A major challenge in paleoclimatology is disagreement between data and models for periods of warm climate. Data generally indicate equable conditions and reduced latitudinal temperature gradients, while models generally produce colder conditions and steeper latitudinal gradients except when using very high CO2. Here we show congruence between temperature indicators and climate model output for the cool greenhouse interval of the latest Cretaceous (Maastrichtian) using a global database of terrestrial and marine indicators and fully coupled simulations with the Community Climate System Model version 3. In these simulations we explore potential roles of greenhouse gases and properties of pre-anthropogenic liquid clouds in creating warm conditions. Our model simulations successfully reproduce warm polar temperatures and the latitudinal temperature gradient without overheating the tropics. Best fits for mean annual temperature are simulations that use 6× preindustrial levels of atmospheric CO2, or 2× preindustrial levels of atmospheric CO2 and liquid cloud properties that may reflect pre-anthropogenic levels of cloud condensation nuclei. The Siberian interior is problematic, but this may relate to reconstructed elevation and the presence of lakes. Data and models together indicate tropical sea-surface temperatures ∼5 °C above modern, an equator-to-pole temperature difference of 25–30 °C, and a mid-latitudinal temperature gradient of ∼0.4 °C per 1° latitude, similar to the Eocene. Modified liquid cloud properties allow successful simulation of Maastrichtian climate at the relatively low levels of atmospheric CO2 indicated by proxies and carbon cycle modeling. This supports the suggestion that altered properties of liquid clouds may be an important mechanism of warming during past greenhouse intervals.


The discrepancy between geologic indicators of temperature and climate models has been a major challenge in the study of warm climates. Ever since the work of Barron and Washington (1982), climate models have generally produced colder temperatures and less equable climates than those inferred from the geologic record, especially for high latitudes and continental interiors (Royer et al., 2012). While early discrepancies could be attributed to problems with both models and data, the problem has continued as models have undergone major improvement and geologic data have become more quantitative (Spicer et al., 2008; Upchurch et al., 1999). Warm climates of the early Eocene have been successfully simulated by some models (e.g., Huber and Caballero, 2011; Lunt et al., 2012), but usually require atmospheric CO2 well above that allowed by proxies and carbon cycle modeling (16× present atmospheric level, PAL). Simulating past warm climates under geologically realistic levels of greenhouse gases is important for predicting the future, because future levels of greenhouse gases are projected to reach Late Cretaceous and Eocene levels.

For the middle and Late Cretaceous, modeling studies have explored the warming potential of different factors. These include elevated atmospheric CO2 (Bice et al., 2006), polar forests (Otto-Bliesner and Upchurch, 1997; Zhou et al., 2012), altered properties of liquid clouds (Kump and Pollard, 2008), and increased levels of non-CO2 trace gases (Beerling et al., 2011).

We explore the sensitivity of modeled climate for the Maastrichtian to increased carbon dioxide and methane, and properties of liquid clouds, and compare model results to a global quantitative database of temperature indicators. We use the Maastrichtian because it has been the subject of numerous model and data studies at the global scale (Amiot et al., 2004; Craggs et al., 2012; Horrell, 1991; Hunter et al., 2008; Markwick, 2007; Puceat et al., 2007; Upchurch et al., 1999), and because understanding Maastrichtian climate is critical to understanding climatic change during the Cretaceous–Paleogene mass extinction.


We greatly expanded and updated a previous global database (Upchurch et al., 1999), focusing on quantitative indicators of terrestrial and marine temperature (Fig. 1; see the GSA Data Repository1). Terrestrial paleotemperatures were based on plant macrofossils, palynomorphs, and δ18O of vertebrate tooth enamel and pedogenic carbonate, and include mean annual temperature (MAT), warm month mean, and cold month mean. Paleobotanical temperatures were based on multiple calibrations and methods, including leaf margin analysis, Climate Leaf Analysis Multivariate Program (CLAMP), digital leaf physiognomy, and our own recalibration of leaf margin temperatures in Wolfe and Upchurch (1987). Sea-surface temperatures (SSTs) were estimated using TEX86 and δ18O of exceptionally preserved carbonate (especially glassy planktonic foraminifera and aragonitic shells of mollusks that lived above the thermocline) and fish tooth enamel, the latter updated with two new calibrations (see the Data Repository). Our isotopic data include previously published global data sets (Amiot et al., 2004; Puceat et al., 2007; Zakharov et al., 2006). We omitted planktonic foraminifera with average preservation because of potential problems with recrystallization in the presence of colder bottom waters (Pearson et al., 2001). Our geochemical data provide a warm estimate of tropical SST and an upper limit on modeled temperatures for the tropics. Our temperature database comprises 66 records (many with alternative calibrations) that span an interval of ∼6–7 m.y., making it comparable to global databases used for the early Eocene (Huber and Caballero, 2011; Lunt et al., 2012). It represents average Maastrichtian climate and does not account for documented temperature fluctuations (e.g., Alsenz et al., 2013).

