Introduction

The Late Miocene, 11.6 to 5.3 million years ago (Ma), is a time period that witnessed significant global changes in tectonics, biological communities, and climate, including the so-called Late Miocene cooling (LMC) that resulted in the establishment of near-modern ocean temperatures between 7 and 5.4 Ma1,2. During this time, ocean temperatures decreased substantially in deep and shallow waters as part of the long-term Cenozoic cooling and atmospheric CO2 decline3,4,5. This period offers a unique opportunity to investigate interactions of climate and global carbon cycling under a warmer-than-modern condition, potentially providing valuable information to inform models and thereby improving predictions related to climate change and terrestrial ecosystem responses6,7,8,9.

Reconstructions of sea surface temperatures (SST) highlight significant cooling in both hemispheres between 7 and 5.4 Ma1,2. A global carbon isotope (δ13C) excursion occurred during the Late Miocene that reveals carbon cycle dynamics involving the terrestrial and deep ocean carbon reservoirs, thought to drive Late Miocene climate cooling2. However, the dynamic processes and driving forces underlying the LMC are debated10. Some studies suggest that a gradual decline in atmosphere CO2 played a central role in driving global cooling11 and associated changes in terrestrial ecosystems12,13. However, the role of enhanced silicate and phosphorus (P) weathering during the LMC is less clear14,15,16. Silicate weathering, if coupled to carbonate formation (as opposed to clay synthesis), contributes to the net consumption of ocean-atmospheric CO217,18,19,20,21, while P cycling affects atmospheric CO2 by regulating oceanic productivity and organic matter burial22,23,24. In this study, we investigate the relative roles of silicate weathering and P cycling in regulating the global climate over the LMC.

In this study, we use phosphorous-to-iron ratio (P/Fe) records of a hydrogenetic ferromanganese crust from the West Pacific (Fig. 1 and Supplementary Fig. 1) to infer oceanic P concentrations and continental P weathering during the Late Miocene. By employing a global biogeochemical model, we demonstrate that increased P weathering alone suffices to lower atmospheric CO2 and explain the LMC. Contrary to the traditional view that assumes a direct coupling between P and silicate weathering, we posit that these processes may have undergone different dynamics during the LMC.

Fig. 1: Modern surface ocean phosphate and general deep Pacific Ocean circulation patterns (modified from ref. 89).
figure 1

Fe-Mn crust MID03 (red star), ODP/DSDP sites (white circles) and Fe-Mn crust (yellow square) are plotted in this figure. Detailed information about these deposits are provided in Supplementary Table 1. Yellow and red circles with a central point represent the upwelling of Lower Circumpolar Deep Water (LCDW) and Upper Circumpolar Deep Water (UCDW), respectively (modified from refs. 90,89). The white arrow shows the flow carrying modified North Pacific Deep Water (NPDW) to mix with UCDW. Arrows indicate wind directions for the East Asian Winter Monsoon (EAWM; light blue) and the prevailing westerlies (dark blue arrow over land for winter; pink arrow over land for summer; thick red arrow over the ocean). The figure was created with Ocean Data View91.

Full size image

Results and discussion

Shifts in oceanic P reservoir and primary productivity

Phosphorus cycling plays a prominent role in regulating oceanic primary production on various time scales, with profound impacts on oceanic biogeochemistry, ecosystem structure, and global climate22,25,26. In all modern aquatic systems, phosphorus is primarily sourced from the weathering of continental materials27,28. Key processes responsible for the internal recycling of dissolved phosphate include mixing of the surface and deep oceans, and the release of organic phosphorus back into the deeper ocean22.

At site MID03 from the North Pacific, crust samples are hydrogenetic (Supplementary Figs. 2 and 3 and Supplementary Text 1), reflecting genesis from cold, deep water. In this crust, Fe2O3 and P2O5 show a strong positive correlation, while a significant inverse relationship is observed between MnO2 and P2O5 (Supplementary Fig. 4 and Supplementary Tables 3 and 4). This indicates that P is mainly hosted in iron oxyhydroxides of the ferromanganese crust. Previous studies have revealed that P/Fe of iron oxyhydroxide precipitates are strongly correlated with ambient deep-water phosphate levels29,30,31,32. We build on these previous studies31,32 by employing P from iron-oxyhydroxide P/Fe ratios in crust MID03 as a proxy for past deep-water phosphate levels.

