RainDNA: Autonomous Atmospheric Acidification Analysis

Historically, acid rain has been monitored independently of all other forms of acid precipitation by primary reliance on mathematical numerical models, resulting in delays for accurate forecasts that determine the risk of acid rain deposition and when water remediation needs to occur after the event has already occurred. The purpose of this research is to create the necessary tools to fill these voids so that an Artificial Intelligence (AI) Framework, labelled as RainDNA, will provide greater accuracy in the short-term and long-term forecasting of the Acid Rain Risk Index (ARRI) and develop standardised water purification protocols confirmed through the procurement of multiple weather-related data sources. To accomplish this task, a fused data set was constructed that included cloud imaging from satellites, real-time atmospheric chemistry data and weather API data. A Hybrid Machine Learning (ML) Technology Architecture was subsequently created to address the extensive spatiotemporal forecasting challenges associated with the generation of the ARRI. While the machine learning algorithms trained their respective models using visual cloud feature extraction by MobileNetV2, Principal Component Analysis (PCA) was utilised to compress the extracted visual features to create a fused data set that can then be used with lagged temporal weather data to generate continuous ARRI forecasts and recursive ARRI prediction for the next ten-days. Finally, an offline generative language model (GLM) was developed to produce accurate neutralization protocols based on simulated chemical analyses of generated water quality metrics. Using multi-modal predictive models has proven to be advantageous over traditional baseline methods, which use only past numbers. The RainDNA model discussed here has demonstrated strong results in continuous risk assessment, with a Root Mean Square Error (RMSE) of 0.1435 and a Coefficient of Determination (R²) of 0.5064. The overall framework is designed to be reliable and fault-tolerant, with a geospatial component focused on regions like Kerala, thereby demonstrating promise for other IoT-based and environmental research applications and further exploring the data collected from atmospheric systems.