Over multi-million-year timescales, climates during the Mesozoic and Cenozoic were affected by a gradual increase in solar forcing and, in general, a gradual decrease in CO₂ forcing Fig. 1. A number of theories have been advanced to explain changes in atmospheric CO₂, and thus climate, over multi-million-year timescales. The gradual breakup of Pangaea and the associated rifting likely resulted in increased volcanic degassing fluxes . The Late Cretaceous and the Cenozoic saw a period of mountain building associated with the latter stages of the breakup of Pangaea. The Laramide orogeny began approximately 70 Ma , and the development of the northern Andes and the Himalayas occurred during the early Cenozoic . The development of new mountain ranges, particularly the Himalayas, has been implicated in the reduction of CO₂ concentrations during the Cenozoic by increasing silicate weathering rates and thus increasing CO₂ drawdown . Changes in weathering fluxes have been linked to a number of climate shifts in the Earth's past, such as the global cooling which occurred during the Carboniferous e.g..

Changes in palaeogeography can have substantial effects . For example, the presence of mountain ranges may affect monsoon circulation and thus precipitation, or the presence of large continental interiors isolated from moisture can lead to large arid deserts. As such, the potential silicate weathering flux, or “weatherability”, of the continents is likely to represent a key climate forcing during the late Mesozoic–early Cenozoic and forms the focus of this study.

At steady state, silicate weathering fluxes are equal to emissions from volcanic degassing e.g.. However, a distinction should be drawn between weathering (here, the chemical breakdown of silicate rocks by water and carbonic acid) and “weatherability”. Weatherability corresponds to the susceptibility of the land surfaces to be weathered, under a given climate (temperature and runoff) . Weatherability can be affected by the presence of land plants impacting relief and erosion rates (through the exposure of fresh material to weathering), by changes in the lithology, or by the palaeogeographic configuration . The role of land plants is difficult to constrain, and studies have produced contrasting results e.g., so vegetation is often excluded or held constant within simulations of weathering.

Global palaeolithologies are poorly constrained and become even less so further back into the geological record. Similarly, factors such as topography and uplift rates, which affect erosion rates e.g., are also poorly constrained into geological time. In contrast, broad-scale climate parameters, such as temperature and precipitation patterns, are comparatively well constrained through proxy data and GCM studies. Climate states are in turn strongly controlled by the positioning of the continents, which can affect oceanic and atmospheric circulation patterns e.g.. The positioning of the continents also affects weatherability by determining the land area in intense weathering environments, such as the tropics e.g..

In this study, we focus on the impact of the palaeogeographic setting on the weatherability, assuming that all the other factors remain constant over the simulated period (this is certainly not the case, but our objective is to quantify exclusively the role of palaeogeography).

For an idealised case where degassing is constant , the global weathering flux will equilibrate the volcanic degassing of CO₂, and the global CO₂ consumption by silicate rock weathering will stay constant. However, the evolving continental configuration modulates weatherability, allowing CO₂ to fluctuate .

In contrast to palaeo-degassing rates, palaeogeographies (within the last 150 million years) are relatively well constrained. Thus, there is scope to investigate the impact of changing weatherability on long-term CO₂ concentrations. Early studies using zero-dimensional global models, such as GEOCARB, used estimations of global mean annual temperature (MAT) and runoff values, but no geographical constraints, to estimate global weathering fluxes and steady-state CO₂ concentrations in geological time . Further studies refined the GEOCARB approach by including palaeogeographical reconstructions, both conceptual (e.g. “ringworld” configurations) and realistic, which demonstrated that continental configurations had significant impact on steady-state CO₂ concentrations e.g. and that palaeogeography and climate interact to determine steady-state CO₂ concentrations . Further model developments provided spatial patterns in climate-forcing data, demonstrating that regional changes in weatherability can have global climate impacts . More recently, such models have shown that the positioning of the continents also affects weatherability by determining the land area in intense weathering environments, such as the tropics e.g..

