Scientific publication

Artificial intelligence maps giant kelp forests from satellite imagery

Automated kelp canopy detection from satellite data for improved long-term ecological monitoring and conservation efforts.

Credits: NASA - National Aeronautics and Space Administration

The impact of climate change on marine ecosystems is undeniable, and one clear manifestation of this is the shifting distribution and abundance of marine species. Kelp forests, crucial players in marine ecosystems, have seen their geographical boundaries transform on a global scale. To comprehend the underlying dynamics of these ecosystems and to predict their responses to ongoing and future climate changes, it is imperative to synthesize long-term time series data of kelp forests.

Traditional methods of mapping kelp from satellite imagery have proven to be time-consuming and expensive. These approaches demand a substantial amount of human effort for image processing and algorithm optimization. In response to this challenge, we propose an innovative solution - the utilization of mask region-based convolutional neural networks (Mask R-CNN) to automate the assimilation of data from open-source satellite imagery, such as Landsat Thematic Mapper, for the purpose of detecting kelp forest canopy cover.

Our research focused on the giant kelp species Macrocystis pyrifera along the shorelines of southern California and Baja California in the northeastern Pacific. Through extensive cross-validation procedures that tested various model hyper-parameters, including data augmentation, different learning rates, and anchor sizes, we optimized our model. The result was a highly accurate detection system for kelp forests with minimal overprediction (Jaccard’s index: 0.87±0.07; Dice index: 0.93±0.04; overprediction: 0.06). This optimized model enabled us to reconstruct a 32-year time series in Baja California, a region renowned for its kelp variability due to El Niño events.

Our framework, based on Mask R-CNN, now stands as a cost-efficient tool for long-term marine ecological monitoring. This technological advancement promises to facilitate well-informed biodiversity conservation, improved ecosystem management, and more informed decision-making processes. By harnessing the power of automation and artificial intelligence, we are better equipped to track and understand the ever-changing dynamics of kelp forests in the face of climate change.

Main reference

Marquez, L., Fragkopoulou, E., Cavanaugh, K. C., Houskeeper, H. F., & Assis, J. (2022). Artificial intelligence convolutional neural networks map giant kelp forests from satellite imagery. Scientific Reports.

  • Featured publications

biodiversityDS.

Jorge Assis [PhD, Associate Researcher]
Centre of Marine Sciences, University of Algarve [Faro, Portugal]
© 2023 Biodiversity Data Science, All Rights Reserved