Our results indicate a pronounced increase in oceanic P concentrations during the Late Miocene (Fig. 2d), with P/Fe ratios increasing from ca. 0.032 to around 0.046 between 7.5 and 4 Ma, followed by a notable decrease from about 0.046 to 0.035 between 4 and 3 Ma. P/Fe further increases to 0.042 at 2 Ma and then decreases to 0.038 at 1 Ma after which it remains roughly stable into the Pleistocene (Supplementary Fig. 5b). The modern deep-sea P concentration is comparatively homogeneous at ~ 3 μM33, although we note that small variations in P are documented globally, especially in the Pacific Ocean (Fig. 3c)15,34,35,36,37. We suggest that P/Fe changes in our MID03 crust possibly reflect a whole deep ocean P shift during the Late Miocene. Therefore, we calculated the evolution of deep ocean P concentrations over the past 10 Myrs using the P/Fe record (Fig. 4).

Fig. 2: Proxy records of nutrient utilization and paleoproductivity in the West Pacific compared to regional and global paleoclimate and paleoceanographic reconstructions since 11 Ma.
figure 2

a Stack sea surface temperature (SST) anomalies for the Northern Hemisphere (NH), Southern Hemisphere (SH), and tropical1. b 3HeET-derived opal fluxes and 3HeET-derived Baxs fluxes from ODP 885/88692. c Nutrient utilization and productivity records BaEF at Sites U142538. d P/Fe ratio in crust MID03 (this study). e Opal flux from East Pacific ODP Site 84940. f CaCO3 accumulation rate from the three ocean basins93. g Paleoproductivity at ODP Site 1088 from the Southern Ocean41. h Redox-sensitive metals (normalized Ni, Mn, Cu composition) and anhysteretic remanent magnetization (ARM), as proxies for the relative magnetofossil abundance in crust MID03 (this study). i Foraminifera-bound δ15N and reconstructed [PO43-] based on the P and Fe contents at DSDP Site 59815.

Full size image
Fig. 3: Compiled global and regional palaeoclimatic indicators of the past 11 Myr.
figure 3

a Global mean surface temperatures (GMST) estimated from benthic δ18O data based on complied from ocean drilling core sites94. b Sea Surface Temperature (SST) compilation over the past 11 Myr1. c Phosphorus accumulation rate (μmol cm−2 kyr−1) in the equatorial Pacific Ocean36,37. d P/Al ratio in crust MID03 (this study). e Ge/Si ratio in opaline silca from diatoms75. f Soil carbonate δ13C from northern Pakistan95; vertical blue bars show the standard view of the presence and extent of full-scale/permanent ice sheets (solid bars) and those thought to have been partial/ephemeral (broken bars). The timing of C4 plant expansion is indicated at the base14.

Full size image
Fig. 4: Evolution of simulated oceanic P contents and atmospheric pCO2 along with reconstructed sea surface temperature (SST) over the last 11 Myr.
figure 4

a Model results for P weathering rate. b Model results for silicate weathering rate. c Oceanic P content. Model result of the deep ocean box (solid line). The points are calculated from the P/Fe data of this study using a linear scale factor, with the modern oceanic P concentration calibrated at 3 μM84. d Atmospheric pCO2. Model result (solid line) compared with proxy data (open circles)68. e Reconstructed SST1 from different latitudes. In this study, a Monte Carlo method was employed by varying the P weathering flux and climate sensitivity. A 5% variation was introduced to the P weathering flux using a normal distribution, while climate sensitivity was adjusted within a range of 4–6 °C also using a normal distribution. An enhancement factor of P weathering was applied to this simulation (Scenario 1), which is shown in Supplementary Fig. 8. Other than the enhanced forcing factor for P weathering, the other forcings include uplift as fixed at modern levels.