The impact of changing weatherability as a result of changes in palaeogeography during the Phanerozoic was previously investigated by using GEOCLIM with climate inputs from the GCM, FOAM, which found that the formation of supercontinents resulted in a continental configuration favourable for high atmospheric CO₂ concentrations (10–25 × PAL), as the converged continental configuration resulted in arid continental interiors and reduced weatherability. Conversely, dispersed continental configurations were more favourable to lower atmospheric CO₂ concentrations (1–8 × PAL) by favouring higher continental runoff and weatherability. Furthermore, found that continents or even smaller landmasses crossing warm humid climate belts can result in significant drawdown of atmospheric CO₂. For example, during the Late Triassic, the gradual northward drift of Pangaea brought larger areas of landmass into more humid zones. In response, CO₂ concentrations fell sharply over ∼ 20 Myr from 19 × PAL to 3 × PAL .

had a relatively low temporal resolution of 22 simulations over the last ∼ 520 Myr (approximately 1 simulation per 20–30 Myr) to produce a simulated CO₂ record and, from that, glean insights on the effects of changing weatherability (through palaeogeographical changes) on long-term CO₂ concentrations and thus global climate. In contrast, this study will use 19 simulations over the Cretaceous–Eocene period (145–36 Ma) in addition to better-constrained and higher-resolution (factor of 3.65 increase) palaeogeography and climate inputs to produce our own simulated CO₂ record, assuming a constant degassing rate. The higher temporal resolution will allow the investigation of the impacts of higher-temporal-resolution changes (∼ 5 Myr) in palaeogeography on long-term CO₂ and weatherability, while the higher spatial resolution will better constrain regional variability in weatherability relative to . The use of climate inputs produced by the more complex HadCM3L in this study (relative to the FOAM inputs used by ) also represents an improvement. These model-based improvements should provide a more accurate picture of how global climate responds to changes in weatherability and should better constrain the role of chemical weathering feedbacks on global climate through geological time.

GEOCLIM is a coupled global climate and biogeochemical model initially developed in the early 2000s and has been used to investigate interactions between climate and geochemistry in deep time settings on geological timescales . GEOCLIM uses temperature and runoff inputs from climate models at specified CO₂ concentrations (e.g. 280, 560 ppm) and models the climate conditions through interpolation between as atmospheric CO₂ concentrations evolve within GEOCLIM simulations. Should CO₂ concentrations in GEOCLIM fall outside of the specified range, it will be unable to update climate conditions and thus will use climate conditions associated with the nearest available CO₂ concentration. GEOCLIM initially used CLIMBER inputs in early studies , but more recent studies have used FOAM inputs .

COMBINE, a model within GEOCLIM, handles biogeochemical processes for both land and ocean environments. Oceans are divided into nine “boxes”: two high-latitude oceans (> 60° N/S) (each separated into a photic layer and a deep-ocean layer); a low- to mid-latitude ocean (60° S–60° N) divided into photic, thermocline, and deep-ocean layers; and an epicontinental sea also divided into photic and deep layers. A final 10th box represents the atmosphere. A range of biogeochemical processes are modelled within the ocean layers. As this study focuses on terrestrial processes, we will omit a detailed description of the ocean processes, but full descriptions can be found in and .

Chemical weathering in GEOCLIM occurs in terrestrial grid cells, and the total sum value of silicate weathering from each terrestrial grid cell, along with a prescribed volcanic degassing value, is used to calculate changes in atmospheric CO₂ at each time step. Ocean carbon and alkalinity are kept balanced at steady state. Chemical weathering calculations for silicate rocks are based on equations from and . 1Fsil(t)=ksil×αj(t)×ρj(t)×Ae-Ea/RT

Equation (1) calculates the granitic weathering flux in moles of C per year at a time step (t) in a given grid cell, Fsil(t); ksil is the silicate weathering constant; αj(t) is the area of a grid cell (10⁶ km²); and ρj(t) is the runoff value of that grid cell (cm yr⁻¹). The final term (dimensionless) is an Arrhenius equation for granite weathering, representing the granitic dependence on dissolution based on air temperature (T) and the universal gas constant (R), which uses the activation energy for granite (Ea) defined by and the air temperature in a given grid cell, Tj. 2Fbas(t)=kbas×αj(t)×ρj(t)×Ae-Ea/RT

A similar calculation exists for basalt (Eq. 2), based on values in , and for carbonate rocks. Both calculations were derived from and calibrated on weathering reaction processes within granite and basaltic watersheds, described in detail in and , respectively. We omit the carbonate equation here, as this study focuses on variations in CO₂ drawdown via changes in silicate weathering fluxes. GEOCLIM assumes that each grid cell has an equal area of granitic, basaltic, and carbonate rocks. While this is obviously a significant simplification of the real-world geology of the time, palaeolithologies are poorly constrained. In the absence of reliable global palaeolithologies from the studied time period, we assume an even distribution of lithologies for simplicity and consistency with . A newer version of GEOCLIM, recently developed, includes an erosion dependency on chemical weathering rates based on equations in . As a key part of this study relates to comparing our new simulations to previous results with GEOCLIM , this version of the model was not used in this study for reasons of consistency. The most recent version of GEOCLIM, GEOCLIM7, is fully detailed in .

