Machine learning-based modeling of chl-a concentration in Northern marine regions using oceanic and atmospheric data

Abstract

Chl-a concentration is one of the key characteristics of marine areas related to photosynthesis, along with oxygen levels and water salinity. Most studies focus on estimating chl-a concentration in closed water bodies, rivers, and coastal areas of the tropical and temperate Earth belts and are therefore limited to specific regions and also require direct measurements and chemical analysis to obtain precise information about marine environmental conditions. Remote sensing techniques and spatial modeling aim to offer tools for rapid and global analysis of climate and ecological changes. In this study, we aim to develop a machine learning (ML)-based approach to estimate chlorophyll-a concentration when satellite data are unavailable. To provide physical parameters that may influence the predicted variable (chl-a concentration), we combined satellite observations from MODIS with geophysical Weather Research & Forecasting (WRF) and Nucleus for European Modelling of the Ocean (NEMO) models. Classical ML and deep learning (DL) algorithms were compared and analyzed for their ability to extract key biogeochemical patterns in the Barents Sea. The proposed approach allows us to forecast chl-a concentration for the next 8 days based on spatial features and measurements from preceding days. The best R2 metric achieved was 0.578 using a LightGBM algorithm, confirming the applicability of the developed solution to map the northern marine region even in cases where MODIS observations are unavailable for the preceding period due to insufficient illumination and dense cloud cover.