A method for prairie fire dynamic risk assessment

By optimizing the DBSCAN clustering and XGBoost model, the interpolation error and dynamic changes in grassland fire susceptibility assessment in alpine grassland ecosystems were resolved, enabling near real-time grassland fire risk assessment and improving the accuracy and real-time performance of the assessment.

CN122155411APending Publication Date: 2026-06-05QINGHAI NORMAL UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
QINGHAI NORMAL UNIV
Filing Date
2026-03-04
Publication Date
2026-06-05

AI Technical Summary

Technical Problem

Existing grassland fire susceptibility assessment technologies suffer from problems such as large interpolation errors of meteorological driving factors in alpine grassland ecosystems, difficulty in reflecting short-term dynamic changes, and insufficient specificity of feature extraction. These issues affect model stability and assessment accuracy, making it difficult to meet the needs of daily or event-scale fire risk assessment.

Method used

The DBSCAN clustering and minimum concave shell area method were used to determine the grassland fire impact area. The XGBoost machine learning model was then used for multiple iterations of training to obtain grassland fire feature data. Dynamic risk assessment was then conducted through optimal time window and parameter optimization.

Benefits of technology

It enables near real-time grassland fire risk assessment of target areas on specified dates, provides dynamic prevention and risk management solutions, and improves the accuracy and real-time nature of the assessment.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122155411A_ABST
    Figure CN122155411A_ABST
Patent Text Reader

Abstract

This invention relates to the field of fire monitoring technology, and more specifically to a method for dynamic risk assessment of grassland fires. The method includes: acquiring multi-source data such as historical active fire points, satellite remote sensing, digital terrain, and grassland distribution data for the monitoring area; preprocessing active fire points in the grassland distribution data as positive samples with time attributes, generating fire impact areas through DBSCAN clustering, constructing negative samples with time distributions consistent with the positive samples from non-impact areas, merging them to form a total grassland fire sample and dividing it into training and testing sets. Based on the time attributes of the samples, remote sensing factors, tassel transformation components, and terrain factors within the time window before the assessment date are extracted as features. These features are input into an XGBoost model for training, and hyperparameters are optimized using grid search and five-fold cross-validation. Through iterative training and verification using multiple time windows, the time window with optimal accuracy is determined. Five typical dates in 2024 are selected, and the optimal model and window are used to conduct grassland fire risk assessment, outputting a fire occurrence probability grid; based on historical fire point clustering characteristics, a grading interval is determined, classifying the probability into five risk levels: extremely high, high, medium, low, and extremely low. This method can dynamically assess grassland fire risk on different dates in near real-time and visualize risk areas, providing support for disaster prevention and mitigation decision-making.
Need to check novelty before this filing date? Find Prior Art