DeepRes-UpwellNet: A Robust Framework for Coastal Upwelling Monitoring Along the Moroccan Atlantic Coast

Upwelling, a pivotal oceanic process, serves as a crucial mechanism for transporting nutrients from the ocean depths to the surface, playing a fundamental role in enhancing primary productivity and bearing immense significance for coastal ecosystems. Within the scope of this study, we introduce a novel deep learning tool tailored to the monitoring of upwelling in the Moroccan coastal waters, utilizing satellite images encompassing both biological and physical attributes. Recent years have witnessed substantial advancements in the field of deep learning, particularly in the domain of segmentation techniques. These methods have proven their efficacy across a spectrum of ocean remote sensing applications, encompassing image segmentation, classification, and detection. Notably, the U-Net architecture has gained widespread recognition for its effectiveness in these tasks. In the context of our investigation, we present DeepRes-UpwellNet, a fully convolutional deep neural network architecture explicitly engineered to automate the detection and precise localization of upwelling regions in sea surface temperature (SST) and chlorophyll-a (chl-a) images. Our primary objective revolves around assessing the performance of deep learning in the accurate identification of upwelling regions. To achieve this, DeepRes-UpwellNet is meticulously trained and optimized using a database of satellite-derived SST and chlorophyll-a data sourced from the Moderate Resolution Imaging Spectroradiometer (MODIS). The empirical results, as obtained from experiments conducted along the Atlantic coast of Morocco, clearly underscore the superior performance of our proposed model in comparison to conventional segmentation methods. By harnessing deep learning-based upwelling detection systems, we usher in a cost-effective, precise, and practical approach to objectively analyze this pivotal oceanic phenomenon.