Decision Support - Stage 4

01 Develop a decision support quality improvement and expansion plan 

Completing a Stage 3 integrated assessment represents a significant institutional achievement, but it also generates its most valuable asset: a detailed, experience-based understanding of where the analysis performed well and where it fell short. Before launching data collection, model refinement, or new decision support tools, take time to conduct a structured post-assessment review with your core technical team and key stakeholders. The goal of Step 1 is diagnostic and strategic — to produce a prioritized improvement plan that guides the more action-oriented work of Steps 2 and 3 below.

Document what data inputs proved reliable and well-suited to the selected modeling platform (whether LEAP-IBC, GAINS, or another tool), which sectoral estimates required the most assumptions or proxy data, and where model outputs diverged most from monitoring observations during the Stage 3 validation exercise. Pay particular attention to sectors that appeared as key emission categories in your inventory (see Emissions Inventory guidance), but where activity data were thin or highly uncertain. These gaps define the highest-priority data improvement targets for Stage 4. For example, global level databases for household energy are available via the WHO Global Database of Household Air Pollution Measurements. If available for your jurisdiction, you can also look to academic literature specific to your country (see Permadi et al., 2017 below for an Indonesian example with district and monthly resolution for several sectors, including residential and commercial combustion, biomass open burning, industry and transportation). Li et al., 2024, provides high resolution biomass open burning data but only for one province. For solid waste, official statistics systematically undercount rural areas, so academic studies like Ramadan et al., 2026, are useful because they use machine learning to blend official statistics on waste generation with other surrogates to provide improved estimates of activity and emissions.

Looking Ahead: Data Needs for Episodic and Near-Term Decision Support

Stage 3 Integrated Assessment Modeling (IAM) tools are optimized for long-range scenario analysis, but the episodic decision support tools introduced in Steps 6 and 7 below require a qualitatively different category of data: higher temporal and spatial resolution inputs that can support real-time event analysis. Begin identifying these data needs now, so that collection efforts in Step 2 below can serve both the IAM refinement agenda and the episodic tools agenda simultaneously.

For episodic support, key data needs include: sub-national or provincial-level activity data disaggregated by season; satellite-derived fire hotspot and burned area products (e.g., VIIRS, MODIS); near-real-time air quality monitoring data for model validation; and receptor modeling outputs from source apportionment studies (see Stage 3, Step 6 of the Decision Support guidance) that can attribute episodic concentration spikes to specific source categories. For the near-term policy and cost analysis tools introduced in Steps 8 and 9 below, the critical gap is typically not in emissions data but in local control technology costs and health impact parameters — jurisdiction-specific cost estimates updated to reflect current market conditions, and local baseline disease burden data that will strengthen cost-benefit analyses.

Building a living data improvement framework

The improvement plan developed in this step should not be a one-time exercise. Establish a standing annual data review cycle in which sectoral data custodians report updates to the modeling team. Assign clear data ownership across ministries and document data provenance and uncertainty ranges in a shared repository. A decision support system is only as strong as the data feeding it, and the progression toward more sophisticated tools depends on treating data quality improvement as an ongoing institutional commitment. 

02 Launch and support data collection and revision initiatives

With the priorities identified in Step 1 above in hand, Step 2 moves from diagnosis to action: launching targeted data collection efforts in the sectors where improved data will most directly strengthen decision support. The focus should fall on the key emission categories identified in your inventory (see Emissions Inventory Guidance Stage 1, Step 8 – updated through Stages 2 and 3), because these are precisely the sectors where both IAM refinement and episodic tool calibration will benefit most.

The partnerships needed at Stage 4 go beyond the academic and interagency relationships established in earlier stages. Cultivate formal data-sharing arrangements with national statistical offices, energy regulatory authorities, trade associations, and sector-specific ministries — transportation, agriculture, industry, and environment — through memoranda of understanding or interagency data protocols that define reporting responsibilities, update frequencies, and quality assurance standards. Where subnational governments control key activity data, such as provincial agricultural departments managing crop calendars and burning practices, or municipal authorities overseeing waste collection and disposal, these entities must be brought into the data governance framework as active contributors rather than occasional consultants.

