Assessing reclamation potential of abandoned drylands using knowledge-guided machine learning (KGML) and remote...

Area of Science:

• Agricultural Science
• Environmental Science
• Data Science

Background:

• Escalating water scarcity and food security demands necessitate assessing abandoned dryland reclamation potential.
• Dryland agriculture faces significant challenges, requiring innovative management strategies.

Purpose of the Study:

• To develop and validate a knowledge-guided machine learning framework for water-carbon spatial optimization in dryland reclamation.
• To identify suitable abandoned dryland areas for reclamation and estimate economic returns.

Main Methods:

• Integrated a knowledge-guided machine learning framework with remote sensing.
• Predicted irrigation water use, net primary productivity, and soil organic carbon at 1-km resolution.
• Validated the framework across the Middle Reaches of the Yellow River Basin.

Main Results:

• Identified approximately 40% (∼800,000 hectares) of abandoned dryland as suitable for reclamation, with an estimated annual economic return of USD 40 million.
• Reclamation performance is optimized under water-conserving regimes and high-potential zones, not solely by vegetation growth.
• Reclamation potential is projected to remain stable under low- and medium-emission scenarios but decline under high-emission trajectories by 2031-2050.

Conclusions:

• The developed framework is transferable, cost-effective, and demonstrates stability for evaluating reclamation potential and guiding climate-adaptive resource management in global drylands.
• Reclamation offers ecological and economic co-benefits but requires substantial upfront investment and long-term fiscal support due to a lengthy payback period.
• The study highlights the need for sustained financial support in the early stages of reclamation to ensure long-term financial self-sufficiency and successful transition.