Study on Particulate Matter Emissions from Forest Combustible Material Combustion in the Xiao Xing’an Mountains Region with Random Forest and XGBoost Model Validation
Description
Laboratory combustion measurements of PM2.5 emissions from 22 forest fuel species collected in the Xiao Xing'an Mountains, Heilongjiang Province, Northeast China (6 conifers, 9 broad-leaved trees, 7 shrubs). Data comprise 6,966 measurements across controlled treatments of plant part, moisture content, fuel load, wind speed, combustion mode, and season. PM2.5 was collected on filters using a JCH-6120 medium-flow gravimetric sampler at 100 L/min. Broad-leaved samples were sampled for 26 min, giving a sampled air volume of 2.6 m3 per measurement; conifer and shrub samples burned more rapidly and required shorter durations. IMPORTANT: PM2.5 units differ by vegetation category - conifer and shrub values are in grams (mass per sample); broad-leaved values are in ug/m3 (mass concentration). Data should be analysed within each category, not pooled. Version 2 supplies a corrected merged dataset (combined_data.csv) together with the full analysis code. See README.txt for variable definitions, the species list, and details of the correction. Version 3 (revision R1 of the article) adds the machine-learning code with grouped cross-validation (final_v5_grouped.R) and the statistical analyses of the revision (revision_statistics.R), with their outputs; the files of the original submission are kept in the folder original_submission_v4. See README_revision_R1.txt for details.
Files
Steps to reproduce
Fuel samples were collected from 22 forest species in the Xiao Xing'an Mountains, oven-dried at 40 C, then conditioned to the target moisture contents by rewetting and equilibration. Samples were burned in a laboratory combustion chamber under controlled treatments of plant part, fuel load, moisture content, wind speed (conifers), combustion mode (shrubs), and season (broad-leaved species). For shrubs, flaming and smoldering were separate tests on separate samples, each collected on its own filter; at 40% moisture content only smoldering tests were made. PM2.5 was collected on filters using a JCH-6120 medium-flow gravimetric sampler at 100 L/min (0.1 m3/min). Broad-leaved samples were sampled for 26 min, giving a sampled air volume of 2.6 m3 per measurement; conifer and shrub samples burned more rapidly and required shorter durations. Filter mass difference was used to derive PM2.5 emissions. Each treatment combination was run in triplicate. To reproduce the results of the revised article: (1) place combined_data.csv in the same folder as the R scripts; (2) run final_v5_grouped.R (packages randomForest and xgboost; seed 123) to obtain the grouped cross-validation results, Figures 4 and 5 and Table S7; (3) run revision_statistics.R (package car) to obtain the linear models, Type II ANOVA, the common-factor analysis (Table S8) and the sensitivity checks. Figures 2 and 3 are produced by figures_2_3.R and the original statistics by statistical_analysis.R. Variable definitions, units and the species list are given in README.txt; the changes made in version 3 are listed in README_revision_R1.txt.
Institutions
- Northeast Forestry UniversityHeilongjiang, Harbin
Categories
Funders
- National Key R&D Program of ChinaGrant ID: 2024YFF1306200, 2024YFF1306203
- Heilongjiang Provincial Natural Science Foundation Joint Fund Support ProjectGrant ID: PL2026C146