HadCM3L is part of the HadCM3 “family” of coupled GCMs originally developed by the UK Met Office . The HadCM3 family of models has been in use for over 15 years, with various modifications to the configuration of the original model, HadCM3, being made in that time. HadCM3L is a version of HadCM3 using a “low-resolution” ocean, where ocean and atmosphere share the same 96 × 73 (3.75° × 2.5°) resolution . Although described as “low-resolution”, 96 × 73 is nonetheless 3.65 times higher resolution than FOAM. However, like FOAM, HadCM3L is used for long-term simulations where higher-resolution models would be too time-intensive to be practical. To that end, HadCM3L has been used extensively in pre-Quaternary climate studies, where long-term climate simulations are required e.g.. This study uses climate inputs from 19 HadCM3L simulations of the Cretaceous–Eocene period at intervals of 3–13 Ma, the same simulations used by . These simulations were run with CO₂ concentrations at 560 and 1120 ppm, which are considered to be a reasonable representation of potential CO₂ concentrations during the Cretaceous–Eocene, which are subject to considerable uncertainty . To match the range of climate conditions at the 11 concentrations used for the FOAM climate inputs (160–1400 ppm) in , the HadCM3L climate inputs were extrapolated from the two CO₂ concentrations (560–1120 ppm) to the same 11 concentrations (Appendix A). This process was devised and evaluated using a sensitivity study in and was found to be robust, although this is largely irrelevant, as the modelled CO₂ concentrations produced remained very close to or within the 560–1120 ppm range. The simulations in (used for this study) slightly differ from those in in that they have a spinup time of 10 422 years, as opposed to 1422 years in . The extended spinup time was to enable the simulations to more closely approach an equilibrium state and to make use of the fully dynamic mode within the coupled vegetation model. Runoff calculations are affected by the vegetation model, with both canopy interactions and soil infiltration considered within HadCM3L. A full description of the runoff parameterisation can be found in . Estimating uncertainties in modelled runoff is challenging, but the higher resolution of HadCM3L relative to FOAM is likely to provide a better parameterisation of features significant to runoff, such as relief, even when accounting for uncertainties in reconstructing palaeo-relief.

The palaeogeographic reconstructions used in this study were developed by Getech for . The reconstructions were produced using methods developed by and include data from well-constrained geological databases, such as The Paleogeographic Atlas Project. These palaeogeographies were originally constructed at 0.5° × 0.5° resolution – topographies, bathymetries, and land–sea masks were generated from these high-resolution palaeogeographies at 3.75° × 2.5° resolution for HadCM3L . Additionally, HadCM3L is coupled to the Top-down Representation of Interactive Foliage and Flora Including Dynamics (TRIFFID), which is a dynamic global vegetation model . The two modules are coupled via the MOSES 2.1 land surface scheme . The TRIFFID model calculates the fraction of five plant functional types (broadleaf trees, needleleaf trees, C₃ grasses, C₄ grasses, and shrubs) in each grid cell. Although these are modern plant types, studies have suggested that models such as TRIFFID can nonetheless perform acceptably in providing vegetation feedback signals back to at least 250 million years . Estimating uncertainties introduced by the vegetation model is challenging; however, indicate that vegetation models adequately reproduce the spatial distributions of major biogeographical areas, such as desert, temperate, and polar regions.

The degassing rate for the palaeogeography simulations was set to a constant value of 1 × 10¹³ mol C yr⁻¹, similar to previous GEOCLIM studies . A value of 1 × 10¹³ mol C yr⁻¹ is within the range of estimates (∼ 1.2–2 times modern) for degassing rates during the Cretaceous–Eocene period . These simulations were also run using variable degassing rates ranging from a low value of 7.82 × 10¹² mol C yr⁻¹ in the latest Eocene to a high of 1.23 × 10¹³ mol C yr⁻¹ in the Early Cretaceous (Appendix Fig. A1). This range is based on values calculated by using reconstructed subduction zone arc lengths.

