| DOI | Resolve DOI: https://doi.org/10.1109/IJCNN60899.2024.10651143 |
|---|
| Author | Search for: Durand, Guillaume1; Search for: Valdés, Julio J.1ORCID identifier: https://orcid.org/0000-0003-2930-0325; Search for: Guyondet, Thomas; Search for: Rice, Olivia Gerry; Search for: Coffin, Michael |
|---|
| Affiliation | - National Research Council of Canada. Digital Technologies
|
|---|
| Funder | Search for: Environment and Climate Change Canada; Search for: National Research Council Canada; Search for: Canadian Food Inspection Agency; Search for: Canadian Space Agency |
|---|
| Format | Text, Article |
|---|
| Conference | 2024 International Joint Conference on Neural Networks (IJCNN), June 30 - July 5, 2024, Yokohama, Japan |
|---|
| Subject | estuary ecosystems; dissolved oxygen time series; non-linear dynamic systems; chaos theory; modeling; machine learning; chaos; analytical models; fluctuations; computational modeling; biological system modeling; time series analysis; predictive models |
|---|
| Abstract | This paper introduces a novel approach for modeling dissolved oxygen (DO) time series data collected in Atlantic Canada. The primary objective is to propose a solution that can predict significant fluctuations in dissolved oxygen levels in estuaries, thereby improving our ability to manage and protect these dynamic ecosystems. Dissolved oxygen serves as a crucial indicator of water quality and is influenced by various factors such as temperature, nutrient concentrations, and biological activities, resulting in a nonlinear dynamic system. By analyzing chaos-related metrics, we demonstrate that the inherent unpredictability and sensitivity of DO time series can be effectively captured. We show that the modeling of the magnitude and duration of DO fluctuations can be anticipated by studying features like bifurcation patterns. Our research effectively characterizes dissolved oxygen series as nonlinear dynamic systems, revealing the internal structure of these systems and enabling differentiation of sites based on their hypoxic behavior. We develop models using both classical and deep learning-based early warning/tipping point indicators, which successfully predict the occurrence of hypoxic events even in degraded sites. The presented results highlight the advantages of our proposed method in anticipating tipping points and predicting the duration of anoxic episodes, particularly in highly degraded sites. |
|---|
| Publication date | 2024-06-30 |
|---|
| Publisher | IEEE |
|---|
| In | |
|---|
| Language | English |
|---|
| Peer reviewed | Yes |
|---|
| Export citation | Export as RIS |
|---|
| Report a correction | Report a correction (opens in a new tab) |
|---|
| Record identifier | 7d255869-22c2-4497-b5d3-aba20a685a6a |
|---|
| Record created | 2024-09-11 |
|---|
| Record modified | 2024-09-12 |
|---|
