Emergence of the Southeast Asian islands as a driver for Neogene cooling

Significance

The Southeast Asian islands are a modern-day hotspot of CO₂ consumption via silicate weathering. Since ∼15 million years ago, these islands have been increasing in size at the same time that Earth’s climate has been cooling. Here, we test the hypothesis that this global cooling could have been driven by tectonic emergence of the Southeast Asian islands. Using a compilation of paleoshorelines, in conjunction with a coupled silicate weathering and climate model, we find that this emergence is associated with a large decrease in pCO₂. Without these changes in tropical island paleogeography, there would not have been large Northern Hemisphere ice sheets as a defining feature of Earth’s climate over the past 3 million years.

Abstract

Steep topography, a tropical climate, and mafic lithologies contribute to efficient chemical weathering and carbon sequestration in the Southeast Asian islands. Ongoing arc–continent collision between the Sunda-Banda arc system and Australia has increased the area of subaerially exposed land in the region since the mid-Miocene. Concurrently, Earth’s climate has cooled since the Miocene Climatic Optimum, leading to growth of the Antarctic ice sheet and the onset of Northern Hemisphere glaciation. We seek to evaluate the hypothesis that the emergence of the Southeast Asian islands played a significant role in driving this cooling trend through increasing global weatherability. To do so, we have compiled paleoshoreline data and incorporated them into GEOCLIM, which couples a global climate model to a silicate weathering model with spatially resolved lithology. We find that without the increase in area of the Southeast Asian islands over the Neogene, atmospheric pCO₂ would have been significantly higher than preindustrial values, remaining above the levels necessary for initiating Northern Hemisphere ice sheets.

The Southeast Asian islands (SEAIs) have an outsized contribution to modern chemical weathering fluxes relative to its area. The confluence of steep topography, a warm and wet tropical climate, and the presence of mafic lithologies result in high fluxes of Ca and Mg cations in the dissolved load and associated CO₂ consumption (1–4). There has been a significant increase in the area of subaerially exposed land within the region since the mid-Miocene associated with ongoing arc–continent collision between Australia and the Sunda-Banda arc system (5–7). Concurrently, after the Miocene Climatic Optimum, a cooling trend began circa (ca.) 15 million years ago (Ma) and accelerated over the past 4 million years (m.y.) leading to the development of Northern Hemisphere ice sheets (8, 9). Many hypotheses have been proposed to explain this cooling trend, including changes in ocean/atmosphere circulation (5, 10, 11), a decrease in volcanic degassing (12), or uplift in the Himalaya (13, 14). Here, we seek to evaluate the hypothesis that emergence of the SEAIs was a significant factor in driving long-term climatic cooling over the Neogene.

Over geologic time scales, CO₂ enters Earth’s ocean–atmosphere system primarily via volcanism and metamorphic degassing and leaves primarily through the chemical weathering of silicate rocks and through organic carbon burial (15). Chemical weathering delivers alkalinity and cations to the ocean, which drives carbon sequestration through carbonate precipitation. Steady-state $\mathit{\text{p}}{\text{CO}}_2$ is set at the $\mathit{\text{p}}{\text{CO}}_2$ level at which ${\text{CO}}_2$ sinks are equal to sources. As ${\text{CO}}_2$ sinks increase and $\mathit{\text{p}}{\text{CO}}_2$ falls, temperature decreases, and the hydrological cycle is weakened, causing the efficiency of the silicate weathering sink to decrease until a new steady state is achieved at lower $\mathit{\text{p}}{\text{CO}}_2$ (16).

Topography, climate, and lithology all affect chemical weathering. High-relief regions generally lack extensive regolith development and, thus, tend to have reaction-limited weathering regimes that are more prone to adjust when climate changes (17–19). High physical erosion rates contribute to high chemical weathering fluxes in these high-relief regions (20). In warm and wet regions, mineral-dissolution kinetics are faster, leading to enhanced chemical weathering (18, 21). Mafic rocks have higher Ca and Mg concentrations and dissolution rates than felsic rocks and, thus, have the potential to more efficiently sequester carbon through silicate weathering (22). These factors have led to the proposal that arc–continent collisions, which create steep landscapes that include mafic lithologies, within the tropical rain belt have been important in enhancing global weatherability, lowering atmospheric $\mathit{\text{p}}{\text{CO}}_2$, and initiating glacial climate over the past 520 m.y. (7, 23, 24), and perhaps in the Neoproterozoic as well (25).

