Aerosol Optical Depth vs. PM2.5: Adaptation of Hybrid Optimization Algorithms for Temporal Prediction

The study also aims to understand the aerosol optical depth (AOD) relationship with land and ocean by considering their mapping with Particulate Matter (PM2.5) for Manali, with Chennai being a coastal area with abundant aerosols for the period 2016–2020. The AOD data is obtained from NASA’s Visible Infrared Imaging Radiometer Suite (VIIRS) Deep Blue Aerosol satellite using its daily datasets. The study is constructed to find the association between AOD and PM2.5 using machine learning algorithms namely artificial neural network (ANN) and support vector machine (SVM). Furthermore, compare the machine learning techniques with a combination of hybrid algorithms, which is a naturally inspired algorithm; BAT acts together with ANN to get the optimized result. The performance of the models was verified using performance indicator namely RMSE. The prediction efficiency shows that out of two machine learning algorithms and one hybrid algorithm, the hybrid algorithm performed better. Overall, the BAT along with ANN performed better than other models in pattern recognition. The correlation coefficient (R2) of PM2.5 and AOD is 0.58. Statistical analysis shows that the mean and standard deviation of AOD and PM2.5 are 55.53 ± 42.11 and 2.89 ± 1.45, respectively, along with their skewness of 3.477 and 0.289, followed by a kurtosis of AOD and PM2.5 at 328 0.97 and 17.06, respectively. The results showed that 98% and 99% of training and 2% and 1% of ANN tests had better results with RMSE of 21.09 μg/m3, 22.05 μg/m3 during training, and 17.95 μg/m3 and 12.54 μg/m3 provided for testing. For SVM, the normalized poly kernel function was found to be the best out of four functions, with an RMSE of ±21.79 μg/m3 in training and ± 15.7 μg/m3 in testing for 98%,99% of training and 2%,1%of tests. The study concludes that the hybrid algorithm model proves AOD’s ability to make near-term future PM2.5 predictions.