We conducted two sets of experiments. We ran one simulation per time slice, fixing the initial atmospheric CO₂ to 2.85 times the pre-industrial value of 280 ppm (∼ 800 ppm). This CO₂ concentration is intermediate between the 560 and 1120 ppm reconstructions. The first set of experiments allows us to isolate the effect of the palaeogeography on the weatherability of the continental surfaces, all other factors being fixed. The second set of experiments was conducted by allowing GEOCLIM to run until atmospheric CO₂ reached steady state to provide an estimate of atmospheric CO₂ under the aforementioned palaeogeographical and climate conditions. Atmospheric CO₂ will adapt in response to weathering until global silicate weathering fluxes compensate for solid Earth degassing.

During the Eocene, within the GEOCLIM simulations, both mean and total global runoff from the fixed CO₂ runs fall, yet silicate weathering fluxes rise, an inverse of the pattern seen during the Cretaceous, where runoff rates and weathering fluxes show identical patterns (Fig. 2). During the Early Cretaceous, silicate weathering fluxes are strongly controlled by non-tropical runoff (> 30° N/S), showing an almost identical pattern until the mid-Cretaceous (Fig. 4). This pattern weakens somewhat around 86 Ma but strengthens by the end of the Cretaceous. Into the Cenozoic, the pattern weakens again and silicate weathering fluxes become anti-phased to non-tropical runoff. In contrast, silicate weathering fluxes in the Eocene appear to be more strongly controlled by runoff rates in the tropics (Fig. 4). Similarly, tropical runoff changes in the Cretaceous do not appear to have a significant impact on silicate weathering fluxes, with the exception of a brief period during the Late Cretaceous (86 Ma). There is a noticeable weakening of the influence of non-tropical runoff on weathering fluxes from the Cretaceous to the Cenozoic, although, due to the small number of simulations in this period (n=6), confidence in this relationship is low.

The shift from non-tropical- to tropical-runoff-controlled weathering appears to represent a regime change in the long-term climate pattern seen during the modelled period. Theoretically, a continental configuration with a greater land area in the low latitudes, where precipitation is generally highest, would favour lower global CO₂ concentrations by increasing silicate weathering rates . However, changes in continental positioning and orogeny will alter ocean–atmospheric circulation and evaporation patterns, which would then alter climatic variables, especially runoff . Furthermore, runoff patterns will also be affected by whether continents are in a dispersed or converged (i.e. supercontinental) configuration.

The shift in weathering controls seen in the GEOCLIM simulations is the result of a change in the effects of palaeogeography on weathering fluxes. During the Cretaceous, the models suggest that increased weathering occurred as a result of the climate becoming more humid due to the breakup of Pangaea. The breakup of Pangaea promoted higher weathering fluxes through evaporation by increased ocean area and more favourable moisture transport to the continental interiors. In contrast, during the Cenozoic, both total land areas and tropical land areas increased. During the same period, modelled total global runoff falls, but tropical runoff increases, resulting in both a greater weatherable area and a more intense weathering environment in the tropics. As such, the shift in weathering controls in the Cenozoic from non-tropical to tropical runoff is the result of the continents moving into a configuration which promotes higher weatherability, while, during the Cretaceous, changes in weathering fluxes were largely caused by climate changes induced by the breakup of Pangaea which increased runoff in the previously arid continental interiors.

While the GEOCLIM simulations in this study represent an improvement over previous studies due to their higher spatiotemporal resolution and better-constrained palaeogeography, they are nonetheless an obvious simplification relative to both full GCM studies (which provide more realistic modelling of climate change processes, and feedbacks in particular, rather than the linear interpolation used here) and real-world settings. Because the primary aim of this study is to assess the impact of palaeogeographic changes on potential global “weatherability”, simplifications such as uniform palaeolithologies were used in the absence of well-constrained data for such fields. Still, it is naturally of interest to compare the modelled CO₂ concentrations produced in this study with those derived from previous modelling studies and proxy data. Such a comparison may provide an indication of the potential impact of changing weatherability on long-term CO₂ concentrations through the Cretaceous–Eocene period.

