Machine learning for land use change analysis in environmental protection areas.

Area of Science:

• Environmental Science
• Remote Sensing
• Machine Learning Applications

Background:

• Anthropogenic land use and land cover change (LULCC) combined with climate variability significantly challenge environmental protection areas (EPAs).
• Traditional LULCC methods often overlook dynamic environmental drivers, limiting the understanding of complex landscape-climate interactions.
• EPAs face ecosystem structure alteration, vegetation degradation, and disrupted climate regulation due to these pressures.

Purpose of the Study:

• To investigate environmental dynamics within the Guaraqueçaba Environmental Protection Area.
• To evaluate automated classification and prediction of land use change impacts.
• To analyze landscape-climate interactions using multitemporal data and machine learning.

Main Methods:

• Analyzed a 15-year (2009-2023) multitemporal dataset including climate variables (precipitation, temperature, humidity) and vegetation indices (NDVI).
• Utilized a balanced subset of 3.6 million records to address class imbalance in predictive modeling.
• Developed and compared predictive models: multiple linear regression (MLR), k-nearest neighbors (KNN), and random forest (RF), assessing performance with accuracy, R², precision, recall, and F1-score.

Main Results:

• Observed pronounced local climate changes, including rising temperatures in modified areas and altered humidity linked to vegetation loss.
• Random Forest (RF) model demonstrated superior predictive performance, achieving up to 96% accuracy and 88.6% R², effectively modeling nonlinear interactions.
• Precipitation and Normalized Difference Vegetation Index (NDVI) were identified as the most significant drivers of LULCC processes.

Conclusions:

• Machine learning approaches, especially RF, are effective for identifying environmental degradation trajectories in protected areas.
• The study provides a robust framework for targeted mitigation strategies and policy development in EPAs.
• Findings highlight the critical interplay between LULCC, climate variability, and vegetation health in environmental management.