7.5-4.0 Ma reconstructed CO₂ with paleosol-CO₂ method in Chinese Loess Plateau deposits (Jiaxian and Shilou sections)

We generate a high-resolution, highly variable terrestrial CO₂ record spanning 7.5–4.0 Ma by applying a machine learning-based multi-proxy paleosol-CO₂ inversion to Chinese Loess Plateau deposits (Jiaxian and Shilou sections).<b>Data Sources:</b><b>Inorganic Isotopes, Elements &amp; Grain Size:</b> The carbonate stable isotopes (d13Cc, d18Oc), elemental concentrations (aFe, aSi, aAl, fFe), grain size (D), and magnetic susceptibility (Xlf) were measured in this study across both sections.<b>Organic Carbon Isotopes (</b><b>d13Co</b><b>):</b>Jiaxian Section: Values are interpolated from the high-resolution nodule organic isotope record published by Gallagher et al. (2021).Shilou Section: Values were measured directly in this study on decarbonated paleosol samples.<b>Formation Temperature (</b><b>Temperature</b><b>):</b> Interpolated from soil temperature records derived from carbonate nodule clumped isotopes (Δ47) (Da, 2025).<b>Atmospheric Carbon Isotopes (</b><b>d13Ca</b><b>):</b> Estimated from high-resolution marine carbonate records (Tipple et al., 2010).<b>Atmospheric CO₂ &amp; Sz Reconstructions:</b> Soil-respired CO₂ (Sz) was predicted using our optimized Gradient Boosting machine learning model, which was then inverted with the isotopic mixing ratio (R) to reconstruct atmospheric CO₂.<b>Column Definitions and Units:</b><b>Sample ID</b>: Unique identifier for each Red Clay paleosol sample.<b>Depth(m)</b>: Stratigraphic depth of the sample (meters).<b>Age (Ma)</b>: Estimated age of the sample (Million years ago).<b>Xlf</b>: Low-frequency magnetic susceptibility. (Xlf_std: standard deviation).<b>Ampl 46</b>: Signal amplitude for mass 46 during isotope ratio mass spectrometry.<b>d13Cc</b>, <b>d18Oc</b>: Carbon and oxygen isotopic composition of pedogenic carbonate (‰, VPDB).<b>d13Cc_std</b>, <b>d18Oc_std</b>: 1σ uncertainties of the carbonate isotopic measurements.<b>d13Cc_n</b>, <b>d18Oc_n</b>: Number of replicate measurements for carbonate isotopes.<b>d13Co</b>: Carbon isotopic composition of soil organic matter (‰, VPDB).<b>d13Co_std</b>: 1σ uncertainty of the organic carbon isotopic composition.<b>Temperature</b>: Estimated formation temperature of pedogenic carbonates (°C).<b>Temperature_std</b>: 1σ uncertainty of the formation temperature (°C).<b>d13Ca</b>: Estimated carbon isotopic composition of atmospheric CO₂ (‰, VPDB).<b>d13Ca_std</b>: 1σ uncertainty of atmospheric CO₂ isotopic composition.<b>aAl, aFe, aSi</b>: Amorphous aluminum, iron, and silicon concentrations (mass %, quantified by their corresponding oxides).<b>fFe</b>: Free iron oxides concentration (mass %, quantified by its corresponding oxide).<b>[Element]_std</b>: 1σ standard deviation for the respective element concentration.<b>[Element]_sem</b>: Standard error of the mean for the respective element concentration.<b>[Element]_n</b>: Number of replicate measurements for the element extraction.<b>n_measurements_max</b>: Maximum number of replicates among the elemental measurements.<b>aFe/aSi, aFe/aAl, aSi/aAl, aFe/fFe, aSi/fFe, aAl/fFe</b>: Calculated stoichiometric ratios of the respective amorphous/free element concentrations.<b>[Ratio]_std</b>: 1σ uncertainties of the elemental ratios, propagated using a Taylor expansion series.<b>D</b>: Volume weighted mean grain size, D[4, 3] (μm).<b>D_std</b>: 1σ uncertainty of the grain size measurement (μm).<b>R</b>: Calculated mixing ratio of atmospheric CO₂ versus soil-respired CO₂.<b>R_std</b>: 1σ uncertainty of the mixing ratio R.<b>R_90_low, R_90_high</b>: Lower and upper bounds of the 90% confidence interval for R.<b>Sz_mean</b>: Machine-learning predicted concentration of soil-respired CO₂ (ppm).<b>Sz_std</b>: 1σ uncertainty of Sz (ppm).<b>Sz_90_low, Sz_90_high</b>: Lower and upper bounds of the 90% confidence interval for Sz.<b>CO2_mean</b>: Reconstructed atmospheric CO₂ concentration (ppm).<b>CO2_std</b>: 1σ uncertainty of the reconstructed atmospheric CO₂ (ppm).<b>CO2_90_low, CO2_90_high</b>: Lower and upper bounds of the 90% confidence interval for the reconstructed CO₂.Gallagher, T. M. <i>et al.</i> Regional Patterns in Miocene-Pliocene Aridity Across the Chinese Loess Plateau Revealed by High Resolution Records of Paleosol Carbonate and Occluded Organic Matter. <i>Paleoceanogr. Paleoclimatology</i> <b>36</b>, e2021PA004344 (2021).Da, J. Rainy spring delayed the late Neogene expansion of C4 biomass over East Asia. 78003 Bytes figshare https://doi.org/10.6084/M9.FIGSHARE.29137106.V1 (2025).Tipple, B. J., Meyers, S. R. &amp; Pagani, M. Carbon isotope ratio of Cenozoic CO₂: A comparative evaluation of available geochemical proxies: CENOZOIC<i>δ</i>¹³C_CO2. <i>Paleoceanography</i> <b>25</b>, (2010).