Dataset: Remote sensing-based prediction of permanganate oxidisable carbon (POXC) in organic cassava plots, Northeast Thailand

This dataset supports a study evaluating multi-source remote sensing (RS) features for predicting soil organic carbon in 248 GPS-referenced organic cassava plots across Yasothon, Amnat Charoen, and Ubon Ratchathani provinces, Northeast Thailand (dry season 2023–2024, November 2023–April 2024; Acrisols/Ultisols, sandy-loam texture). Dataset contents (8 files): 1. README with full documentation 2. Plot-level soil properties (TOC, POXC, pH, BD, soil order USDA/WRB; n=248) 3. 39 RS variables extracted; 24 retained for analysis via Google Earth Engine (Sentinel-2, Sentinel-1 SAR, MODIS LST, SRTM terrain; n=247; 246 used in multivariate analyses (LDA, mediation, RF)) 4. Complete Python analysis pipeline (H1–H4): cv_random_state=109, perm_random_state=1, n_permutations=999 5. GEE extraction script (JavaScript) 6. Statistical results (H1 LDA, H2 Mediation, H3 Moderation, H4 RF) 7. Organic fertilisation records — anonymised (Farmer-01 to Farmer-15; 3 farms per management age group, Years 1–5) 8. QC log Key results: POXC (R² = 0.175, RMSE = 0.326 g kg⁻¹, RMSE/SD = 0.906) was predictable from RS; TOC was not (R² = −0.087). Elevation (MDI = 25.2%), dry-season LST, and inter-annual NDVI were the principal predictors. LDA: CV accuracy = 44.6 ± 7.8%, permutation p = 0.109 (n=999). Farmer names and district are withheld (personal data); province is retained. Full identification available from the corresponding author upon request. Associated manuscript: Aumtong et al. (2026), Remote Sensing Applications: Society and Environment (RSASE; ISSN 2352-9385). Submitted to RSASE; DOI pending acceptance.