Sea-surface pCO2 maps for the Bay of Bengal based on advanced machine learning algorithms

Abstract

Lack of sufficient observations has been an impediment for understanding the spatial and temporal variability of sea-surface pCO₂ for the Bay of Bengal (BoB). The limited number of observations into existing machine learning (ML) products from BoB often results in high prediction errors. This study develops climatological sea-surface pCO₂ maps using a significant number of open and coastal ocean observations of pCO₂ and associated variables regulating pCO₂ variability in BoB. We employ four advanced ML algorithms to predict pCO₂. We use the best ML model to produce a high-resolution climatological product (INCOIS-ReML). The comparison of INCOIS-ReML pCO₂ with RAMA buoy-based sea-surface pCO₂ observations indicates INCOIS-ReML's satisfactory performance. Further, the comparison of INCOIS-ReML pCO₂ with existing ML products establishes the superiority of INCOIS-ReML. The high-resolution INCOIS-ReML greatly captures the spatial variability of pCO₂ and associated air-sea CO₂ flux compared to other ML products in the coastal BoB and the northern BoB.