Targeted surveys and bottom-up verification

For sectors where top-down national data have historically driven emission inventory development, Stage 4 is the moment to invest in bottom-up verification of how and where people are using energy through targeted surveys. These efforts will differ from typical inventory development because the goal is now to inform actual policies that the government can adopt, so a better understanding of existing and potential technologies and their operation is needed.  The various technology options or operational details can then be calibrated against the inventories that have been developed (usually) by different staff. 

As an example, household energy surveys in rural and peri-urban communities can reveal fuel stacking behaviors, seasonal switching, and clean technology adoption rates that aggregate statistics consistently obscure. These surveys need not cover entire populations to be analytically useful: statistically designed sampling across representative sub-populations — stratified by income level, geography, urban-rural gradient, or community where cooking and heating practices differ — can generate locally validated emission factors and activity multipliers that substantially reduce uncertainty in residential sector modeling (see Eshetu, 2024, below). Similar survey-based approaches apply to small and medium enterprises in the informal industrial sector. More resources - such as survey examples - can be found in the first step (baseline assessment) of the AQMx Sectoral Guidance on e-Cooking.

In the agriculture sector, ground-truthing of satellite-derived burned area and fire frequency data through field surveys and engagement with farming communities is essential for improving temporal and spatial resolution of burning emission estimates. Farmer interviews and participatory mapping, conducted in collaboration with agricultural extension services, yield insight into the timing, spatial distribution, and drivers of crop residue burning that remote sensing alone cannot provide (see Santiago-De La Rosa et al., 2018, below). Again, additional resources for estimating the extent of CRB are available in the AQMx Sectoral Guidance on Alternatives to Crop Reside Burning (see Step 1, baseline assessment). For the waste sector, periodic waste characterization studies — systematic sampling and compositional analysis at disposal sites and transfer stations — provide the empirical foundation for refining open burning emission estimates. Partnerships with university environmental engineering or public health programs can reduce the cost of these studies while building domestic technical capacity. Additional resources can be found in the AQMx Sectoral Guidance on Open Waste Burning.

Some sectors may be able to rely on nationally consistent and uniformly collected top-down data (e.g. on road transportation sources where rigorous vehicle registration data are typically available); however, this will vary by jurisdiction. Across all sectors, ensure that data collected through these initiatives are documented in formats directly compatible with your IAM platform, with version control and provenance records maintained so that successive model updates can be traced to their underlying data sources. 

03 Refine Integrated Assessment Modeling and scenario models based on updated data

With updated sectoral data in hand from Step 2 above, Step 3 applies those improvements directly to the IAM platform — recalibrating the model, upgrading emission factor tiers where possible, and updating the economic and technology assumptions that underpin scenario projections. This means aligning model assumptions with current and planned regulation and legislation, especially source-based emission standards (see Legal Framework and Policy Design Guidance - Stage 4, Step 3). Focus first on key category sectors that are most impactful for air pollution but treat this modeling cycle also as an opportunity to improve coverage of categories that received only cursory attention during the initial Stage 3 assessment. This may be the time to integrate Tier 2 emission inventory data for sectors where local measurements have become available, or to update spatial surrogates if improved GIS data have been developed (see Emissions Inventory Guidance Stage 3, Steps 3 and 4).

Begin the recalibration by revisiting the baseline year: compare revised activity data and emission factors against observed air quality monitoring data and, where available, updated source apportionment results (see Source Attribution Guidance, Stage 2). Discrepancies that persisted through the Stage 3 validation are the most informative starting points — systematic over- or underestimation in a specific sector often signals a data input problem rather than a structural model error.

For key category sectors, the move from Tier 1 to Tier 2 emission factors is among the highest-leverage improvements available. Tier 2 factors — derived from local stack testing, fuel sampling, or technology-specific measurements — can substantially reduce uncertainty bands for dominant source categories such as industrial combustion, brick kilns, or biomass burning. Where Tier 2 inventory data have been developed through companion emission inventory work, ensure these are formatted and ingested consistently to avoid internal inconsistencies between the standalone inventory and the model’s internal emission accounting.

