Air Quality Forecasting - Stage 3

01 Update your plan for Stage 3

In Stage 3, evolve from relying on externally available tools to focusing primarily on public information functions. Move toward an operational organization supported by your own tools for air quality management, pollution-episode response, and government decision-making.  For this first step, update your plan to reflect these objectives and define the operational services your forecasts should support. For example, national circumstances may prioritize forecasting Identify the technical infrastructure, software systems, staffing, and computational resources (see this example assessment from NOAA), as well as the standard operational procedures needed to support routine forecasting operations, preparedness, communication, and coordination of responses. Include clear milestones and operational performance targets in your plan. 

02 Secure funding

Air Quality Forecasting requires sustained financial support to remain reliable and effective (Clean Air Fund, 2025). In this stage, transition forecasting activities from the pilot phase towards stable funding through regular government budgets or sustainable institutional agreements. Funding must account for staffing, operational software and computer hardware, internet, cloud, communications infrastructure, and training resources.  Replacement cycles and operational redundancy must be accounted for in budgets. Of particular importance are staff retention and archiving institutional knowledge; maintaining this knowledge enables your team to ensure reliable service delivery and maintain public trust. 

03 Strengthen operational forecasting infrastructure

Build a robust operational infrastructure to support forecasting, including data management and communication (see WMO’s Training Materials and Best Practices for Chemical Weather/Air Quality Forecasting, Chapter 8).  Work on strengthening computing systems, data storage, data transfer capabilities, workflow automation, and overall system reliability. Automatically ingest observational data required for forecasts, including meteorological data, external data (e.g., satellite data), and air quality monitoring data at forecast production speed to reduce workloads and improve forecast consistency.  Coordinate closely with the air quality monitoring team to improve access to near-real-time air quality observations for forecast verification and event detection.  Importantly, implement backup systems and continuity procedures to reduce the risk of downtime (e.g., due to equipment failures or internet outages).   

04 Transition to locally run forecasting tools 

With secure funding and infrastructure in place, begin transitioning to locally run forecasting tools while simultaneously (for now) continuing to use externally available tools. Select forecast tools and models at the local or national level appropriate for your technical capabilities and computing capacity (see WMO’s Training Materials and Best Practices for Chemical Weather/Air Quality Forecasting, Chapter 4). Forecast system approaches can include a wide spectrum of tools, ranging from s​​tatistical models (i.e. empirical relationships between pollutant concentrations and predictor variables like temperature, winds, etc.) and simplified dispersion models (i.e. models which combine meteorology, emissions, and simplistic chemical processes to derive pollutant concentrations deterministically) to highly complex chemical transport and coupled meteorology-chemistry models (see Chapter 4 of the US EPA Guidelines for Developing an Air Quality Forecasting Program). Using simpler models, such as dispersion models like AERMOD, may be more feasible where staff, emissions inputs, monitoring data, or computing resources are limited. In contrast, more advanced models, such as CMAQ, WRF-Chem, and GEOS-Chem, can represent multiple pollutants, atmospheric chemistry, emissions, meteorology, and regional transport but require greater expertise and operational support. Begin with tools that are realistic for your current circumstances and identify what additional data, staff training, and infrastructure would be needed to move toward more advanced modeling in later stages.

Distinguish between an operational forecast configuration and diagnostic modeling configurations. The operational configuration should be optimized for timely, reliable prediction of expected concentrations. Separate retrospective or experimental configurations may be used for source attribution, emissions sensitivity testing, model evaluation, or policy analysis. These configurations may use the same underlying CTM, meteorological model, and emissions-processing system, but they have different objectives, computational requirements, and quality-assurance procedures.

No matter the forecasting approach taken, design the forecast domains, pollutants considered, and forecast products (outputs) to meet local and end-user needs.  As much as possible, forecast systems should increasingly prepare to incorporate locally developed products (e.g., from air quality monitoring, source attribution, and emissions inventory), either formally as part of model setup or for forecast verification and improvement.  Importantly, maintain well-documented workflows (e.g., reproducible model configurations including assumptions, input datasets, model settings, and update procedures). Well-documented workflows support operational consistency, troubleshooting, and future upgrades to the air quality forecasting system. 

