Multi-temporal image analysis of wetland dynamics using machine learning algorithms

Area of Science:

• Environmental Science
• Remote Sensing
• Geospatial Analysis

Background:

• Wetlands are vital ecosystems providing critical services like water purification and habitat provision.
• Global wetland decline necessitates advanced mapping and monitoring techniques.
• Machine learning and earth observation data offer new solutions for wetland assessment.

Purpose of the Study:

• To analyze wetland dynamics in Pakistan's Thatta region (Haleji & Kinjhar Lake).
• To evaluate the effectiveness of different classification systems for wetland mapping.
• To predict future wetland changes under various environmental and anthropogenic scenarios.

Main Methods:

• Utilized Google Earth Engine and Landsat imagery for wetland analysis.
• Applied spectral indices and four classification techniques for mapping.
• Employed Random Forest algorithm for accurate wetland classification.
• Conducted change detection analysis from 1990-2020 and future scenario modeling.

Main Results:

• Random Forest achieved 87% accuracy in wetland classification.
• Significant wetland loss (352.8 sq.km) observed in the Thatta region between 1990-2020.
• Key drivers of loss include agriculture, urbanization, groundwater extraction, and climate change.
• Future projections indicate continued wetland deterioration under various scenarios.

Conclusions:

• Urgent conservation and restoration efforts are required to mitigate wetland loss.
• Satellite data analytics and machine learning provide crucial insights for sustainable wetland management.
• Effective policies are needed to address anthropogenic pressures and climate change impacts on wetlands.