Full size image

Our reconstructed rise in oceanic P concentrations can well explain a previously documented productivity enhancement observed at multiple locations during the Late Miocene. For example, previous studies have shown increased opal accumulation at ODP 885/886 (Fig. 2b) and barium enrichment (BaEF; an established productivity proxy) in the Sea of Japan (Fig. 2c)38. The widespread and prolonged productivity peak in the low-latitude Pacific Ocean during the Late Miocene also aligns with a “biogenic bloom” in the eutrophic upwelling systems of the Indian37 and Pacific oceans39 (Fig. 2f and Supplementary Table 1). Specifically, in the eastern equatorial Pacific40 and Southern Ocean41, opal and paleoproductivity deposition (Fig. 2e and g) reached a peak between 7.0 and 6.4 Ma42,43 and declined after that until ~ 4 Ma. Given that P is a crucial limiting nutrient for primary productivity44,45, we propose that increased oceanic P levels were a main forcing of enhanced primary productivity beginning in the Late Miocene.

Nutrient cycling dynamics

Iron, phosphorus, and nitrogen are essential nutrients for phytoplankton growth and carbon sequestration46. Late Miocene phosphatization has also been observed along Western Pacific47,48 Chatham Rise49,50, Blake Plateau51, southwest Africa52, northwestern Iberia53, and southern California54. Major climatic shifts were thought to have initiated ocean phosphogenesis by promoting intense oceanic vertical mixing that mobilized a large amount of phosphorus previous stored in the deep ocean55.

To assess ocean circulation and upwelling processes over the LMC, we use the εNd recorded in crust MID03 and other samples from the West Pacific as a water mass tracer56,57. Our crusts from the West Pacific show a relatively radiogenic Nd isotope composition during the Late Miocene (Supplementary Fig. 5e and Fig. 7b), with consistency between lead and iron isotope data suggesting they are affected by similar changes in the source province58. The increase in εNd during 8–4 Ma likely indicates enhanced input of more radiogenic island arc materials at low latitudes58, pointing to a shift from marginal northwest Pacific deep waters to low-latitude equatorial deep waters in the region. We propose that climate cooling and the formation of Pacific deep water influenced deep-water upwelling in the Western Pacific during the Late Miocene38. These deep-water oscillations likely enhanced nutrient circulation within the ocean, which, together with elevated oceanic phosphorus concentrations, further stimulated primary productivity and phosphogenesis.

Moreover, the increase in oceanic P concentration could be closely connected with oceanic redox changes during the Late Miocene. Concurrent with the rise in oceanic P at ~ 7.5 Ma at the West Pacific site MID03 (Fig. 2d), a relative reduction of our measured oxygen-sensitive index in each subsample59 (calculated following the method of ref. 59) of crust MID03 indicates a simultaneous decrease in bottom water oxygen concentrations (Fig. 2h and Supplementary Fig. 5c). We have also measured the anhysteretic remanent magnetization (ARM) of the crust MID03, which is a proxy of the magnetotactic bacteria (MTB). The results (Fig. 2h and Supplementary Fig. 6) indicate the presence of MTB, further supporting a decrease in bottom water oxygen concentrations, as suboxic conditions inhibit aerophilic MTB biomineralization34. This occurred despite surface ocean cooling (Fig. 3a)1 and intensified ocean circulation (Supplementary Fig. 7b)36, attesting to enhanced deep-water deoxygenation in the Pacific Ocean in the Late Miocene15. Our results suggest that the amplified oceanic P fertilization effect and resultant higher biological production have outweighed this oceanic vertical mixing, leading to Pacific Ocean deoxygenation during the Late Miocene.

Dominant role of phosphorus fertilization in driving CO2 consumption

The Late Miocene (~ 11.6–5.3 Ma) witnessed significant climatic, tectonic, and environmental changes at a global scale5,10,60. This time interval was characterized by more intense, high-latitude cooling and glacial events as well as by atmospheric CO2 decline61. However, discerning the forcing behind these changes remains unresolved2. Asian monsoon studies suggest that chemical weathering, including both silicate and P weathering linked to Neogene crustal exhumation, played a pivotal role in reducing atmospheric CO217,20. It has been proposed that this additional sink might have contributed to the substantial consumption of atmospheric CO2 and the rapid global cooling observed in the Late Miocene16.

