Title: Modeling Biosphere Extremes Using Reservoir Computing

Duration: 01.06.2021 – Present

Research Area: Earth and Environmental Sciences

The PhD project “Modeling Biosphere Extremes using Reservoir Computing” investigates novel methods to predict extreme events in ecosystems. It utilizes reservoir computing, particularly echo state networks (ESNs), a specialized recurrent neural network adept at analyzing time-series data. The research focuses on how ESNs effectively capture and model complex ecological dynamics, highlighting their capability in forecasting extreme environmental conditions. This project “Modeling Biosphere Extremes using Reservoir Computing” combines advanced computing techniques with environmental science, showcasing the potential of ESNs in understanding and predicting ecosystem behavior.

Aims

1. Developing innovative methods for understanding and predicting extreme events in ecosystems using reservoir computing.
2. Show the effectiveness of echo state networks (ESNs), a type of reservoir computing, in analyzing complex time-series data from environmental systems.
3. Providing efficient computational tools to the community to use reservoir computing models
4. Bridging advanced computational techniques with ecological research, enhancing the understanding of ecosystem behavior.

Problem

1. Complexity of Ecological Systems: Addressing the inherent complexity and unpredictability in ecosystem behaviors, especially under extreme conditions.
2. Predictive Challenges: Overcoming limitations of traditional methods in forecasting extreme environmental events accurately.
3. Data Analysis: Effectively analyzing complex time-series ecological data.
4. Utilization of Advanced Computing: Employing reservoir computing, specifically echo state networks, to enhance predictive capabilities in ecological modeling.
5. Bridging Disciplines: Integrating advanced computational techniques with ecological and environmental science.
6. Addressing the Lack of Tools: Developing a generalized and efficient software library, ‘ReservoirComputing.jl’, to address the gap in tools for exploring and implementing reservoir computing models effectively in ecological research.

Practical example

• ReservoirComputing.jl
• rnn-ndvi
• SpectralIndices.jl

Technology

• Julia
• Recurrent Neural Networks
• Reservoir Computing

Publications

• Martinuzzi, F., Mahecha, M. D., Camps-Valls, G., Montero, D., Williams, T., & Mora, K. (2023). Learning Extreme Vegetation Response to Climate Forcing: A Comparison of Recurrent Neural Network Architectures. EGUsphere, 2023, 1-32.
• Martinuzzi, F., Rackauckas, C., Abdelrehim, A., Mahecha, M. D., & Mora, K. (2022). ReservoirComputing. jl: an efficient and modular library for reservoir computing models. The Journal of Machine Learning Research, 23(1), 13093-13100.

Team

Lead

• Francesco Martinuzzi

Team Members

• Karin Mora
• Prof. Dr. Miguel Mahecha

Partner

• University of Valencia