Research on Pine Wilt Disease Spread Prediction Based on an Improved Light Gradient Boosting Machine Model.

Area of Science:

• Forest pathology and ecology
• Computational biology and machine learning

Background:

• Pine wilt disease poses significant ecological and economic threats to China's forests.
• Traditional disease prediction models lack the necessary accuracy for effective management.

Purpose of the Study:

• To develop and validate an enhanced machine learning model for predicting pine wilt disease trends in China.
• To identify key anthropogenic and natural factors influencing disease spread.
• To provide a theoretical basis for proactive disease monitoring and prevention.

Main Methods:

• Collected county-level pine wilt disease occurrence data (2017-2022).
• Incorporated anthropogenic (wood imports, road density, adjacent counties, wood factories) and natural (temperature, humidity, wind speed) factors.
• Utilized Pearson correlation and Light Gradient Boosting Machine (LGBM) for feature selection (17 factors identified).
• Performed spatial analysis on epidemic subcompartments (2022-2023) to understand distribution and relationships.
• Enhanced the LGBM model using Bayesian, sparrow search, and hunter-prey optimization algorithms.

Main Results:

• The enhanced LGBM model demonstrated superior accuracy, precision, recall, sensitivity, and specificity compared to traditional methods.
• Spatial analysis revealed disease concentration near roads and spatial links between new and old epidemic areas.
• Identified 17 significant factors influencing pine wilt disease spread.
• Current disease hotspots are in central-southern and northeastern China.

Conclusions:

• The enhanced LGBM model provides a robust tool for predicting pine wilt disease outbreaks.
• Future spread is predicted to expand into northeastern and southern regions of China.
• The findings support enhanced monitoring and targeted prevention strategies to mitigate disease impact.