Air quality and health impacts in Nepal’s urban valley: PM2.5 pollution patterns and public health risks

• Monsoon "relief" reduces PM2.5 by 65% but still exceeds WHO guidelines by 3.4 times. • Random Forest achieves 94% accuracy (R2=0.941) via systematic hyperparameter optimization. • All vulnerable populations face year-round health risks (HQ>1), including the monsoon season. • Systematic annual increases (6.7 μg/m3/year) threaten to eliminate seasonal relief by 2030. • Meteorological thresholds identified but insufficient; emission controls urgently needed. The Kathmandu Valley exemplifies how monsoon climate dynamics create deceptive pollution patterns that mask persistent public health threats. Seasonal PM 2.5 variability suggests that the monsoon "relief" pollution, but even the cleanest periods pose severe health risks. We analyzed 1,710 daily PM 2.5 observations spanning 2020-2024 using machine learning and advanced statistical methods, developing predictive models, identifying meteorological thresholds, and assessing health risks across Nepal's five distinct seasons. PM 2.5 measurements were obtained using a BAM-1020 Beta Attenuation Monitor with rigorous quality control procedures ensuring 97.3% data completeness. Extreme seasonal variability was observed, ranging from 51.5 ± 30.5 μg/m 3 during the monsoon to 146.7 ± 27.3 μg/m 3 in winter; however, across all seasons, the values were 3.4-9.8 times higher. Temperature exhibited the strongest PM 2.5 control (r = -0.710, p < 0.001), while Random Forest models achieved superior prediction accuracy (R² = 0.941, RMSE = 12.04 μg/m 3 , MAPE = 10.3%) compared to linear regression (R² = 0.572) following systematic hyperparameter optimization across 18 parameter combinations using 5-fold cross validation (CV R 2 = 0.933 ± 0.017). Critical meteorological thresholds were identified: rainfall exceeding 15.9 mm/day and wind speed exceeding 5.8 m/s reduce pollution, but not sufficiently for safety. Concentrations show a systematic annual increase of 6.7 μg/m 3 (p < 0.001), with all vulnerable populations facing year-round health risks (hazard quotients >1), and people with asthma experiencing the highest risks (HQ = 4.4 in winter). Multi-year patterns demonstrate that monsoon washing effects, while substantial, cannot compensate for extreme baseline pollution levels, revealing the urgent need for emission controls that transcend seasonal pollution management approaches.