Although both silicate and P weathering can lead to the consumption of atmospheric CO2, the relative importance of these two processes on climate dynamics has not been established62,63. For instance, elevated P weathering linked to volcanic eruptions has been proposed as the driver for climate change during the Late Ordovician64. However, P and silicate weathering can be decoupled as P is primarily hosted in apatite and organic phases and may exhibit different reaction kinetics compared to silicates65. To assess the impact of P weathering on primary productivity and CO2 consumption during the Late Miocene, we use a global biogeochemical model66,67 and compare it to our reconstructed record of ocean P levels. The model is calibrated using contemporary biogeochemical fluxes, ocean circulation patterns, reservoir sizes, and seawater ion concentrations66,67. Our model results reveal decoupled P and silicate weathering regime across the Late Miocene.

First, we conducted model simulations by introducing a forced increase in P weathering using an enhancement factor (Supplementary Fig. 8), which was adjusted to match the observed changes in oceanic P concentrations. The results reveal that a ~ 30% rise in continental P weathering is required to produce the reconstructed peak oceanic P concentration at 4 Ma (Fig. 4c). This would lead to a significant ~ 60 ppm drop in atmospheric CO2 during the Late Miocene, effectively explaining all of the change in atmospheric CO2 record over this interval68 (Fig. 4d). Next, to explore the impact of coupled enhanced silicate and P weathering, we conducted a set of model simulations with a forced increase in both P and silicate weathering using the same enhancement factor (Supplementary Fig. 9) or directly scaled the rates of P and silicate weathering (Supplementary Fig. 10). When adjusted to match our P/Fe record, the simulation with a forced increase in both P and silicate weathering using the same enhancement factor results in a decrease in atmospheric CO2 from ~ 300 ppm to less than 150 ppm (Supplementary Fig. 9), a magnitude far too large to be reconciled with our proxy records (Fig. 4b). In the simulation with scaled P and silicate weathering, the model fails to reproduce the observed shift in seawater P concentration (Supplementary Fig. 10). This occurs because fully coupled P and silicate weathering should be balanced with degassing69, and thus, with a fixed degassing rate, significant changes are unlikely. Consequently, our P/Fe record, along with the global biogeochemical model, do not support a coupled shift in P and silicate weathering. Currently, continental margin sediments account for about half of the total reactive P buried in the ocean70. We speculate that sea-level change in the Late Miocene influenced the global ocean P cycle because the fraction of P eroded from shelf sediments was prone to transport to the open ocean65. This substantiates the predominant role of enhanced P weathering in contributing to the late Miocene cooling.

Considering the nature and timing of significant climatic and tectonic events in Asia, we posit that regional tectonic-induced P weathering71 was the primary trigger for the rapidly cooling climate and ice-sheet expansion along West Antarctica and in the Arctic by ~ 7 Ma (Fig. 3a and Supplementary Fig. 7a)72. The uplift and weathering of the Himalayan-Tibetan Plateau, as well as the Andes, may have accelerated the net flux of the bio-limiting nutrient phosphorus to the ocean, which is also supported by a noteworthy rise in P burial flux within marine sedimentary records (Fig. 3c and d)73,74. In this case, the temporary increase in nutrient input to the ocean accelerated dissolved element fluxes to the ocean and enhanced global oceanic productivity (Fig. 3e)75. Several mechanisms contribute to amplifying P weathering flux compared to silicate weathering through mountain uplift. Firstly, the elevation of mountains triggers a surge in soil erosion76, delivering a significant amount of reactive particulate P to the ocean77. This P can subsequently be released as dissolved P during diagenetic regeneration77. Secondly, extensive exposure to P-rich basic volcanic rocks in regions such as the Tibetan Plateau and the Andes results in enhanced P weathering relative to silicate weathering78. Exposure of sea-level lowlands due to initial cooling can further intensify the erosion of unconsolidated sediments, facilitating the transport of reactive particulate P to the ocean65.

