2026-10-06T04:19:00Z https://api.figshare.com/v2/oai

oai:figshare.com:article/34059390 2026-10-02T19:08:51Z category_25780 category_25924 category_26896 item_type_3 month_year_10_2026

Datasets generated for O'Mara et al. (2026), Fire biomarkers in marine sediments record terrestrial fire activity. AGU Advances Nicholas OMara (18793204) Organic geochemistry Palaeoclimatology Fire ecology PAH Anhydrosugar Fire proxy Dataset of organic fire proxy measurements from a transect of marine sediment core tops from the Iberian through Western African margins (40°N-34°S). Measurements include concentrations and 230ThXS-normalized mass accumulation rates of a suite of polycyclic aromatic hydrocarbons (PAHs) and anhydrosugars. Core samples were provided by the Lamont-Doherty Core Repository of Columbia University. Compound concentration measurements were made via gas chromatography-mass spectrometry at the Lamont-Doherty Earth Observatory (LDEO) Organic Geochemistry Lab using an Agilent 7890A GC coupled to an Agilent 5975C MS. File descriptions: Table S1. Compound info for PAH and anhydrosugar compounds analyzed in this study including name, abbreviation, compound type, molecular weight, ring number, and integration ion. Table S2. Sample info including laboratory identification number, latitude (°N), longitude (°E), water depth (m), biomarker concentrations (ng/g), 1 sigma concentration uncertainties, and 230ThXS-derived mass accumulation rates (ng cm-2 kyr-1) for all compounds analyzed. Table S3. Source-area environmental data: annual and monthly climate, vegetation and satellite fire-metric values within each modeled source area (500, 1500 and 3000 m atmospheric back-trajectory sheds and river basins) for every core-top site. Table S4. Compositional PCA biomarker loadings. PCA loadings of the seven biomarker variables (two amount variables and five compositional log-ratios) on PC1 and PC2, internal to the ordination. The ordination is a correlation-matrix PCA of the z-scored variables. Table S5. Fire-metric envfit statistics. R² and permutation p-values for each fire metric fitted as an external vector overlay onto the compositional PCA (envfit, Yeo-Johnson transformed), by source area. Table S6. Random Forest validation and out-of-sample skill. For each fire metric and source area: the out-of-sample R² of the biomarker Random Forest, estimated by per-site held-out cross-validation against a mean-only baseline; the out-of-sample R² of the transferable-predictor null Random Forest (bulk sediment mass accumulation rate, %TOC, water depth, distance offshore, accumulation duration, dust flux, terrestrial n-alkane mass accumulation rate and precipitation at the core site) evaluated identically; the gain of the biomarker model over that null; and a flag marking the combinations deemed skillful, defined as an out-of-sample R² that exceeds 0.1, exceeds the transect mean predictor, and exceeds the transferable-predictor null. Table S7. Leave-one-region-out spatial cross-validation. For each fire metric and source area, the out-of-sample R² of the biomarker Random Forest and of the transferable-predictor null under leave-one-region-out cross-validation over six ecoregion-based region groupings (Iberian margin, Saharan, Sahelian, central African forest, southern African savanna and Cape, see Figure S11), each withheld in turn, together with the gain of the biomarker model over that null. Directly comparable to the interpolation-based estimates in Table S6. 2026-10-02T19:08:51Z Dataset Dataset 10.6084/m9.figshare.34059390.v1 https://figshare.com/articles/dataset/Datasets_generated_for_O_Mara_et_al_2026_Fire_biomarkers_in_marine_sediments_record_terrestrial_fire_activity_i_AGU_Advances_i_/34059390 CC BY 4.0