We ran fully coupled simulations of Maastrichtian climate using a lower resolution version of the Community Climate System Model version 3 (CCSM3; Kiehl and Shields, 2013). The model employs a T31 (truncation; 3.75° × 3.75°) horizontal resolution for the atmosphere and a nominal 3° ocean model with 25 levels. This lower resolution version of CCSM3 yields a realistic simulation of the present climate. The model runs presented here used prescribed levels of CO2 and CH4 and fixed plant life forms.

Paleoelevation and land and sea distributions for model simulations were identical to those used for the late Maastrichtian (Upchurch et al., 1999) except (1) in North America and Africa, the seaways were widened to better simulate ocean circulation; and (2) in South America, the narrow seaways east of the cordillera became land when converted to T31 resolution (Fig. 1). Vegetation was based on the most recent global reconstruction (Upchurch et al., 1999) adapted to the CCSM3, which prescribed polar forests for the Northern and Southern Hemispheres. The vegetation model used modern physiology for plant functional types. Not included are effects of elevated CO2 on plant physiology, and evolution of angiosperm physiology during the Cretaceous (Feild et al., 2011).

For the atmospheric model, CO2 was set at 560 ppm and 1680 ppm (2× and 6× PAL) to encompass the full range of values proposed for the Maastrichtian (Breecker et al., 2010; Nordt et al., 2003; Royer, 2010). Atmospheric CH4 was set at 760 ppb (PAL) and 2000 ppb, the lower range of the modeled level for the Late Cretaceous (Beerling et al., 2009, 2011). For liquid clouds, droplet size and droplet concentration were set at (1) the standard value for CCSM3, and (2) values characteristic of pristine regions well removed from anthropogenic cloud condensation nuclei (explained in detail in Kiehl and Shields, 2013). These warm pole clouds tested the proposed role of altered liquid clouds in Cretaceous global warming (Kump and Pollard, 2008), and determined the extent to which model-data discrepancies might be due to cloud parameterizations tuned to a modern dirty atmosphere with abundant anthropogenic cloud condensation nuclei.

We ran six Maastrichtian simulations (Table 1) and a modern-day reference simulation. We evaluated model performance by comparing individual temperature values from the database with values from the corresponding model grid for each simulation, and calculating statistics, including the average of model minus data and standard deviation. For geologic data, average and extreme values were plotted, using alternative calibrations and estimates of error, to provide a range of uncertainty. Outside the tropics, the latitudinal gradient of MAT was estimated for model and proxy data using simple linear regression for model grid cells containing one or more proxy data points.


Tropical SST is estimated as 28–32 °C, ∼5 °C above the modern value, based on exceptionally preserved carbonate in planktonic foraminifera and shelf invertebrates, fish tooth enamel, and TEXH86. Mean annual temperature for the Arctic basin ranges from 2.6 °C (δ18O of dinosaur tooth enamel) and ∼6 °C (paleobotany) to 17 °C (TEXH86). The difference between the highest estimated equatorial temperature and the mid-range for the Arctic basin is ∼26 °C. The latitudinal gradient of MAT from 30°N to 80°N is 0.40 °C per 1° latitude if the Arctic TEX86 data point is removed (R2 = 0.66) and 0.38 °C per 1° latitude if it is included (R2 = 0.62). This total evidence gradient is similar to gradients based on δ18O of terrestrial vertebrate and marine fish tooth enamel (Amiot et al., 2004; Puceat et al., 2007), soil carbonate (Nordt et al., 2003), and leaf margin analysis (see the Data Repository). It is steeper than gradients based on δ18O of planktonic foraminifera with typical preservation (∼0.1 °C) and CLAMP (0.2 °C) (Puceat et al., 2007; Spicer and Herman, 2010).

Model simulations produce an average global mean surface temperature of 20–28 °C, 7–15 °C warmer than in the modern simulation (Table 1). The mean surface temperature for the tropics (20°S to 20°N) is 28–35 °C, 3–10 °C higher than in the modern simulation. The average simulated temperature difference between the tropics and the Arctic is 23–28 °C. Polar regions have above-freezing MAT in all simulations except 2× CO2. Warm pole clouds increase global mean surface temperature by 4 °C, with 3–4 °C warming at the equator and >6–8 °C warming at high latitudes, similar to the Eocene (Kiehl and Shields, 2013). Winter sea ice is minimal in all simulations except for 2× CO2, where it forms over most of the Arctic Ocean and in oceans adjacent to Antarctica.

Comparison of model grid temperatures with temperature proxies indicates that the 6× CO2 simulations with modern clouds and the 2× CO2 warm pole (WP) simulation agree best with data (Fig. 2; see the Data Repository). The 6× CO2 simulation with preindustrial levels of methane has an average of model minus data closest to 0, while the 2× CO2 WP simulation has the lowest standard deviation. In the tropics, the 2× CO2 simulation produces temperatures at the low end allowed by data, whereas the 2× CO2 WP simulation and the 6× CO2 simulations with anthropogenic clouds produce mean annual surface temperatures at the high end. The 6× CO2 WP simulations produce tropical temperatures that are well above proxies. For middle and high latitudes, the 2× CO2 WP simulation and the 6× CO2 simulations agree well with data, but the 2× CO2 simulation is too cold and produces too steep a latitudinal temperature gradient (Fig. 2; see the Data Repository).