Quantitatively estimating the magnitude of decrease in steady-state $\mathit{\text{p}}{\text{CO}}_2$ associated with the emergence of a region with a high carbon-sequestration potential, such as the SEAIs, requires constraints on changing tectonic context and accounting of associated feedbacks. As this region emerges, the total global silicate weathering flux will transiently exceed the volcanic degassing flux, causing $\mathit{\text{p}}{\text{CO}}_2$ to initially decline until a new steady state is established, where the total magnitude of the ${\text{CO}}_2$ sinks is the same as before the change. However, the sensitivity of the silicate weathering flux in any particular location to this change in $\mathit{\text{p}}{\text{CO}}_2$ is variable and dependent on the specific topography, climate, and lithology at that location. Furthermore, how regional climate responds to this change in $\mathit{\text{p}}{\text{CO}}_2$ is itself spatially variable. Therefore, the magnitude of $\mathit{\text{p}}{\text{CO}}_2$ change that is required to balance the total global silicate weathering flux with the volcanic degassing flux will depend on the specific spatial distribution of topography, climate, and lithology at the time of emergence. As a result, any attempt to meaningfully estimate the decrease in steady-state $\mathit{\text{p}}{\text{CO}}_2$ associated with the emergence of the SEAIs must model spatially resolved climatology and silicate weathering fluxes in tandem and account for the spatial distribution of the factors that affect these interconnected systems.

GEOCLIM Model

To estimate the decrease in steady-state $\mathit{\text{p}}{\text{CO}}_2$ associated with the increase of subaerially exposed land area in the SEAIs, we use the global spatially resolved GEOCLIM model (26). GEOCLIM estimates changes in steady-state $\mathit{\text{p}}{\text{CO}}_2$ associated with coupled changes in erosion, chemical weathering, and climatology by linking a silicate weathering model to climate model runs at multiple $\mathit{\text{p}}{\text{CO}}_2$ levels.

Silicate Weathering Component.

The silicate weathering component of GEOCLIM calculates ${\text{CO}}_2$ consumption resulting from silicate weathering for subaerially exposed land. We assume that Ca and Mg are the only cations that consume ${\text{CO}}_2$ over geologic time scales, such that each mole of Ca or Mg that is dissolved by silicate weathering consumes 1 mol of ${\text{CO}}_2$. While reverse weathering is another potential sink for Ca or Mg (27), its parameterization is unclear, and it has been interpreted to be a relatively minor flux in the Cenozoic (28), and we do not include it in our model. In previous versions of the model, silicate weathering was a function of temperature and runoff only, and all bedrock was assigned identical chemical compositions (26). More recent versions of GEOCLIM implement regolith development and soil shielding (Fig. 1), which introduces a dependence on erosion rate (and, therefore, topographic slope) (29). While this introduction of regolith development into GEOCLIM is important for assessing the impact of tropical arc–continent collisions on $\mathit{\text{p}}{\text{CO}}_2$, the relatively high Ca+Mg concentration in arc rocks relative to other lithologies must also be considered.

Fig. 1.

A schematic representation of the silicate weathering component of GEOCLIM in a single profile at steady state. A rock particle leaves the unweathered bedrock with production rate $P_r$ and transits vertically through a regolith of height $h$. Regolith production and physical erosion ($E_p$) are equal at steady state. As a particle transits upwards, some fraction of the primary phases ($x$) are chemically weathered ($W$), with the flux of dissolved Ca+Mg being $W$ multiplied by the concentration of Ca+Mg in unweathered bedrock ($X_{CaMg}$). Details of the formulation for the silicate weathering component of GEOCLIM can be found in Materials and Methods.

We therefore implement variable bedrock Ca+Mg concentration into GEOCLIM (SI Appendix). The spatial distribution of lithologies is sourced from the Global Lithologic Map (GLiM) (30) and is represented by six categories: metamorphic, felsic, intermediate, mafic, carbonate, and siliciclastic sediment. Each land pixel is assigned these lithologic categories at a resolution of $0.1^{°}\times 0.1^{°}$. The Ca+Mg concentrations of felsic, intermediate, and mafic lithologies are assigned based on the mean of data of these lithologic categories compiled in EarthChem (www.earthchem.org/portal). Given that GLiM does not distinguish ultramafic lithologies, such rocks are grouped with mafic rocks. As a result, the Ca+Mg concentration is likely an underestimate in regions of obducted ophiolites, such that the estimated effect of these regions on changing steady-state $\mathit{\text{p}}{\text{CO}}_2$ could be conservative (31). The weathering of carbonate does not contribute to long-term ${\text{CO}}_2$ consumption, and its Ca+Mg concentration is ignored. The Ca+Mg concentrations of metamorphic and siliciclastic sediment lithologies are more difficult to define, since their chemical composition is strongly dependent on protolith composition and, in the case of siliciclastic sediment, the degree of previous chemical depletion. We explore a range of feasible Ca+Mg concentrations for metamorphic rocks and siliciclastic sediment during calibration of the silicate-weathering component of GEOCLIM.

Calibration.

The values of four parameters within the silicate-weathering component that modify the dependence of silicate weathering on temperature, runoff, erosion, and regolith thickness are poorly constrained. Rather than prescribing single values, we selected multiple values for each of these four parameters along with the Ca+Mg concentration of metamorphic and siliciclastic lithologies from within reasonable ranges (SI Appendix, Table S2). We then permuted all possible combinations of these values for the six parameters, leading to 93,600 unique parameter combinations (i.e., permutations). For each combination, we computed spatially resolved long-term ${\text{CO}}_2$ consumption associated with Ca+Mg fluxes using present-day runoff, temperature, and slope. We sum-computed ${\text{CO}}_2$ consumption over watersheds for which data-constrained estimates are available (1, 32) and then calculated the coefficient of determination ($r^2$) between computed and measured ${\text{CO}}_2$ consumption in each of these watersheds. After eliminating parameter combinations that resulted in low $r^2$, 573 parameter combinations remained (SI Appendix, Fig. S3). The resulting global ${\text{CO}}_2$ consumption of these filtered model runs all overlap with independently derived estimates of the global ${\text{CO}}_2$ degassing flux (33), as they should for a steady-state long-term carbon cycle (SI Appendix, Fig. S3).

