FROM ATMOSPHERIC CONDITIONS TO HEALTH ALERTS — AN END-TO-END AQI AND WEATHER DATA PIPELINE FOR ENVIRONMENTAL EXPOSURE AND RISK MONITORING - (AIRWARE)

Abstract

Air pollution is one of the leading environmental risks to public health, and its effects depend not only on pollutant levels but also on weather conditions such as temperature, humidity, wind speed, and atmospheric stability. In Indian cities, fine particulate matter often rises to unhealthy levels during winter inversions, crop residue burning, and festival periods, while summer heat increases ozone formation. Although monitoring stations publish readings every hour, the data is scattered across portals, has frequent gaps, and is rarely translated into personal health advice. This paper presents AirWare, an end-to-end data pipeline that collects air quality and weather data, computes exposure indicators, forecasts the Air Quality Index, and turns the results into health alerts for different groups of people. The pipeline ingests hourly pollutant concentrations for PM2.5, PM10, NO2, SO2, CO, and O3 from continuous ambient air quality monitoring stations, together with meteorological variables from weather services and, where available, readings from low-cost sensors. Each source has its own format, time zone handling, and units, so the ingestion layer standardises timestamps and units, tags every reading with its station and location, and records its origin. Cleaning rules remove negative and physically impossible values, flag sensor drift, and fill short gaps using interpolation and neighbouring stations, while longer gaps are left as missing and reported. Cleaned data is stored in a time-series warehouse and used to compute sub-indices and the overall AQI according to the National Air Quality Index method. Spatial interpolation produces ward-level estimates between stations. A forecasting model based on gradient boosting, with an LSTM network for comparison, predicts AQI for the next 24 to 72 hours using lagged pollutant values, weather forecasts, wind direction, and calendar effects. An exposure module combines predicted AQI with user profiles, such as age group, respiratory or cardiac conditions, and outdoor activity, to produce a personal risk level.