05 Integrate emissions inventory and source attribution information

In this step, focus on formally incorporating local data (air quality monitoring data, local emissions inventories and source apportionment studies) into forecast products to improve forecast realism and interpretation.  Locally developed gridded emissions inventories​​​​, especially those that are more spatially and temporally refined, can improve the representation of seasonal and episodic sources (e.g., biomass burning, residential heating, agricultural activity, industrial operations, and traffic patterns) in forecast models. Locally developed inventories are typically more representative than global inventories such as EDGAR and CEDS. ​​Source attribution data can improve the interpretation of forecasted conditions and help identify likely contributors to individual elevated pollution events. As new data is incorporated into forecast systems, routinely update all assumptions and operational procedures. 

06 Establish formal operational procedures

At this stage, formalize Standard Operating Procedures (SOPs) to ensure consistent and reliable forecasts. SOPs should: define schedules, approval procedures, quality assurance checks, communication protocols, operational responsibilities, and procedures for escalating issues. See example SOPs from India,  Texas, and the scientific literature (Duncan et al., 2021). Work with Decision Support and Public Engagement & Communication colleagues to establish formal procedures for communicating pollution forecasts, especially for high-pollution events, to health agencies, emergency responders, transportation agencies, and other decision-makers.  As always, maintain documentation to archive forecast decisions, technical issues, and deviations from standard operations.  Institute routine quality-check procedures to identify data issues, communication failures, or problematic forecast results.      ​  

07 Support episode management and early warning

At this stage, focus on supporting the operational management of major pollution events, such as stewarding early-warning activities (see WMO’s Sand and Dust Storms Compendium, Chapter 10). Forecast products should strive to identify the likelihood, timing, and severity of events. In coordination with the Decision Support Guidance (Stage 4, Step 6), estab​​lish operational coordination with relevant agencies or ministries (e.g., health and transport authorities, emergency management agencies).  Using the Public Engagement and Communications guidance (Stage 4, Step 3), develop messaging for targeted populations during severe pollution events (e.g., wildfire smoke, dust storms, crop residue burning). This messaging should clearly explain the forecast, associated uncertainty, and recommended exposure reduction actions for the event (see WHO’s Personal Interventions and Risk Communication on Air Pollution, 2019). 

08 Implement routine verification and skill assessment

While forecast verification activities occurred in Stage 2, formalize them in Stage 3 to ensure that forecast skill is routinely tracked by pollutant, location, season, and forecast lead time (e.g., see WMO forecast verification training and Section 5.6 of the EPA guidelines).  Multiple operational metrics can be considered: forecast bias, forecast error, event detection performance, false alarm rates, and exceedance prediction performance.  Regularly document verification and evaluation results and ensure they are reviewed by staff and management.  Where sufficient data are available, supplement concentration-based verification with a diagnostic evaluation of the processes and sources that affect forecast performance. For example, assess whether the model correctly represents the occurrence of biomass-burning influence, dust transport, traffic-related peaks, secondary aerosol formation, or regional inflow identified by source apportionment studies. Agreement in total concentration can sometimes result from compensating errors among source sectors or atmospheric processes. Use the results from forecast performance assessments to prioritize future system upgrades, identify operational weaknesses, improve communication, and, overall, help modernize the forecasting system. 

09 Establish a cycle of continuous improvement

In this final step of Stage 3, focus on adopting an approach that incorporates continuous improvement across operational performance, user needs, technical methodologies, and communication. To maintain operational stability and transparency, move forward by incrementally updating the forecast system (methods, models, operational procedures, input datasets, verification procedures, and communication strategies), and always document any updates thoroughly. At regular intervals and whenever feasible, seek out stakeholder feedback and incorporate it into operations (see this example from weather forecasting). Begin preparing, through internal team capacity building and infrastructure/compute improvements, for Stage 4 modernization, which will include the use of more advanced modeling systems, expanded observational inputs, regional coordination, and links to advanced decision-support applications (see Campbell at al., 2022).