Blue intelligence

An ocean of data

We are living in the era of “Big data”, where the amount of data has increased and will continue rising exponentially in volume, velocity and variety. Within Marine & Coastal Ecosystems (MCEs), a huge amount of information for environmental monitoring and analysis are collected daily by satellite, aerial remote sensing tools, monitoring stations, ships, and buoys to serve marine and coastal-related sectors.

To exploit the potential of big data and overcome limitations in current mainstream frameworks, the research community has started focusing on cutting-edge Machine Learning (ML) approaches, offering a new way of looking at complex environmental systems, while providing useful predictive insight into the functioning of MCEs. ML algorithms have been applied to better understand MCEs, using an array of data types and processing methods to unravel unknown patterns and complex relationships in the data.

ML can be used to analyse the impacts of climate change (CC) on MCEs, with methods including ecological modelling  and multi-risk assessments (MRA) that allows evaluating CC risk across different sectors (e.g., fisheries, coastal and marine development, climate adaptation, and risk insurance) and marine eco-regions. 

With technological advancements, the scientific community has come to the consensus that analysing big data (including environmental, climatic, meteorological, and socio-economic data) with ML has the potential to solve real-world issues, paving the way for improved understanding and more effective management of CC risks and their interactions with the socio-ecological system.

Figure 1: Exploiting heterogeneous ‘big data’ with Machine Learning (ML) within MCEs

Boosting climate risk assessment with ML

Global warming is exacerbating weather, and extreme climatic events and is projected to aggravate risks across multiple sectors. Assessing and managing the multiple risks posed by interacting anthropogenic and natural drivers (including climate change) is one of the major challenges that the research community is currently facing.

This is particularly crucial for MCEs, where little is known about the complex inter-relationships between CC, biodiversity, and ecosystem services (ES) flow, due to limited data availability. 

The complexity of MCEs, and their spatio-temporal dynamics causes major challenges when trying to understand cumulative risks in these systems. Challenges include identifying sites at high risk of cumulative effects (hotspots), and determining the relationships and synergies between multiple pressures that may interact to cause severe impacts.

What are the competitive benefits of ML, and how it may help to better analyse and manage current and future climate change risks on MCEs?

Thanks to the current digitization of EU and international society, ML-based methods can provide an alternative approach to characterizing complex environmental systems and provide reliable quantification of the effect of human activities on MCEs along with the interacting climate-driven and local/global anthropogenic factors affecting MCEs.

ML methods can be broadly applied across different environments, (e.g., river basins, lakes, mainland, coastal areas, and urban areas), and they can be used to carry out diverse tasks while taking into account various stressors, and hazards, exposure, and vulnerability factors (e.g., climatic, economic, social, demographic, cultural, institutional, governance and environmental).

ML for MaCoBioS

Within the MaCoBioS project, we exploit the potential of ML, as well as the huge amount of data that is available for environmental observation and monitoring (e.g. remote sensing data from Copernicus CMEMS and Sentinel missions, USGS Earth Explorer, among others). ML models will be used to assess the response of MCEs to the cumulative impacts caused by CC and human activities, including resulting changes to ES capacity and flow in these systems.

We will use scenario-analysis to assess the response of MCEs under multiple ‘what-if’ multi-risk scenarios, modelled by considering different CC conditions and management strategies. Furthermore, by using worst-case scenarios, ML models can explore the resilience and tipping points of marine coastal socio-ecological systems to multiple risks.

Figure 2: Multi-tiers workflow for ML model development in the MaCoBioS eco-regions and case studies

The outputs of our ML-based applications will comprise a set of GIS-based multi-risk screening scenarios, including eco-region and local scale maps and risk metrics, simplifying understanding and communication of risks induced by changing climate and management conditions in the investigated cases. Data produced will be made available through the MaCoBioS Web-GIS Platform. This information will facilitate the identification of areas and MCEs where management actions and adaptation strategies would be best targeted, and will be used as input data for the Nature Based Solutions suitability mapping as envisaged in WP3.

Figure 3: Expected outcomes from the ML application in the MaCoBioS eco-regions and case studies

An exciting challenge ahead

The availability of spatio-temporal datasets is expected to increase thanks to the rising availability of advanced technologies for real-time environmental data acquisition (e.g., satellites, drones). This process might even be accelerated by ML optimizations in data collection and pre-processing systems. The combination of big -remote sensing- data, ML algorithms able to handle them, together with field survey data feeding validation processes, show a high predictive potential to evaluate and manage short-, medium- and long-term multiple risks due to climate change. In this sense, methods based on artificial neural networks with the use of multiple layers in the network (i.e., deep learning) could be the most promising methods, offering a higher ability to learn from data and understand highly nonlinear behaviours. Deep learning will likely be applied even more frequently to discover intricate climate, environmental and socio-economic structures in large data sets. Under the perspective of a rising abundance of data and ML models’ complexity, researchers will have the possibility (and duty) to enhance the understanding of future climate variability and risks, improving capabilities in climate forecasting, predictions, and projections, with the main aim of providing accurate and sound multi-risk scenarios that will allow for more robust adaptation planning and sustainable management of MCEs.

Share this post

Other News