Evaluation method for ecology-agriculture-urban spaces based on deep learning.

Area of Science:

• Environmental Science
• Urban Planning
• Agricultural Science

Background:

• Increasing global population and land degradation necessitate balancing urban development, food security, and ecological conservation.
• Coordinating these elements is vital for achieving sustainable development goals.
• Assessing the integrated Ecology-Agriculture-Urban (E-A-U) space is critical for effective land management.

Purpose of the Study:

• To develop and validate an advanced deep learning model for evaluating the E-A-U space.
• To apply the model to Yulin City, China, as a representative case study.
• To demonstrate the model's superiority over traditional methods in spatial evaluation.

Main Methods:

• Developed a Self-Attention Residual Neural Network (SARes-NET) model based on the "Double Evaluation" framework.
• Applied the SARes-NET model to assess the E-A-U space in Yulin City.
• Conducted comparative validation against Logistic Regression, Naive Bayes, GBDT, RF, and ANN models.

Main Results:

• The SARes-NET model demonstrated superior simulation performance compared to five other models.
• The model effectively captured complex non-linear relationships within the E-A-U spatial data.
• Spatial analysis revealed distinct agricultural dominance in the northwest and urban/ecological areas in the southeast of Yulin City.

Conclusions:

• The deep learning-guided E-A-U spatial evaluation offers an innovative approach for national spatial planning.
• SARes-NET provides a robust method for assessing land use suitability and ecological-urban-agricultural interactions.
• This approach has significant implications for national-level territorial assessments and sustainable development strategies.