Beyond emission factors, revisit the economic and technology assumptions embedded in your scenarios. Control technology cost estimates age quickly, and local market conditions — import duties, subsidy regimes, supply chain availability — may have shifted since the original model build. Engage relevant industry associations or ministry procurement records to update these parameters, particularly for the transport and industrial sectors (See e.g. ICCT, 2016 or Table 3.2 in US EPA, 2022 below). Also consider extending coverage to previously residual categories: fugitive dust from unpaved roads, construction activity, and agricultural soils is frequently underrepresented in first-cycle IAM runs despite being a significant contributor to ambient PM₁₀ in many developing-country contexts.

Coordinate the refined IAM outputs closely with your air quality modeling team to ensure that gridded emission outputs are in formats compatible with the chemical transport models used for source attribution and episode analysis (see Source Attribution Guidance Stage 3, Step 7). Also ensure outputs remain aligned with your Legal Framework as you did when designing scenarios. This compatibility is essential groundwork for the episodic decision support tools introduced in Steps 6 and 7 below. 

04 Assess progress and establish new regulatory and policy goals

As with your emission inventory, decision support tools should operate on a cycle of continuous improvement. Step 4 uses source attribution and modeling tools to conduct historical analyses that reveal the impact of already-implemented policies, validate decision support outputs against real-world outcomes, and generate the evidentiary foundation for the next generation of regulatory and policy targets.

Where Stage 3 focused on building the analytical case for policy action, Stage 4 demands a retrospective turn: using the same tools to ask whether adopted policies are delivering the emission reductions and air quality improvements that the models predicted. This back-casting analysis — comparing modeled projections against observed monitoring trends and updated inventory estimates — is among the most powerful validation exercises available to an air quality management program, and it builds institutional credibility for future rounds of scenario analysis.

Begin by assembling a multi-year time series of monitored ambient concentrations at key regulatory sites, aligned with emission trends from successive inventory updates. Where statistically significant declining trends in PM₂.₅, black carbon, or other priority pollutants are detectable, attribute these — as far as the data allow — to specific sectoral policies: vehicle emission standards, cookstove transition programs, waste collection improvements, or crop residue burning restrictions. Alternatively, documenting any increases in PM or tropospheric ozone can point to missing sources or changes in regional or global background sources that need to be accounted for in AQM plans. Updated source apportionment results can disaggregate trend signals by source category, providing stronger evidentiary grounding for attribution claims (see Source Attribution Guidance, Stage 2). Where trends are absent or worsening, this signals implementation gaps, rebound effects, or the emergence of new source categories not captured in earlier assessments.

Re-run your IAM under a “policies implemented” scenario, populated with actual technology adoption rates, compliance levels, and enforcement records from regulatory monitoring. Compare results against original projections to identify where the model performed well and where it diverged from reality. Systematic divergences — for instance, a modeled residential sector improvement that did not materialize because fuel switching rates were lower than assumed — should trigger revision of behavioral and adoption-rate parameters in future scenarios. This calibration against policy outcomes substantially improves the predictive reliability of your tool for the next planning cycle (see Klimont et al., 2018) or illustration of improved emission categories – such as kerosene lamps and gas flaring – relative to prior estimates).

With retrospective analysis complete, translate findings directly into updated regulatory and policy goals. Where early no-regrets measures have delivered significant health, climate, environmental/ecosystemic and visibility co-benefits, document and communicate these outcomes to build momentum for more ambitious targets (see Legal Framework and Policy Design Guidance Stage 3, Steps 3 - 5 and Public Engagement and Communication Guidance Stage 3, Step 9). Where gaps remain — underperforming sectors, persistently elevated exposures, or pollutants not yet addressed by regulation — translate these into concrete priority commitments, with measurable targets, assigned institutional responsibilities, and a defined reassessment schedule. 

05 Staffing check — Ensure your agency is fit for purpose

Steps 1 through 4 consolidate and advance the long-cycle IAM work that justifies medium- to long-term regulatory action. Steps 6 through 10 introduce a fundamentally different operational mode: near-real-time and near-term decision support for managing high pollution episodes. Before proceeding, pause to assess whether your agency has the human capacity needed to implement and sustain these new capabilities without compromising the foundational work already underway. This guidance should be read alongside the AQMx Air Quality Forecasting Guidance and Legal Framework and Policy Design Guidance.

