Ephemeral prediction of aerosol optical depth associated with precipitation: A data-driven approach

The study investigates the influence of aerosol optical depth (AOD) in coastal areas where the aerosols have an extensive range owing to wind movements together with precipitation, for a period from June 2016 to Dec 2020. Therefore, the consequences of climate change at the microscale are determined in this research using AOD for the southern coastal area, while the studies are more prevalent only in the Indo-Gangetic Plain. The entire investigation used data from NASA's VIIRS Deep Blue Aerosol satellite. The study examined whether precipitation played a role in AOD prevalence analysis. Four models with variables, namely AOD550, AOD ocean, precipitation, temperature and sea surface temperature, were implemented in this study. Machine learning algorithms were adopted for the prediction analysis, namely artificial neural network (ANN) and support vector machine (SVM) including Polykernel, Radial Basis kernel Function and Pearson VII Universal Kernel (PUK). The results show that precipitation (similar to 0.03) had a negative correlation with AOD550 land but a positive relationship with AOD ocean (0.554). The main conclusions were that ANN does not agree well with the AOD prediction, although R2 and RMSE performed better than average, ANN proved to overperform during the testing phase of all the ratio splits of 60-40%, 70-30%, 80-20%, 90-10%, 98-2% and 99-1%. However, SVM offers more reliability with fewer data variability than ANN thus providing a better result for 98-2% (31 days) and 99-1% (16 days) prediction. The PUK function out-performed the overall SVM function for all 4 models with an RMSE of +/- 0.005 during training and +/- 0.0035 during testing, whose R-2 value was also around +/- 0.90 during training and 1 during testing.