Prediction of Arctic kelp forest occurrence using Extreme Gradient Boosting

Abstract

Kelp forests are one of the most productive marine habitats of the world that provide number of valuable ecosystem services for diverse range of species. Understanding the physicochemical factors influencing kelp forest occurrence is vital for comprehending its ecosystems' dynamics. That seems especially important in Arctic environments which are strongly influenced by climate change. Therefore, a high-Arctic fjord (Isfjorden), was selected as a model system to investigate the influential parameters for kelp forest occurrence using a binary classification model - Extreme Gradient Boosting (XGBoost). For this purpose, a set of physicochemical parameters, including water masses flow velocity, electrical conductivity (EC), pH, oxygen, light intensity and temperature, were measured at various depths and locations within kelp forest sites and areas without them. Analyses have shown the possibility of effectively predicting kelp forest occurrence using machine learning based on the measured values of the physicochemical parameters. Additionally, the feature importance analysis of the developed XGBoost model revealed the significance of each parameter in the kelp forest occurrence prediction. The created model demonstrated exceptional predictive performance, accurately distinguishing between kelp forest-associated and kelp forest-barren sites with an AUC (Area Under the Curve) of 0.999. This study serves as a foundation for further research on kelp ecosystems worldwide, emphasizing the significance of employing mathematical modeling approaches to unravel the factors governing kelp forest distribution and growth.