A deep learning-based model for estimating pollution fluxes from rivers into the sea and its optimization

Area of Science:

• Environmental Science
• Marine Biology
• Data Science

Background:

• Pollution from rivers is the primary source of contaminants in nearshore marine environments.
• Understanding the spatiotemporal dynamics of pollution fluxes is crucial for coastal ecosystem health.

Purpose of the Study:

• To estimate pollution fluxes into the sea and analyze their spatiotemporal heterogeneity.
• To develop a deep learning model for simplifying pollution flux estimation.
• To identify the contribution rates of pollution from different spatial gradients.

Main Methods:

• Utilized a source-sink process model for basin-estuary-coastal water systems.
• Developed a deep learning model incorporating socio-economic and meteorological data.
• Proposed a method to estimate pollution flux contribution rates based on spatial factors.

Main Results:

• Total nitrogen and phosphorus fluxes in the Bohai Sea Rim Basin showed variations between 1980-2020.
• Rural life and livestock were identified as major contributors to total nitrogen and phosphorus pollution.
• The deep learning model achieved over 90% accuracy in estimating runoff pollution fluxes.
• Areas with 0-100m elevation, 50-100km from the coast, and coastal districts showed the highest pollution contribution rates.

Conclusions:

• Pollution fluxes are significantly influenced by human economic activities and environmental policies.
• The developed deep learning model offers a simplified and accurate method for estimating pollution fluxes.
• Identifying high-contribution regions provides critical data for adaptive management of nearshore ecosystems.