Climate Model Component.

Having calibrated the silicate weathering component of GEOCLIM, we used it to estimate the decrease in steady-state $\mathit{\text{p}}{\text{CO}}_2$ associated with emergence of the SEAIs. For the climate-model component, we used temperature and runoff from a subset of the Geophysical Fluid Dynamics Laboratory (GFDL) CM2.0 experiments (34) (SI Appendix). These experiments are well-suited for this analysis because all non-${\text{CO}}_2$ forcings are held constant at values representative of preindustrial conditions, allowing the effect of changing $\mathit{\text{p}}{\text{CO}}_2$ on climatology to be isolated. Furthermore, the experiments were run long enough for the final system to approximate steady state.

Paleoshorelines

To determine the position of paleoshorelines in the SEAIs over the past 15 m.y., we used terrestrial and marine sedimentary deposits (Fig. 2; SI Appendix). The paleoshoreline data indicate that the Sunda-Banda Arc and New Guinea are primarily responsible for the increase in area since 15 Ma. Exhumation of the modern Sunda-Banda Arc is the result of ongoing arc–continent collision with the Australian Plate (37). Most of Sumatra and Java along with the nonvolcanic islands of the Outer Banda Arc were elevated above sea level after 5 Ma (38). In New Guinea, emergence in the mid-Miocene is associated with collision between the Melanesian Arc and Australia’s distal margin (39), which drove exhumation of the Irian-Marum-April Ophiolite Belt. Exhumation accelerated over the past 4 m.y. in the New Guinea Central Range, due to slab breakoff and buoyant uplift, and in eastern New Guinea due to jamming of the north-dipping subduction zone (39). We also include changes in areas of presently submerged continental shelves, such as the Sunda Shelf, that were previously exposed (SI Appendix, Fig. S7). These tectonic drivers and others throughout the region led to progressive emergence over the past 15 m.y. that accelerated following 5 Ma (Fig. 2B). This trend mirrors broad cooling over the Neogene that resulted in the initiation of Northern Hemisphere ice sheets (Fig. 2C).

Fig. 2.

The emergence of the SEAIs (also referred to as the Maritime Continent in the climate-science literature) from the mid-Miocene to present. Past shorelines at 5, 10, and 15 Ma are shown in A, with associated land area summarized in B. A significant increase in area over the past 5 m.y. is coincident with cooling and the onset of Northern Hemisphere glaciation, as reflected in the benthic oxygen isotope record (35) shown in C.

We used GEOCLIM to estimate $\mathit{\text{p}}{\text{CO}}_2$ associated with the reconstructed subaerial extent of the SEAIs at ca. 15, 10, and 5 Ma (“paleo-SEAIs” scenarios; Fig. 3). Because we used a climate model forced with modern geography, the position of the tectonic blocks remains fixed. Although there has been motion of these tectonic blocks since 15 Ma, they have remained within tropical latitudes, such that this fixed scenario is a good approximation of the paleogeography (SI Appendix, Fig. S10). We also tested an end-member scenario, in which all islands associated with arc–continent collision in the region were removed (“removed SEAIs” scenario; Fig. 3).

Fig. 3.

Steady-state $\mathit{\text{p}}{\text{CO}}_2$ estimates from GEOCLIM for the various scenarios discussed in the text. For each of the seven scenarios, each point represents an estimate from one of the 573 unique parameter combinations that most closely matched estimates of present-day ${\text{CO}}_2$ consumption in 80 watersheds around the world (SI Appendix). The box encloses the middle 50% of the $\mathit{\text{p}}{\text{CO}}_2$ estimates (i.e., the interquartile range), and the notch represents the median with its 95% CI. The whiskers extend to the 2.5th and 97.5th percentile values. Glaciation thresholds (36) are shown on the x axis.

$\mathit{\text{p}}{\text{CO}}_2$ Estimates

Using the 573 unique parameter combinations, the paleo-SEAIs scenarios resulted in 526 to 678 parts per million (ppm) for 15 Ma, 457 to 516 ppm for 10 Ma, and 391 to 434 ppm $\mathit{\text{p}}{\text{CO}}_2$ for 5 Ma (Fig. 3). These results indicate a progressive decrease in $\mathit{\text{p}}{\text{CO}}_2$ over the Neogene associated with the emergence of the SEAIs and suggest that, without this emergence, preindustrial $\mathit{\text{p}}{\text{CO}}_2$ would have been $\sim$526 to 678 ppm. These paleo-SEAIs scenarios do not account for Neogene changes outside of the SEAIs (e.g., changes in ocean/atmosphere circulation, volcanic degassing, and weathering fluxes elsewhere on Earth, discussed in Alternative Mechanisms for Neogene Cooling). Therefore, these results are not estimating $\mathit{\text{p}}{\text{CO}}_2$ at 15 Ma, but, rather, are quantifying $\mathit{\text{p}}{\text{CO}}_2$ change associated with emergence of the SEAIs.

