Air pollution is a global environmental challenge that disproportionately affects low- and middle-income nations. While urban air quality degradation is well documented, rural regions often remain largely unmonitored, leading to inequitable health risks and potential environmental injustice. This research presents a real-time assessment, profiling, and predictive modelling of air pollution in a rural resource-constrained region (Khairpur Mirs) of Pakistan, utilizing in-situ measurements from a continuously operating station installed at BBS-UTECH ( \(27.510998^{\circ }\text {N}, 68.741506^{\circ }\text {E}\) ). The research makes three key contributions. First, it provides data-driven profiling of air quality, along with assessments of compliance with WHO Air Quality Guidelines and Pakistan’s National Clean Air Policy (NCAP-23) targets. Second, it develops a multi-parame-ter dust-event detection framework and presents a multi-faceted analysis and characterization of extreme pollution events in the region. Finally, we propose a Storm-Aware Hybrid Model (SAHM), which integrates an ensemble machine learning approach with an Extreme Value Theory (EVT) module in a regime-switching framework to address the dual challenge of forecasting normal air quality dynamics and capturing extreme pollution signatures. Compared with a RF baseline predictive model, SAHM achieved a significant improvement (53.3%) in prediction performance, with an RMSE of 7.19 for hourly AQI forecasts. Alarmingly, over a nearly full annual cycle, the WHO safe guidelines for \(\text {PM}{10}\) were not met 95.4% of the time, while the NCAP \(\text {PM}{10}\) targets were achieved 48.4% of the time. These findings reinforce the urgent need for expanded monitoring and accurate forecasts that can help in targeted interventions for protecting public health in rural regions.
