Biochar application age in tropical longan orchards: soil physical, chemical, and carbon data

This dataset supports the research article "Artificial Intelligence-GuidedClassification of Biochar Application Age from Routine Soil Properties:A Random Forest Framework for Re-application Decision Support in TropicalLongan Orchards" submitted to Biochar (Springer Nature).

Study design: Factorial randomised complete block design (RCBD). 143 soilsamples were collected from 12 longan orchards in Mae Pang subdistrict,Phrao district, Chiang Mai Province, northern Thailand, across a 3-yearbiochar chronosequence. Orchards received a single woody-biochar applicationof 15 kg per tree (approximately 2.34 t/ha at 8 x 8 m spacing) by trenchincorporation. Sampling was stratified by 4 years-after-application classes(YAA: 0, 1, 2, 3) and 4 soil-depth increments (0-15, 15-30, 30-45, 45-60 cm).Soils are classified as Typic Hapludults (Ultisols) of the Mae Taeng soilseries, characterised by strongly acidic surface horizons (pH 4.2-4.8) andlow cation exchange capacity (4-8 cmol(+)/kg).

Contents: (1) Main dataset biochar_cleaned.csv with 143 rows x 35 variables(soil physical, chemical, organic-carbon fractions, and derived ratios);(2) full variable codebook in CSV and Markdown formats; (3) all derivedanalysis outputs reported in the manuscript tables (PCA eigenvalues andloadings, Random Forest feature importance from MDI, permutation, and SHAPmethods, per-class metrics, and a 3-way pseudo-replication sensitivitycomparison including tree-level ANOVA, plot-level ANOVA, and cluster-robustregression).

Key findings supported by this dataset:- A Random Forest classifier using 17 routine soil features predicted  years-after-application with 83.2 +/- 7.5% accuracy (F1-macro = 0.833;  Cohen's kappa = 0.776), confirmed by nested 5x3-fold CV (82.5 +/- 8.5%)  and depth-grouped CV (83.2%).- Convergent feature-importance rankings (MDI, permutation, SHAP)  consistently identified SO4 (sulphate), pH, and silt as dominant  predictors across all three methods.- Soil organic matter (SOM) at 0-15 cm peaked at Year 1 (+59.8% above  baseline) and remained elevated at Year 3 (+47.0%); available phosphorus  accumulated by +122% over the 3-year period (reaching 466 mg/kg).- PCA revealed a positive multivariate soil-quality trajectory (PC1  centroid: Year 0 = -0.65 to Year 3 = +0.49).- Effects were confined to 0-15 cm depth; depth dominated overall  variation (p < 0.001).

Reproducibility: All manuscript analyses can be reproduced from this datasetusing Python 3.12 with scikit-learn 1.8, xgboost 3.2, pandas 2.3,matplotlib 3.10, seaborn 0.13, shap, and statsmodels; random seed = 42throughout. See data/biochar_cleaned_codebook.md for variable descriptionsand data quality notes, and README.md for a complete overview and citationinformation.

Funding: This research was self-funded by the authors under the project"Study on the duration after biochar application on physical and chemicalproperties of longan soil" (Project code: OT-69-010), registered withThe Office of Agricultural Research and Extension, Maejo University,Thailand. No external funds or grants were received.

License: CC BY 4.0 — free to share and adapt with attribution.