Episodic decision support places qualitatively different demands on staff than integrated assessment modeling. It requires personnel who can interpret monitoring data rapidly, communicate actionable guidance under time pressure, and coordinate across multiple agencies simultaneously. Conduct an honest internal assessment of whether your current staff complement can absorb these responsibilities without creating single points of failure during high-stress episode events.

As a reference point, the South Coast Air Quality Management District (SCAQMD) in California — one of the most technically advanced air quality agencies in the world — maintains approximately 10 to 15 staff covering planning, rule development, and implementation, representing roughly 15 percent of total agency staff. Critically, these positions typically split their time between decision support functions and legislative and regulatory development, meaning the effective full-time equivalent dedicated to decision support at any given moment is a fraction of that headcount. For agencies in earlier stages of institutional development, the implication is clear: episodic decision support cannot be added to existing workloads without either dedicated new hires or a deliberate reallocation of capacity.

Map your current staffing against the functional requirements of Steps 6 through 10. Core competency profiles needed include: atmospheric scientists or air quality modelers capable of running and interpreting chemical transport or dispersion models; data analysts able to integrate monitoring streams, satellite products, and meteorological forecasts; communications and public affairs staff experienced in translating technical findings into timely advisories; and enforcement liaisons who can coordinate with regulatory and compliance teams when episode management triggers operational restrictions. These concepts are more fully explored in the World Bank Report (2024) below under the “Committed Executive” aspect of air quality management governance.  

Where gaps are identified, consider a phased capacity-building approach. Formal partnerships with universities or national research institutes can provide surge analytical capacity during episodes while longer-term staffing plans are developed (see Air Quality Monitoring Guidance, Stage 2). Cross-training existing inventory and modeling staff in near-real-time data interpretation broadens the pool of contributors during episode response. Regional cooperation arrangements, where neighboring jurisdictions share modeling or forecasting capacity, can reduce individual agency burdens for transboundary events. Document your staffing review findings formally and present them to agency leadership alongside the technical roadmap for Steps 6 through 10 below. 

06 Explore multi-day air quality forecasting for public awareness and self-protective action 

While integrated assessment models address emission reduction over years to decades, some municipalities have downscaled air quality forecast models (see examples at the Copernicus website including AirPortal, a downscaled version of their global products) and added local emissions context — including active fire counts from satellite observations, real-time monitoring data from neighboring districts, and local point source inventories — to produce operational multi-day air quality forecasts. These systems, typically providing outlooks two to five days ahead, serve a distinct public health function: giving residents, public health authorities, schools, and event organizers sufficient lead time to take self-protective action before a pollution episode or heat wave arrives.

The primary audience for multi-day air quality forecast output is the general public (especially during heat waves -see Air Quality Forecasting Guidance Stage 5, Step 6) and health-sensitive sub-populations for specific pollution events, so be sure to follow the separate guidance and collaborate with staff developing air quality forecasts communication plans (see Public Engagement and Communications Guidance Stage 4, Step 5). On days when the Air Quality Index (AQI) forecast indicates elevated concentrations or heat warnings, a jurisdiction might issue advisories recommending that the elderly, children, and those with respiratory or cardiovascular conditions avoid outdoor exertion; cancel or relocate outdoor sporting events and school activities; recommend the use of masks outdoors or air purifiers indoors; seek cooling shelters; or issue voluntary emission reduction appeals such as refraining from open burning or residential solid fuel combustion. These measures require lead time and pre-determined authorization — which is exactly what multi-day forecasting provides — but they do not require the near-real-time source attribution detail needed for enforcement deployment, which is the domain of Step 7 below.

