Overview
This case is focused on improving the accuracy of ocean colour forecasting, with a current focus on chlorophyll-a and primary production. These parameters are vital for gaining a deeper understanding of oceanic ecosystems, carbon cycles, and fisheries management. The goal is to generate longer forecasts in a more resource-efficient manner by applying machine learning. This effort contributes to the broader objectives of the European Digital Twin of the Ocean (DTO) to provide more accurate insights into seasonal ocean dynamics.
Challenge
Achieving accurate forecasts of chlorophyll-a and primary production can be significantly challenging and resource-intensive. However, accurate predictions are crucial for a comprehensive understanding of oceanic ecosystems, which play a vital role in global processes like carbon cycling and are essential for supporting our fisheries. The inherent variability of interconnected physical ocean conditions and biological processes makes precise forecasting a complex task.
Solution
The proposed solution involves the strategic application of machine learning algorithms. These resource-efficient algorithms will not only significantly enhance the accuracy of these forecasts, but will do so at a fraction of the computational cost. This is possible because machine learning can effectively leverage the vast amounts of available data from various sources, including remote sensing, and generate connections between these datasets without explicitly needing to be guided. Encouraging preliminary positive results have been obtained through the application of convolutional neural networks to predict seasonal chlorophyll-a from physical ocean forecasts. The plan is to expand these forecasts to encompass other variables of interest, including primary production, nutrients, and various plankton types. Additionally, efforts will be directed towards refining the models to achieve increased regional accuracy, ensuring that the forecasts are relevant and useful at finer spatial scales.
Biological monitoring and sensor resources
So far, the primary data resource has been remote sensing data obtained from Copernicus Marine Services, with GlobColour as the primary dataset at this stage.
Data Sources
The data sources to be utilised will be a combination of ocean colour and physical ocean data:
- Ocean colour remote sensing data from Copernicus Marine Services (GlobColour). This data provides the historical and current observations of chlorophyll-a concentrations that will be used to train and validate the machine learning models.
- Physical ocean data will be sourced from two key datasets:
- GLORYS12, which is an ocean reanalysis product from Mercator Ocean and will be used for training the neural network. This reanalysis provides a comprehensive historical record of physical ocean conditions.
- SEAS5, which consists of multi-member forecasts generated by the ECMWF (European Centre for Medium-Range Weather Forecasts) and will be used to generate the biogeochemical predictions.
- Additional observational datasets will be incorporated as this project progresses.
Analysis Tools
Machine learning algorithms will be used, with different approaches to be tested and selected based on the available data.
Expected outputs
The expected outputs are seasonal forecasts of chlorophyll and primary production. These forecasts will be generated by the developed machine learning algorithms and will be based on the input of physical ocean forecasts. The aim is to provide predictions that extend beyond typical weather forecasts, offering insights into the likely conditions of these key biogeochemical variables in the coming seasons.
Target Stakeholders
The identified target stakeholders for the outputs include:
- Copernicus Marine Services, as a key provider of ocean data and services.
- Local and regional administrations and decision-makers, with Mercator Ocean as an example. These groups can utilise the forecasts for various planning and management purposes.
- It is acknowledged that specific actors and stakeholders will be more clearly defined as the project progresses and initial results are obtained. This is because this DUC was initiated later in the project and is still in a research and development phase.
Digital Twin Features demonstration
It is expected to demonstrate several key features of a Digital Twin:
- It aims to contribute to the ability to sense and anticipate significant events by providing improved monitoring and understanding of ecosystem responses to climate variability. The seasonal forecasts should help predict changes in chlorophyll and primary production that may be linked to broader climate patterns and that can have important downstream effects on ecosystems.
- This capability could directly assist with the development and implementation of management strategies based on forecasted data in various sectors, including fisheries, aquaculture, the prediction of algal blooms, and overall ecosystem surveillance.
- To accelerate research outputs, it aims to enable "What-if" scenarios for different physical ocean states. The research expects that using machine learning models to explore how changes in physical conditions might affect chlorophyll and primary production, researchers could gain valuable insights into ecosystem sensitivities and potential future scenarios.
Status
Ready for implementation.
Leaders:
- Gabriela Martinez Balbontin (MOi).
- Stefano Ciavatta (MOi).