Proxy-based estimates of the magnitude and trajectory of $\mathit{\text{p}}{\text{CO}}_2$ change from the Miocene to the Pliocene are variable between techniques and associated assumptions underlying their interpretation (SI Appendix, Fig. S11). The $\mathit{\text{p}}{\text{CO}}_2$ values from the 5 Ma paleo-SEAIs scenario overlap with many proxy-based estimates (40), as well as values that emerge from approaches that assimilate climate and ice-sheet model output with benthic ${\delta }^{18}\text{O}$ data (41, 42). The modeled $\mathit{\text{p}}{\text{CO}}_2$ values for 15 Ma resemble the higher end of proxy-based $\mathit{\text{p}}{\text{CO}}_2$ estimates for the early to mid-Miocene, indicating that the increase in subaerially exposed land area and tectonic topography of the SEAIs is sufficient to explain long-term cooling of Earth’s climate over the Neogene. The $\mathit{\text{p}}{\text{CO}}_2$ threshold for Antarctic glaciation is estimated to be $\sim$750 ppm, with that for Northern Hemisphere glaciation being significantly lower at $\sim$280 ppm (36). These modeled values of decreasing $\mathit{\text{p}}{\text{CO}}_2$ associated with emergence of the SEAIs are, therefore, consistent with the record of Neogene climate with Miocene ice sheets on Antarctica (43), followed by Northern Hemisphere ice sheets developing in the Pliocene (44), as $\mathit{\text{p}}{\text{CO}}_2$ subsequently decreased.

The results of our paleo-SEAIs scenarios highlight the importance of the combination of topography, runoff, and lithology in setting Earth’s climate state. To independently explore the effect of the modern-day surface exposure of lower-relief basaltic lavas on steady-state $\mathit{\text{p}}{\text{CO}}_2$ (45), we replaced mafic volcanics associated with the Deccan Traps, Ethiopian Traps, and Columbia River Basalts with the Ca+Mg concentration of bulk continental crust in GEOCLIM (Fig. 3). The resulting $\mathit{\text{p}}{\text{CO}}_2$ is $\sim$300 to 500 ppm, indicating that the presence of mafic rocks in these igneous provinces affects steady-state $\mathit{\text{p}}{\text{CO}}_2$, as has been suggested to be important for Paleogene cooling (45). However, the higher 526- to 678-ppm values for the 15 Ma paleo-SEAIs scenario illustrate that higher relief and a wet tropical climate significantly increase the efficiency of ${\text{CO}}_2$ consumption, especially when paired with high Ca+Mg lithologies. As such, arc–continent collisions in the tropics are likely more important for driving long-term changes in $\mathit{\text{p}}{\text{CO}}_2$ than the eruption of flood basalts (7, 46).

Previous work has estimated that the decrease in $\mathit{\text{p}}{\text{CO}}_2$ since 5 Ma associated with the emergence of the SEAIs and enhanced silicate weathering is $\sim$19 ppm (5), in which case their emergence would be a relatively minor contributor to Neogene cooling. This 19-ppm estimate was obtained by using an equation that assumes a direct linear relationship between mean global temperature and changes in weathering-rate-weighted land area, scaled by a factor that is intended to account for the influence of both runoff and temperature. $\mathit{\text{p}}{\text{CO}}_2$ was then estimated from the calculated temperature by using a simple energy-balance equation. However, the relationship between mean global temperature (or $\mathit{\text{p}}{\text{CO}}_2$) and weathering-rate-weighted land area is not linear. Furthermore, this simple linear relationship ignores spatial variability in topography and climatology and only crudely accounts for spatial variability in lithology. In fact, the 19-ppm estimate is closer in magnitude to the decrease in $\mathit{\text{p}}{\text{CO}}_2$ that we estimate if mafic volcanics associated with the Deccan Traps (a relatively flat area outside of the warm and wet tropics) are replaced with the Ca+Mg concentration of bulk continental crust (22 to 70 ppm; Fig. 3). The significant difference in steady-state $\mathit{\text{p}}{\text{CO}}_2$ estimated between the removed Deccan Traps scenario and the paleo-SEAIs scenarios (Fig. 3) demonstrates that considering changes in the spatial distribution of lithologies alone is not adequate for estimating changes in steady-state $\mathit{\text{p}}{\text{CO}}_2$. Instead, spatially varying topography and climatology significantly modulates silicate weathering rates and must be accounted for when estimating $\mathit{\text{p}}{\text{CO}}_2$ change associated with paleogeographic change.