India’s AIRWISE framework, developed for Delhi’s National Capital Region by the Indian Institute of Tropical Meteorology, exemplifies what is achievable at the sophisticated end of the spectrum. AIRWISE accurately forecasts very poor air quality episodes up to three days in advance with high accuracy at street-level resolution, assimilating satellite aerosol optical depth retrievals, fire information, surface data from 320 monitoring stations, and high-resolution emissions data. The system provides daily attribution of PM₂.₅ contributions from Delhi itself, surrounding NCR districts, and biomass burning in neighboring states — giving forecasters not only a pollution prediction but an understanding of which sources are driving it. At the simpler end of the spectrum, Spain’s national CAMS collaboration programme integrates Copernicus Atmosphere Monitoring Service forecast products into the operational systems of regional air quality managers, in line with national regulations requiring short-term action plans during high-pollution episodes. This approach — leveraging globally available forecast products calibrated to local monitoring data — is accessible to jurisdictions that lack the resources to operate dedicated regional chemistry transport models.

Whatever the technical approach, the communications architecture matters as much as the science (see Public Engagement and Communications Guidance Stage 3, Steps 3-5). Multi-day forecast information should be disseminated through channels that reach the public effectively: mobile applications, public display boards, SMS alert services, and media partnerships. Decision makers should first establish clear, pre-agreed AQI threshold levels that automatically trigger specific public health advisories, removing ambiguity from the decision of when and what to communicate. Forecast information not linked to clear public guidance has limited protective value. This step lays the public-facing foundation upon which the more operationally intensive response tools covered in Steps 7 and 8 below. 

07 Explore air quality nowcasting and automated early warning systems

Nowcasting occupies a different temporal niche than the multi-day forecasting described in Step 6, and serves a different operational purpose. Nowcasting has been defined as forecasting with local detail, by any method, over a period from the present to approximately six hours ahead. This short window shifts the primary function from public communication to operational response — enabling enforcement deployment, alert escalation, and early activation of regulatory measures before peak concentrations are reached.

The US EPA’s NowCast system is the basis of their AirNow website and illustrates the core concept: it shows air quality for the most current hour using a weighted average of recent monitoring observations, weighting the most recent data more heavily when air quality is changing rapidly — such as during a wildfire, dust event, or rapid industrial emission spike. This approach captures the dynamic character of pollution episodes that multi-day forecasts cannot resolve and provides the near-real-time situational awareness needed by enforcement and emergency response personnel.

Many nowcast systems rely on dense monitoring networks that identify sharp concentration gradients or rapid changes upwind of target jurisdictions. When concentrations at upwind sites begin rising sharply, automated alerts can be issued to operational response teams with a one- to two-hour lead time — sufficient to pre-position enforcement staff, alert district officials, or initiate early stages of a Graded Response Action Plan (see Step 8 below). Low-cost sensor networks, where validated and quality-controlled, can substantially extend the spatial resolution of nowcast inputs beyond what reference monitor networks alone can provide, enabling finer-grained identification of hotspot areas and source directions. Satellite-derived fire radiative power and smoke plume products from VIIRS or MODIS (see FIRMS website) further augment nowcast situational awareness during biomass burning seasons.

The institutional requirements for effective nowcasting are demanding: a staffed monitoring desk during high-risk periods, clear escalation protocols, pre-authorized response actions, and reliable communication links between the monitoring center and enforcement authorities. Agencies should assess these requirements carefully against the staffing review conducted in Step 5 above. A well-designed system with modest technical sophistication and robust institutional protocols will consistently outperform a technically sophisticated system that lacks the human infrastructure to act on its outputs in real time. 

08 Explore Graded Response Action Plans (GRAP)

Consider linking your Decision Support System (DSS) and air quality nowcasting capacity to a structured policy response framework such as a Graded Response Action Plan (GRAP). GRAPs include a range of policy measures of increasing stringency, implemented in a phased and stepwise sequence based on ambient concentration thresholds. By deciding in advance — based on an understanding of typical pollution episodes in your jurisdiction — which measures are needed at each threshold level, agencies can respond rapidly without requiring fresh deliberation during an unfolding crisis.

