Automated detection of coastal upwelling in the Western Indian Ocean: Towards an operational “Upwelling Watch” system

Abstract

Coastal upwelling is an oceanographic process that brings cold, nutrient-rich waters to the ocean surface from depth. These nutrient-rich waters help drive primary productivity which forms the foundation of ecological systems and the fisheries dependent on them. Although coastal upwelling systems of the Western Indian Ocean (WIO) are seasonal (i.e., only present for part of the year) with large variability driving strong fluctuations in fish catch, they sustain food security and livelihoods for millions of people via small-scale (subsistence and artisanal) fisheries. Due to the socio-economic importance of these systems, an "Upwelling Watch" analysis is proposed, for producing updates/alerts on upwelling presence and extremes. We propose a methodology for the detection of coastal upwelling using remotely-sensed daily chlorophyll-a and Sea Surface Temperature (SST) data. An unsupervised machine learning approach, K-means clustering, is used to detect upwelling areas off the Somali coast (WIO), where the Somali upwelling – regarded as the largest in the WIO and the fifth most important upwelling system globally – takes place. This automatic detection approach successfully delineates the upwelling core and surrounds, as well as non-upwelling ocean regions. The technique is shown to be robust with accurate classification of out-of-sample data (i.e., data not used for training the detection model). Once upwelling regions have been identified, the classification of extreme upwelling events was performed using confidence intervals derived from the full remote sensing record. This work has shown promise within the Somali upwelling system with aims to expand it to the rest of the WIO upwellings. This upwelling detection and classification method can aid fisheries management and also provide broader scientific insights into the functioning of these important oceanographic features.