existing decision support systems, such as the Air Quality Warning and Integrated Decision Support System for Emissions (AIRWISE) and the Decision Support System for Delhi, India.  ​  

07 Strengthen public-facing services 

Mature air quality forecasting systems provide targeted and accessible data products for the public.  For example, geographically specific alerts and messaging tailored to different user groups and vulnerable populations are common.  Aim to improve your communication with the public, both to enhance usability and to increase understanding, including dashboards, maps, mobile applications, and other impactful visualizations. Implement inclusive measures, including multilingual accessibility and products appropriate for varying levels of literacy, disability, and internet access (see Pennington et al., 2019, for an example from the US). Working with the Public Engagement & Communication team, conduct user testing to ensure that forecast products are both useful and understandable.

At this stage, you can also begin to adapt or develop a locally appropriate Air Quality Index (see Public Engagement & Communications Guidance, Stage 3, Step 10; and Eder et al., 2010). The AQI should translate pollutant concentrations into clear categories, health messages, and recommended actions that can be used consistently in forecasts, advisories, and public alerts. Work with the air quality monitoring team to support the technical calculation of the AQI from validated monitoring data. Communications staff should help ensure that AQI categories, colors, messages, and recommended actions are understandable to the public and aligned with local health guidance. The AQI can then be used as a common framework for issuing early warnings, communicating forecast severity, and triggering preparedness actions during high-pollution episodes.  

08 Institutionalize verification and reporting

Institutionalize formal verification and reporting by benchmarking your operational performance against external peers and participating in independent technical reviews (see Section 5.6 of the EPA guidelines). Such reviews help to identify weaknesses in methods and models, communication systems, and operational procedures. Make sure you archive all forecasts, observations, model setups, evaluation results, technical documentation, and operational procedures.  Incorporate the results of forecast verification into organizational planning, improvement, and modernization activities. For transparency, routinely publish your forecast performance metrics, summaries of your operational system, and technical evaluation results (e.g., see Kukkonen et al., 2012). 

09 Maintain a formal 1-2-year modernization cycle

Implement formal, frequent modernization cycles in Stage 4 to ensure that technical methods and models, operational procedures, and priorities are current and useful.  Make sure you routinely engage stakeholders to reassess priorities and identify emerging user needs.  Periodically update operating procedures, technical standards, and governance to reflect evolving science, technology, and operational requirements.  To prepare for Stage 5, begin developing capacity for future advances, including advanced analytics, in-house research and development, and artificial intelligence/machine learning (AI/ML) applications (WEF 2026 and Shahbazi et al., 2024).