An important caveat for these estimates of $\mathit{\text{p}}{\text{CO}}_2$ is that our modeling is determining the climatology in the GFDL CM2.0 model at which steady state is achieved—a climatology that has an associated $\mathit{\text{p}}{\text{CO}}_2$ value in the model. However, climate models are variable in their response to changes in $\mathit{\text{p}}{\text{CO}}_2$. One way to summarize this variability is through the equilibrium climate-sensitivity value—the steady-state change in global mean surface air temperature associated with a doubling of $\mathit{\text{p}}{\text{CO}}_2$. A range of 1.5 to 4.5 °C per $\mathit{\text{p}}{\text{CO}}_2$ doubling was proposed in the landmark Charney report (47), and this range was considered to be the credible interval (>66% likelihood) in the last Intergovernmental Panel on Climate Change report (48). Integrating constraints both from understanding of climate feedback processes and the climate record, a recent comprehensive review estimates the 66% probability range of climate sensitivity to be 2.6 to 3.9 °C per $\mathit{\text{p}}{\text{CO}}_2$ doubling, with a 5 to 95% range of 2.3 to 4.7 °C per $\mathit{\text{p}}{\text{CO}}_2$ doubling (49). The equilibrium climate sensitivity associated with the CM2.0 climate models is 2.9 °C per $\mathit{\text{p}}{\text{CO}}_2$ doubling, which falls within these ranges, although these ranges remain broad. An alternative way to consider the results from our analysis would be that an estimate of 572 ppm (2$\times$ preindustrial $\mathit{\text{p}}{\text{CO}}_2$) for the 15 Ma paleo-SEAIs scenario is implying that Earth would be $\sim 2.9\hspace{0.17em}°\text{C}$ warmer. If Earth’s climate sensitivity is at the higher end of the probable range and higher than in the CM2.0 model, as it is in some climate models, this same amount of Neogene cooling resulting from the emergence of the SEAIs could have been driven by a smaller change in $\mathit{\text{p}}{\text{CO}}_2$.

Alternative Mechanisms for Neogene Cooling

Ocean/Atmosphere Circulation.

Some hypotheses to explain ice-sheet growth over the Neogene invoke changes in ocean/atmosphere circulation, including further climatic isolation of Antarctica due to strengthening of the circumpolar current (11); increased atmospheric moisture in the Northern Hemisphere due to intensified thermohaline circulation following Panama Isthmus emergence (10); and cooling of North America resulting from a strengthened Walker Circulation associated with emergence of the SEAIs (5). Such changes in ocean/atmosphere circulation are likely to modulate $\mathit{\text{p}}{\text{CO}}_2$ thresholds for glacial initiation and ice-sheet growth (36). However, the prolonged time scale of the cooling trend since 15 Ma (Fig. 2C) is most readily attributable to decreasing $\mathit{\text{p}}{\text{CO}}_2$ associated with evolving geological sources and sinks of carbon, modulated by the silicate weathering feedback (16, 50–53).

Volcanic Degassing.

A decrease in volcanic degassing (12) has also been proposed as a driver for Neogene cooling. However, proxy-based estimates of the evolution of volcanic degassing fluxes throughout the Neogene are inconsistent with each other, such that not even the sign of the change in volcanic degassing over the past $\sim$15 m.y. is without ambiguity (26). For example, it has both been estimated that the volcanic degassing flux was $\sim$25% lower (54) and $\sim$10% higher (55) at 15 Ma relative to the present day.

Our model framework provides an opportunity to estimate the decrease in volcanic degassing flux necessary to achieve the same change in $\mathit{\text{p}}{\text{CO}}_2$ predicted for the increase in global weatherability associated with the emergence of the SEAIs over the past 15 m.y. If we use the parameter combination that had the highest $r^2$ between computed and measured ${\text{CO}}_2$ consumption in watersheds around the world during calibration (Calibration and SI Appendix, Fig. S4), GEOCLIM estimates a preindustrial volcanic degassing flux of 4.1$\times {\mathrm{10}}^{12}$ mol/y to balance the silicate weathering flux at 286 ppm $\mathit{\text{p}}{\text{CO}}_2$. If we then assume that this volcanic degassing flux did not change over the past 15 m.y., then GEOCLIM estimates that the increase in global weatherability associated with the emergence of the SEAIs led to a change in $\mathit{\text{p}}{\text{CO}}_2$ of $\sim$280 ppm (“increase in weatherability only” scenario in Fig. 4). If we, instead, assume that global weatherability did not change over the past 15 m.y., then we estimate that the volcanic degassing flux needs to have been $\sim$13% greater at 15 Ma relative to the preindustrial to drive the same $\sim$280 ppm change in $\mathit{\text{p}}{\text{CO}}_2$ (“decrease in degassing only” scenario in Fig. 4). This $\sim$13% value is higher than 10%, the highest current estimate for the volcanic degassing flux at 15 Ma relative to the present day (55).

Fig. 4.