Large-scale changes in terrestrial environments and ecosystems in subtropical and tropical regions1 possibly played a crucial role in elevating P inputs into the ocean. We note that the timing of this P shift corresponds with a global benthic δ13C decline and the expansion of C4 plants in subtropical regions occurred between 8 and 6 Ma (Fig. 3e)13. It has been postulated that such expansion altered the properties of land surface (e.g., increased the proportion of less dense grasslands), leading to greater soil erosion and higher soil organic matter input to the oceans79. As soil organic matter contains a significant amount of P, the remineralization of soil organic carbon would have also increased P input into the ocean, leading to a transient imbalance between input and export of P from the marine reservoir.

In summary, our data provide evidence for a persistent step-like increase in dissolved marine phosphate concentrations during the Late Miocene. Our findings suggest a prominent decoupling between P and silicate weathering during the LMC, highlighting the significance of terrestrial environmental and ecosystem changes in regulating P weathering and climate change. Our results provide a more comprehensive understanding of past oceanic nutrient variability and associated changes in weathering and climate impacts over the Cenozoic. Multiple P/Fe records from global crust samples would be helpful to further validate the shift in marine phosphate concentrations. Moreover, future high-resolution proxy studies may shed light on short-term changes in P weathering since the Miocene, and the importance of P weathering relative to silicate weathering during earlier global cooling episodes in Earth’s history.

Methods

Age model

A total of seven micro-drilled Fe-Mn oxide samples were taken from the upper 50 mm of crust MID03, with a 2 mm layer thickness sample (Supplementary Fig. 2a). Subsample processing and Be isotopic measurements were carried out at the Xi’an Accelerator Mass Spectrometry Center (Xi’an-AMS Center), Institute of Earth Environment, Chinese Academy of Sciences. Seven subsamples were pretreated according to the protocol proposed by ref. 80 and updated by ref. 81. 10Be/9Be ratios were calibrated against ICN-01-5-1 material, using a certified ratio of 2.709 × 10[−11 82,. All the results (including correction for procedural blanks) are presented in Supplementary Table 2, and the resulting age models are shown in Supplementary Fig. 2. The subsamples display a continuous decline in 10Be/9Be with increasing depth in the crust from 0 to 37 mm (Supplementary Fig. 2b). Average growth rates and ages were calculated based on linear interpolation between those data points (Supplementary Fig. 2c), with growth rates between 2.5 and 8.3 mm/Ma being higher than the typical range of 1–5 mm/Ma for central Pacific crusts83.

Element concentration measurements

The Fe-Mn MID03 crust was cut vertically at 1-2 mm intervals, and a total of 25 subsamples were scraped continuously from top to bottom for major and minor element analyses. Major elements were determined using a Thermo-Fisher IRIS II Intrepid XSP ICP-OES. Additional trace elements, including rare earth elements and yttrium (REY), were analyzed by inductively coupled plasma mass spectrometry using a Perkin-Elmer ELAN 9000 ICP-MS at Qingdao Sparta Analysis & Test Co., Ltd. in China. The dissolution procedure was as follows: 0.05 g of sample powder was digested in a sealed Teflon beaker with 0.5 ml HF, 0.5 ml HNO3, and 1.5 ml HCl on a hot plate (120 °C) for ~ 12 h. Following heating to dryness, 1 ml HNO3 and 1 ml H2O were added, and the beaker was sealed and heated on a hot plate (150 °C) for ~ 12 h to dissolve the residue. For internal calibration, we use the reference material GBW07315, GBW07316, BCR-2, BHVO-2, GBW07295, GBW07296, NOD-P-1, and NOD-A-1. Reference materials and parallel sample analysis are used for quality control during analysis. For major elements, the error range of the reference materials between the test value and the reference value is within 5%, and the accuracy can reach 95%. The relative standard deviation (RSD) of repeated analysis of this batch of samples was less than 2.5%. Estimated changes in oceanic P concentration are calculated from the P/Fe data of this study using a linear scale factor (Fig. 4), with modern oceanic P concentration calibrated at 3 μM84.