The 2× CO2 WP simulation is the only successful simulation that uses CO2 levels in line with current baseline estimates for the Maastrichtian, i.e., ∼2–3× PAL and <1000 ppm (Breecker et al., 2010; Pagani et al., 2014, their figure 10; Royer, 2010, his figure 1). It also fully predicts the distribution of crocodilians and palm macrofossils, in contrast to all earlier model simulations of the Maastrichtian (see the Data Repository).

The Vilyui Basin of Siberia is a region of conflict between climate model output and paleobotanical data for continental interiors. The megaflora used for model data comparisons (Spicer et al., 2008) is Cenomanian (Spicer and Herman, 2010), so its temperature value must be corrected for differences in age and paleolatitude (see the Data Repository). The adjusted value is 11.3 °C, 6.3 °C warmer than model value for the 2× CO2 WP simulation, and slightly less than the model value for the 6× CO2 WP simulation (Table 1). Paleogeographic reconstructions for the Late Cretaceous place the Vilyui Basin at a low elevation, with mountains to the east and south. Low-resolution models smooth topography, so we ran the 6× CO2 WP simulation without mountains in Siberia to determine sensitivity of model surface temperature to elevation. Removing mountains increases MAT for the Vilyui Basin by 1.8 °C. High seasonality remains an issue, but this may reflect failure of model simulations to include the narrow seaway or large lake system that was present in the Vilyui Basin during the Late Cretaceous (Kazmin et al., 1998).


Our simulations are the first to fully simulate Maastrichtian high-latitude warmth and avoid creating tropical temperatures well above those documented by proxies. This results from warmer estimates of tropical SSTs by TEX86 and stable isotopes in exceptionally preserved carbonate and phosphate, and estimates of moderate latitudinal temperature gradients through combined stable isotopic and paleobotanical data. This also results from greater physical realism in model simulations. It was suggested that increased levels of latitudinal heat transport by oceans caused high-latitude warming during the Cretaceous (Herman and Spicer, 1996). However, in CCSM3, the atmosphere and ocean together transport no more heat than for the modern day, implying that increased radiation is the dominant mechanism.

In our model, warming results from increased absorbed long-wave radiation from elevated levels of greenhouse gases, or lower levels of absorbed long-wave radiation coupled with increased absorbed short-wave radiation due to liquid cloud properties. In the absence of strong evidence for high levels of atmospheric CO2 during the latest Cretaceous and Eocene, altered cloud properties are an appealing mechanism for maintaining warm, ice-free conditions (Kiehl and Shields, 2013; Sagoo et al., 2013). Prior to humans, the major source of cloud condensation nuclei would have been biogenic trace gas emissions, sea salt, and dust particles (Kump and Pollard, 2008), limiting the number of cloud condensation nuclei relative to the modern day. Low numbers of cloud condensation nuclei would decrease the number of cloud droplets and increase mean droplet size, as is seen today in pristine areas over the ocean (Kiehl and Shields, 2013). These changes in liquid clouds cause a decrease in total clouds due to increased precipitation efficiency, and an increase in absorbed solar radiation due to decreased cloud reflectivity.

We suggest that future studies explore additional warming mechanisms. One is reduced transpiration in Cretaceous vegetation, resulting from reduced vein densities in angiosperms and high abundances of gymnosperms and ferns (Boyce and Lee, 2010; Feild et al., 2011; Wing et al., 2012). Reduced transpiration can cause surface warming by increased sensible heat, and increase absorbed surface radiation through reduced liquid clouds (Boyce and Lee, 2010; Upchurch et al., 1999).

Our results corroborate earlier studies indicating that warm poles and reduced latitudinal temperature gradients can be maintained if equatorial temperatures are greater than those of the modern day (Pagani et al., 2014, and references therein). Our equator-to-pole temperature difference of ∼25 °C and latitudinal temperature gradient of ∼0.4 °C per 1° latitude are also similar to those of the Eocene. Our study adds to a growing body of evidence that past greenhouse climates were characterized by warmer tropics and significant equator-to-pole temperature differences, and reinforces the suggestion that pre-anthropogenic cloud properties may be an important overlooked component of past global warming (Kiehl and Shields, 2013).

We thank Kenneth MacLeod, Robert Spicer, Marina Suarez, Greg Ludvigson, and Mark Pagani for helpful discussions. Shields was supported through a grant from the National Science Foundation (NSF) EAR Sedimentary Geology and Paleobiology program. Scherer was supported by NSF grant ESI-9731321 to Joseph Koke and Dana García (Texas State University, San Marcos). The National Center for Atmospheric Research is supported by the National Science Foundation.

1GSA Data Repository item 2015238, Table DR1 (paleotemperature data), explanation of methods, recalculated leaf margin temperatures, Vilyui Basin paleogeography, evaluation of model output, and sources of data, is available online at www.geosociety.org/pubs/ft2015.htm, or on request from editing@geosociety.org or Documents Secretary, GSA, P.O. Box 9140, Boulder, CO 80301, USA.