Weatherability curves for the modern and “paleo-SEAIs” scenarios shown in Fig. 3. Lower expands the lower $\mathit{\text{p}}{\text{CO}}_2$ range (x axis) of Upper. Details on how these curves were generated are described in Materials and Methods. Each of the four curves represent a different tectonic boundary condition (i.e., the reconstructed paleoshorelines of the SEAIs; Fig. 2A) and, therefore, a different global weatherability. The curves show the resulting $\mathit{\text{p}}{\text{CO}}_2$ for a given volcanic degassing flux such that the input flux is balanced by the silicate weathering output flux. Point B represents the preindustrial, in which $\mathit{\text{p}}{\text{CO}}_2$ is 286 ppm. The arrow from point ${\mathrm{A}}_1$ to B represents the “increase in weatherability only” scenario, in which global weatherability increases as the SEAIs emerge, but the volcanic degassing flux does not change over the past 15 m.y. In this scenario, the $\mathit{\text{p}}{\text{CO}}_2$ decreases from the value dictated by the 15 Ma paleo-SEAIs weatherability curve (568 ppm). The arrow from point ${\mathrm{A}}_2$ to B instead represents the “decrease in degassing only” scenario, in which global weatherability remains the same as the preindustrial, but the same change in $\mathit{\text{p}}{\text{CO}}_2$ as the “increase in weatherability only” scenario is achieved by decreasing the volcanic degassing flux from a value $\sim$13% greater than the preindustrial. The arrow from Point ${\mathrm{A}}_3$ to B represents the “increase in weatherability and degassing” scenario, in which a change in $\mathit{\text{p}}{\text{CO}}_2$ from 400 to 286 ppm is achieved by increasing both global weatherability from our 15-Ma tectonic boundary condition and the volcanic degassing flux from a value $\sim$7% smaller than the preindustrial flux.

However, changes in the volcanic degassing flux would have modulated changes in $\mathit{\text{p}}{\text{CO}}_2$ associated with changes in global weatherability. For example, some proxy-based approaches, as well as some model-data assimilation approaches, estimated that mid-Miocene $\mathit{\text{p}}{\text{CO}}_2$ was lower than 568 ppm (SI Appendix, Fig. S11). Take a scenario in which $\mathit{\text{p}}{\text{CO}}_2$ was 400 ppm at 15 Ma. If we assume that the $\mathit{\text{p}}{\text{CO}}_2$ decrease to the preindustrial value of 286 ppm was driven by the increase in global weatherability associated with emergence of the SEAIs in conjunction with an increase in volcanic degassing, which counteracts cooling by increasing the flux of ${\text{CO}}_2$ to the atmosphere (“increase in weatherability and degassing” scenario in Fig. 4), the volcanic degassing flux would have had to have been $\sim$7% smaller than the preindustrial. More robust constraints on $\mathit{\text{p}}{\text{CO}}_2$ (SI Appendix, Fig. S11) and/or volcanic degassing rates over the past 20 m.y. are needed to constrain which of the “increase in weatherability only” or “increase in weatherability and degassing” scenarios (Fig. 4) is more representative of the mechanisms driving Neogene cooling.

Himalayan Uplift.

Marine ⁸⁷Sr/⁸⁶Sr has overall been increasing since ca. 35 Ma (56). The traditional explanation for this trend is that it reflects increased weathering of radiogenic (i.e., high ⁸⁷Sr/⁸⁶Sr) silicate rocks (13, 57). Associated with this explanation is the proposal that increasing weathering of radiogenic silicate rocks in the Himalayas was the primary driver of Neogene cooling (58). It could be argued that increasing marine ⁸⁷Sr/⁸⁶Sr is inconsistent with the hypothesis that increasing weathering of juvenile (i.e., low ⁸⁷Sr/⁸⁶Sr) silicate rocks in the SEAIs was an important driver of Neogene cooling. However, the globally averaged ratio of silicate weathering fluxes from radiogenic cratonic rocks versus juvenile arc lithologies can be at least partially decoupled from marine ⁸⁷Sr/⁸⁶Sr via the regional weathering of isotopically unique lithologies. For example, in addition to highly radiogenic granites and gneisses (57), unusually radiogenic carbonates are abundant in Himalayan strata, and it is estimated that $\sim$75% of Sr coming from the Himalayas can be attributed to carbonate rather than silicate weathering (59–61). As such, there are challenges in interpreting the marine ⁸⁷Sr/⁸⁶Sr record as a direct proxy for silicate weathering fluxes. Nevertheless, steadily increasing marine ⁸⁷Sr/⁸⁶Sr is interrupted ca. 16 Ma, when the slope of the ⁸⁷Sr/⁸⁶Sr curve decreases (56). This decrease in slope has been attributed to coincident exhumation of relatively low ⁸⁷Sr/⁸⁶Sr Outer Lesser Himalaya carbonates (62, 63), but could also be at least partially driven by the emergence of low ⁸⁷Sr/⁸⁶Sr lithologies in the SEAIs during arc–continent collision. Increasing seawater Mg/Ca since ca. 15 Ma (64) is consistent with an increasing proportion of the global silicate weathering flux being derived from mafic and ultramafic sources.