For instance, India has developed a fully operational GRAP for the Delhi National Capital Region, linked directly to the AIRWISE forecasting and nowcasting platform. The GRAP was designed to allow pollution control authorities to reduce the magnitude of predicted air pollution for different AQI categories by imposing temporary control measures, with activation of different GRAP stages triggered by forecast thresholds so that effective measures can be identified and implemented in advance of peak concentrations. Under the Delhi GRAP, emergency measures are automatically enforced when AQI stages are crossed, triggering a variety of measures with increasing stringency, e.g., odd-even vehicle rationing, bans on construction activity, closure of brick kilns and stone crushers, and intensification of public transport and road cleaning. The pre-constructed matrix of enforcement actions enables rapid response without ad hoc deliberation — a design feature that jurisdictions developing their own GRAPs should replicate.

When designing a GRAP, ground it in your jurisdiction’s own seasonal episode climatology, established through source attribution and forecasting work. Define threshold levels that reflect locally calibrated AQI breakpoints rather than directly importing thresholds from other contexts (see Public Engagement and Communications Stage 3, Step 10). Assign clear institutional responsibility for each stage of activation and ensure enforcement agencies have been engaged in GRAP design and hold pre-authorized authority to act. Test the system through tabletop exercises before the first high-pollution season, and review performance — both forecast accuracy and response timeliness — after each episode. A GRAP that requires fresh ministerial approval to activate during an unfolding episode will consistently be deployed too late to materially reduce peak exposures. 

09 Align GRAP measures with CCAC sectoral strategies and AQMx sectoral guidance

A well-designed GRAP addresses the immediate question of how to reduce concentrations during an unfolding pollution episode. But the specific measures it draws upon — restrictions on open burning, vehicle traffic management, industrial curtailment — are most durable and most enforceable when they are embedded within longer-term sectoral transformation strategies that address the structural drivers of pollution year-round. Step 9 ensures that the episodic response framework built in Step 8 is coherent with, and reinforced by, the sectoral policies and strategies being implemented by national and local authorities. Consider each major GRAP sector in turn.

Agriculture is frequently the most significant contributor to episodic PM₂.₅ during post-harvest seasons. The AQMx Sectoral Guidance on Alternatives to Crop Residue Burning provides practical instructions for implementing proven policies, enforcement mechanisms, and economic incentives for alternative residue management. GRAP burning restrictions become far more enforceable — and more publicly acceptable — when farmers simultaneously have access to viable alternatives. The CCAC Agriculture Hub additionally provides strategies for reducing methane and black carbon from livestock management, paddy rice practices and fertiliser use.

Household energy — including biomass combustion in traditional cookstoves — is a major contributor to winter and dry-season episodes and accounts for nearly 50% of global anthropogenic black carbon emissions. GRAP restrictions on residential solid fuel burning during episodes should be paired with the clean cooking and heating transition pathways described in the AQMx Household Energy Guidance and the CCAC Household Energy Hub. Episodic restrictions imposed without credible alternatives risk disproportionately burdening low-income households and generating the compliance resistance that undermines both the emergency response and the longer-term transition program. For example, even if stove stacking remains a persistent challenge, the practice can only be curtailed during episodes when a cleaner alternative is present in the home.

For road transport, GRAP traffic management measures — odd-even schemes, high-emitter bans, diesel vehicle restrictions in urban cores — should be consistent with the vehicle emission standards and fuel quality improvement trajectories described in the AQMx Sectoral Guidance on Fuel Quality and Vehicle Emission Standards. The Global Strategy to Introduce Low Sulfur Fuels and Cleaner Diesel Vehicles (UNEP/CCAC, 2025) is designed to reduce particulate and black carbon emissions from on-road vehicles by over 90% by 2030, and has supported implementation in more than 60 countries. Aligning GRAP transport measures with this trajectory ensures that episode response is not working at cross-purposes with long-term fleet and fuel transitions.

For waste, GRAP restrictions on open waste burning should connect directly to the structural waste management improvements described in the AQMx Guidance on Open Waste Burning and supported through the CCAC Waste Hub. Episode bans on waste burning are only enforceable at scale when collection services have expanded to the point where burning is no longer the default disposal option.  

Treat Step 9 not as an administrative alignment exercise but as an opportunity to ensure that every GRAP measure has a corresponding long-term structural policy pathway.