Figure 1 presents the CO₂ output produced in this study (henceforth referred to as the GEOCLIM output) plotted against three CO₂ records derived from proxy data and the 145–34 Ma portion of the record from . The Anagnostou and Hoenisch records only cover the Cenozoic and the latest Cretaceous (in the case of the Hoenisch record). Through the Cretaceous–Eocene period, there are three periods in which the trends in the GEOCLIM output agree with the proxy records: the Early Cretaceous, the Late Cretaceous, and the mid-Eocene. Both the GEOCLIM output and the record from indicate relatively high (> 800 ppm) CO₂ concentrations in the earliest Cretaceous, followed by a general decreasing trend until 125 Ma. The high CO₂ concentrations in the Early Cretaceous are likely the result of the converged continental configuration, which is unfavourable for high weathering fluxes. Both records indicate a rise in CO₂ just after the earliest Cretaceous (138 Ma, although only one GEOCLIM simulation is available between 140 and 135 Ma). At this time, the continental interiors become more arid, as reflected by global runoff falling from ∼ 82 × 10³ to ∼ 64 × 10³ cm yr⁻¹ over a period of 5 Myr (Figs. 2 and 3). The fall in global runoff leads to a reduction in silicate weathering fluxes, resulting in a CO₂ increase.

A rise in CO₂ during the Late Cretaceous (∼ 85 Ma) occurs in both the GEOCLIM output and the record from . During the same period, a drop in global weathering fluxes (especially in the tropics) and total tropical runoff occurs (Fig. 4). In both records, the rise in CO₂ is relatively brief (10–20 Myr) and of approximately the same magnitude (∼ 200 ppm). There is notable disagreement between the Foster and Hoenisch records at the end of the Cretaceous, with showing a significant drop in CO₂ concentrations and showing a more modest decrease down to around 700 ppm. There is significant uncertainty in the Hoenisch proxy record at this time, and the GEOCLIM output is similar to the mean value suggested by the Hoenisch record.

The proxy records in the early Eocene show different CO₂ concentrations, ranging from around 600 ppm in the Foster record to 1800 ppm in the Anagnostou record. The GEOCLIM output shows relatively steady CO₂ concentrations at this time, while the Anagnostou and Hoenisch records show much higher CO₂ concentrations and a sharper decrease. A sharp increase in CO₂ concentrations is shown in the Hoesnich record, but this occurs between simulations in the GEOCLIM output; thus our climate simulations may not be capturing this high-CO₂ period. All records and the GEOCLIM output indicate a drop in CO₂ of varying magnitudes towards the mid-Eocene, with the GEOCLIM output showing the smallest drop in CO₂ (< 100 ppm) and the records from and indicating a much larger drop of around 600 ppm and 400 ppm, respectively. A fall in CO₂ from the early to mid-Eocene is often attributed to increased silicate weathering e.g.. Indeed, our modelled Eocene weathering fluxes are generally high and with a general rising trend towards the end of the Eocene, which coincides with falling CO₂ concentrations (Figs. 1 and 2). Zonal weathering plots indicate a wider area of high weathering fluxes in the tropics (focused on eastern Asia and India) at this time relative to other time periods (e.g. mid-Cretaceous) in this study (30° N/S) (Appendix Fig. A4).

The GEOCLIM CO₂ concentrations presented in this study reflect the changes in global “weatherability” through the Cretaceous–Eocene period, indicating a general trend towards increased global weatherability during that time, punctuated by a decrease at ∼ 80 Ma (Fig. ). Although comparison with CO₂ records from proxy data suggests that not all changes in long-term CO₂ can be ascribed to changes in global weatherability, the three records indicate that changes in global weatherability have sufficiently powerful impacts on geologically short timescales (∼ 10 Myr) to be expressed in long-term CO₂ records (i.e. hundreds of ppmv), further supported by the three periods of agreement between the records. These changes in weatherability are primarily associated with the breakup of Pangaea during the Early Cretaceous and with changes in tropical land areas and runoff rates in the mid-Cretaceous and in the Eocene, respectively. Based on these findings, the implications for palaeoclimates are 2-fold.

Firstly, this study reinforces the conclusions of previous studies that continental configuration is a strong control on long-term CO₂ concentrations by influencing runoff . Supercontinental configurations typically reduce global weatherability and favour high CO₂ concentrations by limiting global runoff, while dispersed configurations typically increase global weatherability and favour lower CO₂ concentrations as the dispersed configuration favours moisture transport to continental interiors.

Secondly, this study demonstrates the potential for relatively localised changes in weatherability to have global impacts. Such local changes may result in a temporary decoupling of globally averaged runoff rates from globally averaged weathering rates. During the Eocene, non-tropical runoff totals fall substantially, but tropical runoff increases significantly, likely associated with an increase in tropical land areas at the same time. Similarly, during the mid-Cretaceous, tropical runoff totals remain stable, but non-tropical rainfall increases. The increase in non-tropical rainfall is sufficient to increase silicate weathering fluxes and lower atmospheric CO₂ concentrations by 200 ppm over 13 Myr.