Himalayan uplift would have affected geological carbon sinks, either via increased weathering of silicate rocks (58) or enhanced burial of organic matter in the Bengal Fan (14). Increased weathering of the emerging SEAIs would have occurred in tandem with such changes in the Himalayas, such that the effects of these paleogeographic changes on geochemical proxy records, like marine ⁸⁷Sr/⁸⁶Sr, become difficult to disentangle. In addition, given the large uncertainty associated with changes in regional climatology across Asia due to Himalayan orogeny, developing quantitative estimates of the evolution of global silicate weathering fluxes associated with Himalayan orogeny remains a major challenge.

The Geologic Carbon Cycle

If geological carbon sources remain approximately constant, global alkalinity delivery from silicate weathering needs to be approximately constant as well to keep the long-term carbon cycle in steady state (16). Enhanced silicate weathering in a region such as the SEAIs is compensated by a decrease in silicate weathering elsewhere. Global alkalinity delivery from silicate weathering does not change, but occurs more efficiently and, thereby, at lower $\mathit{\text{p}}{\text{CO}}_2$. Given that carbonate weathering is disconnected from the long-term carbon-cycle mass balance, changes in carbonate accumulation through time (65) could be driven by changes in carbonate weathering.

The long-term carbon-cycle mass balance can be perturbed via mechanisms that are disconnected from changes in volcanic degassing and silicate weathering rates. For example, sulfide oxidation coupled to carbonate dissolution could act as a source of ${\text{CO}}_2$ on million year time scales (66). Similarly, the weathering of sedimentary organic matter could serve as a source of ${\text{CO}}_2$ (67). On the other hand, enhanced burial of organic matter enabled by higher sediment and nutrient delivery could be an important sink of ${\text{CO}}_2$, as has been suggested in the Bengal Fan (14) and Taiwan (68). The fluxes of ${\text{CO}}_2$ represented by these processes are not accounted for in our model framework and could have been affected by emergence of the SEAIs and/or Himalayan orogeny. $\mathit{\text{p}}{\text{CO}}_2$ changes that result from these processes would be superimposed on $\mathit{\text{p}}{\text{CO}}_2$ changes associated with evolving silicate weathering fluxes. However, our coupled weathering-climate model indicates that the $\mathit{\text{p}}{\text{CO}}_2$ change associated with increased global weatherability driven by emergence of the SEAIs is sufficient to explain the majority of Neogene cooling (Fig. 3). Without this emergence, $\mathit{\text{p}}{\text{CO}}_2$ would have remained above the level necessary for the growth of Northern Hemisphere ice sheets.

Conclusions

Coupled geological constraints and modeling experiments demonstrate that the SEAIs have been a growing hot spot for carbon sequestration due to silicate weathering from the Miocene to present. Changes in volcanic degassing and paleogeography elsewhere on Earth, particularly in the Himalayas and Central America, would have also affected geological carbon sources and sinks. Yet, not only does the history of emergence of the SEAIs coincide with Neogene cooling and the onset of Northern Hemisphere glaciation, but our coupled weathering-climate model also indicates that the associated steady-state $\mathit{\text{p}}{\text{CO}}_2$ change is sufficient to explain much of this cooling. These results highlight that the Earth’s climate state is particularly sensitive to changes in tropical geography.

Materials and Methods

GEOCLIM Silicate Weathering Component.

The silicate weathering component of the GEOCLIM model has been modified from the published version (20). The component implements the model of Gabet and Mudd (2009) (17) for the development of a chemically weathered profile. We refer to this chemically weathered profile as regolith, where the base of the regolith is unweathered bedrock. In the model of Gabet and Mudd (2009), material enters the regolith and leaves either as a solute through chemical weathering of the material during its travel from the bedrock toward the surface, or as a physically weathered particle once it reaches the top. We used the DynSoil implementation of the Gabet and Mudd (2009) model, which integrates chemical weathering within the regolith (18). The transient time-varying version of this regolith model is described by three equations:

$$\frac{dh}{dt}=P_r-E_p,$$ [1]

$$\frac{\partial x}{\partial t}=-P_r\frac{\partial x}{\partial z}-K{\tau }^{\sigma }x,$$ [2]

$$\frac{\partial \tau }{\partial t}=-P_r\frac{\partial \tau }{\partial z}+1.$$ [3]

Eq. 1 is a statement of material conservation, where $h$ is the total height of the regolith (m), $t$ is the model time (y), $P_r$ is the regolith production rate (m/y), and $E_p$ is the physical erosion rate (m/y). Eq. 2 describes how the residual fraction of weatherable phases ($x$, unitless) changes as a function of time ($t$, y) and depth ($z$, m). $K{\tau }^{\sigma }$ is the dissolution rate constant, which depends on the local climate (captured by K, y^−1−σ) and the time that a given rock particle has spent in the regolith ($\tau$, y) to some power $\sigma$ (unitless) which implements a time dependence. Eq. 3 describes how the time that a given rock particle has spent in the regolith changes as time in the model progresses.