Nd isotope measurements

Nd isotopes were analyzed on the Nu Plasma HR MC-ICP-MS with a DSN-100 desolvation nebulizing system, using a three-cycle dynamic procedure. Instrument bias and mass fractionation were corrected by normalization of raw ratios to 146Nd/144Nd = 0.7219. Eleven measurements of JNdi-1 yield an average 143Nd/144Nd = 0.512113 ± 9 (n = 11), which is consistent with consensus value: 0.512115 ± 7. Lab standard Ames Nd Metal was used as a rotary instrument drift monitor, which corresponded with JNDi-1. 17 analyses yield an average of 143Nd/144Nd = 0.511966 ± 16; this value was used as a calibration reference for instrument drift, which is usually less than 15 ppm.

Mineral magnetic measurements

The subsamples were dried in an oven at a temperature below 40 °C and then placed in 2 × 2 × 2 cm3 nonmagnetic plastic cubes to conduct magnetic measurements. Low- and high-frequency mass-specific magnetic susceptibility (χlf and χhf) were measured using a multifunction Kappabridge (MFK-FA) with frequencies of 976 Hz and 15,616 Hz, respectively. An alternating field (AF) was applied to each sample using a D-2000 AF demagnetizer, using a peak field of 100 mT and a direct current (DC) bias field of 0.05 mT, to obtain the anhysteretic remanent magnetization (ARM), which is regarded as representative of the stable single-domain (SD) ferrimagnetic content85. The susceptibility of anhysteretic remanent magnetization (χARM) was obtained by dividing the ARM by the DC bias (0.05 mT). Saturation isothermal remanent magnetization (SIRM) was imparted to the Z-axis for each sample in a DC field of 1 T using an IM-10-30 Impulse Magnetizer and was measured using a JR-6A Spinner Magnetometer (AGICO). This SIRM, named SIRM1T, was subsequently demagnetized with backfields of -100 and -300 mT, and the corresponding remanences (IRM-100mT and IRM-300mT, respectively were measured). Hysteresis loops and first-order reversal curves (FORCs) were measured up to maximum applied fields of 1.0 T for samples using a vibrating sample magnetometer (VSM, LakeShore 8600). The saturation magnetization (Ms) and coercivity (Bc) were obtained after high-field slope correction. The remanence coercivity (Bcr) was obtained by direct current demagnetization of SIRM1T back to -1.0 T. For every sample, a total of 100 FORCs measured with an averaging time of 500 ms and a maximum field of 400 mT were averaged. The FORC diagrams were processed using the software ‘FORCinel'86 with VARIFORC smoothing87. Principle component analysis (FORC-PCA) was also performed using ‘FORCinel’ following the protocols of ref. 88. Magnetic measurements were performed at the Center for Marine Magnetism (CM2), Southern University of Science and Technology in China.

Model simulations

The global biogeochemical model utilized in this study is derived from66,67 and encompasses the cycling of carbon, oxygen, phosphorus, iron, and sulfur through five oceanic boxes and one atmospheric box (Supplementary Tables 5–8). For an in-depth understanding of the model, please refer to refs. 66,67.

For the simulations aimed at assessing the isolated impact of a prescribed rise in phosphorus (P) weathering on oceanic P concentration and climate, we introduced an enhancement factor for P weathering (Supplementary Fig. 8), which was tuned to mach the observed changes in oceanic P concentrations. It is important to note that in this particular simulation, we maintained the formulas for silicate and carbonate weathering without alteration. The formulas for silicate weathering and carbonate weathering were consistent with those outlined in ref. 66, relying on functions of temperature and runoff.

In the model run designed to simulate the combined effects of P and silicate weathering, we applied the same enhancement factor (Supplementary Fig. 8) to P, silicate, and carbonate weathering. This approach ensures a comprehensive examination of the integrated impact of these processes on oceanic P concentration and climate dynamics.