The net weathering rate in the regolith column (W, m/y) can then be calculated with:

$$W={\int }_0^{h}K{\tau }^{\sigma }x\hspace{0.17em}dz.$$ [4]

The regolith production rate can be expressed as the product of the optimal production rate ($P_0$) and a soil-production function ($f\left(h\right)$):

$$P_r=P_0\hspace{0.17em}f\left(h\right),$$ [5]

$$P_0=k_{rp}\hspace{0.17em}q\hspace{0.17em}{e}^{-\frac{E_a}{R}\left(\frac{1}{T}-\frac{1}{T_0}\right)},$$ [6]

$$f\left(h\right)=e^{\frac{-h}{d_0}}.$$ [7]

$P_0$ is the “optimal” regolith production rate (m/y), which is defined to be the regolith production rate when there is no overlying regolith. In Eq. 6, where $k_{rp}$ is a proportionality constant (unitless), $q$ is the runoff (m/y), $E_a$ is the activation energy (J/K/mol), $R$ is the ideal gas constant (J/mol), $T$ is the temperature (K), and $T_0$ is the reference temperature (K), we parameterize the optimal regolith production rate (69). $f\left(h\right)$ is the soil-production function (unitless), which describes how regolith production decreases as the thickness of the regolith increases. It takes an exponential form, as is commonly applied in the literature (17). In Eq. 7, $d_0$ is a reference to regolith thickness (m) (70).

Our implementation of the erosion rate is parameterized based on runoff and slope ($s$; m/m):

$$E_p=k_e\hspace{0.17em}{q}^m\hspace{0.17em}{s}^n.$$ [8]

$k_e$ is a proportionality constant [(m/y)^1−m] and $m$ and $n$ are adjustable exponents that are kept as 0.5 and 1 (29). This formulation is directly inspired by the stream power law (71). This formulation and these exponent values are supported by compilations, but variability in the proportionality constant is difficult to capture at a global scale (72).

The $K$ in the dissolution rate constant in Eq. 2 describes the dependence of the chemical weathering on climate:

$$K=k_d\left(1-e^{-k_{w}q}\right)e^{-\frac{E_a}{R}\left(\frac{1}{T}-\frac{1}{T_0}\right)}.$$ [9]

Eq. 9 is an empirical simplification of mineral-dissolution rates derived from kinetic theory and laboratory experiments (18), where $k_d$ is a proportionality constant that modifies the dependence of dissolution rate on runoff and temperature (y^−1−σ), and $k_w$ is a proportionality constant that modifies the dependence of dissolution rate on runoff (y/m).

In this study, we are interested in obtaining the steady-state solution rather than the transient time-varying solution. The steady-state solution for DynSoil can be calculated analytically by setting the time derivatives equal to zero, resulting in the following set of equations:

$$h=max\left(0,\hspace{0.17em}{d}_0\log \left(\frac{P_0}{E_p}\right)\right),$$ [10]

$$x\left(z\right)=\exp \left(-\frac{K}{\sigma +1}{\left(\frac{z}{E_p}\right)}^{\sigma +1}\right),$$ [11]

$$W=E_p\left(1-x\left(h\right)\right)=E_p\left(1-x_s\right).$$ [12]

$x\left(z\right)$ is the abundance profile of primary phases inside the regolith, varying with height upward from the base of the regolith, as shown in Fig. 1. Setting $z$ equal to the regolith thickness ($h$) gives $x_s$, which is the proportion of primary phases remaining at the top of the regolith column.

Weatherability Curves.

To create the 15-Ma paleo-SEAIs curve shown in Fig. 4, we used the reconstructed paleoshorelines of the SEAIs at 15 Ma (Fig. 2A). We then selected the parameter combination that had the highest $r^2$ between computed and measured ${\text{CO}}_2$ consumption in watersheds around the world during calibration (Calibration and SI Appendix, Fig. S4) and fixed $\mathit{\text{p}}{\text{CO}}_2$ at the three $\mathit{\text{p}}{\text{CO}}_2$ levels at which the GFDL CM2.0 climate model experiments were computed (SI Appendix). We then ran GEOCLIM at each of these $\mathit{\text{p}}{\text{CO}}_2$ levels until steady state was achieved (i.e., until the volcanic degassing flux was equal to the silicate weathering flux). We then repeated this process for the 10- and 5-Ma paleo-SEAIs paleoshorelines and the present-day shorelines to generate the three other weatherability curves. Each estimated $\mathit{\text{p}}{\text{CO}}_2$ in Fig. 3 is the result of underlying weatherability curves that change with the different chemical weathering parameters.

SI Appendix.

A detailed description of the implementation of lithology into the silicate weathering component of GEOCLIM, the calibration of the silicate weathering component of GEOCLIM, the GFDL CM2.0 climate model, and the paleoshoreline reconstructions can be found in SI Appendix.

Data Availability.

The code for the GEOCLIM model used in this study can be found at GitHub, https://github.com/piermafrost/GEOCLIM-dynsoil-steady-state/releases/tag/v1.0. The code that generated the inputs and analyzed the output of the GEOCLIM model, as well as these input and output files, can be found at GitHub, https://github.com/Swanson-Hysell-Group/2020_Southeast_Asian_Islands, and Zenodo, https://doi.org/10.5281/zenodo.4021653.