A proxy-model comparison for mid-Pliocene warm period hydroclimate in the Southwestern US

Hydroclimate proxy reconstructions and paleoclimate models of the midPliocene warm period provide insight into how, under a moderate greenhouse warming scenario, Earth-system feedbacks may impact regional hydroclimate. However, in the Southwestern United States there is discord between these two types of information: proxy data have been interpreted to indicate much wetter conditions, while the most recent generation of mid-Pliocene warm period climate models simulates drying. We use a water and energy balance framework to directly compare paleoclimate model output to a refined compilation of proxy records of the presence and areal extent of mid-Pliocene lakes. Within this framework, we quantify uncertainties in the proxy system model parameters and in the interpretation of available proxy records. We find that despite these significant uncertainties, most paleoclimate models simulate a regional balance between precipitation and evaporative demand that could not have sustained the extent of recorded lakes from this time. Moreover, the extensive lakes included as boundary conditions in mid-Pliocene warm period climate models are inconsistent with the regional climate simulated by those same models. This study identifies and quantifies the remaining unknowns in our picture of regional mid-Pliocene warm period hydroclimate, with implications for analyses of climate dynamics during this time.


Introduction
1 Climate models project that global warming and regional drying in the Amer-2 ican Southwest will exacerbate ongoing problems of water scarcity, drought, 3 and wildfire in coming decades (e.g., Seager and Vecchi, 2010;Williams et al., 4 2020), but natural variability on decadal timescales muddles our predictions 5 of the timing and magnitude of anthropogenic changes. Past climate states, 6 as understood through proxy reconstructions and paleoclimate models, provide 7 additional insight into how regional hydroclimatic conditions change over long 8 timescales, and into the forcings and mechanisms responsible for these changes. The mid-Pliocene warm period 3.3-2.9 Ma is the most recent example of the 10 long-term Earth-system response to elevated warmth and near-modern CO 2 11 concentrations (Tierney et al., 2020); consequently, research efforts in proxy 12 reconstruction and paleoclimate modeling have focused on this time period and 13 region.
14 In contrast to climate model projections of future drying, published recon-15 structions of the mid and late Pliocene (3.6-2.6 Ma) hydroclimate indicate that 16 the Western US was generally wetter than present by multiple measures. Com-17 pilations of Pliocene flora and faunal records, as well as stable isotope data, 18 have been interpreted as evidence for higher-than-present mean annual precip-19 itation (Molnar and Cane, 2007;Salzmann et al., 2008;Winnick et al., 2013). 20 The presence and size of pluvial lakes record climate-driven changes in water 21 availability over time, so records from outcrops, shorelines, and cores of these 22 lakes provide another archive of past climate conditions. Today, the internally- 23 draining Great Basin region in the Western United States includes a few mod-24 estly sized lakes, but published records indicate more extensive lakes during 25 the Pliocene between 3.6-2.6 Ma (Pound et al., 2014). Under conditions of in-26 creased warmth and higher evaporative demand, these lakes could have only 27 been sustained by higher-than-modern precipitation (Ibarra et al., 2018). 28 The Mediterranean-type climate of the western US is shaped by dynamical 29 processes associated with the subtropics and midlatitudes: subtropical highs 30 bring hot, dry summers, while extratropical storm tracks bring most of the re-31 gion's precipitation during the winter (Seager et al., 2019). The climate mod- Pliocene (e.g., Wara et al., 2005;Fedorov et al., 2006), others attribute wetter-42 than-modern conditions in western North America to extratropical teleconnec-43 tion patterns similar to those associated with modern El Niño events (Molnar 44 and Cane, 2007;Goldner et al., 2011;Winnick et al., 2013). Although this work 45 has been cited to explain mid-Pliocene warm period conditions and analogize 46 to future warming (Ibarra et al., 2018;Tierney et al., 2020), the magnitude of 47 tropical Pacific sea surface temperature (SST) gradients in the Pliocene is still 48 disputed: some analyses of proxy data argue for only modest reductions in the 49 zonal gradient (Zhang et al., 2014;O'Brien et al., 2014;Tierney et al., 2019), 50 while others argue for considerable reductions (Ravelo et al., 2014;Wycech 51 et al., 2020;White and Ravelo, 2020). Meanwhile, Pliocene climate models are 52 similarly equivocal: both older (Brierley et al., 2015) and more recent gener-53 ations of models show a general decrease in El Niño-Southern Oscillation am-54 plitude and a slight reduction in the zonal SST gradient, but do not agree on 55 whether there was a shift to an El Niño-like mean state (Brierley et al., 2015; western North America, leading to a regional decrease in extreme precipitation 69 and drier conditions in the annual mean (Menemenlis et al., 2021). Taken to-70 gether, modeling studies of the mid-Pliocene hydrologic cycle indicate that the 71 hydroclimate of the Western United States is sensitive to dynamical changes in 72 tropical and extratropical circulation patterns, but that the interplay between 73 these processes during the mid-Pliocene warm period is still ambiguous.

74
The Pliocene climate was more stable than the Pleistocene, but the mid- 75 Pliocene nonetheless experienced variability due to changes in Earth's obliq-76 uity and precessional cycle (Haywood et al., 2002). Orbital changes would 77 have influenced temperatures, precipitation patterns, and seasonality through 78 feedbacks from ice sheets and vegetation (Willeit et al., 2013;Haywood et al., 79 2013a;Prescott et al., 2014). Given the sensitivity of major spatial features 80 of modeled midlatitude terrestrial hydroclimate to such processes (e.g. , Feng 81 et al., 2021;Chan and Abe-Ouchi, 2020;Menemenlis et al., 2021), it is impor-82 tant to consider how regional hydroclimate might have changed over orbital 83 timescales during the mid-Pliocene warm period. While the lack of precise tem-84 poral constraints on terrestrial proxies poses a challenge for model-data com-85 parison (Haywood et al., 2013a;Salzmann et al., 2013), the timing of pluvial 86 lakes can be assessed with relative accuracy using a combination of paleomag-87 netic and stratigraphic methods. In cases where proxy records are dated with 88 < 10 kyr accuracy, new climatic interpretations are made possible. For example, 89 Knott et al. (2018Knott et al. ( , 2019 (Haywood et al., 2020), largely driven by a decrease in cool-season precipita-107 tion (November-April, see Figure S1). This leaves a discrepancy between the 108 general evidence for wetting in the Pliocene western US, with widespread lakes 109 in the region included as boundary conditions in PlioMIP2 models (Pound et al.,110 2014; Dowsett et al., 2016), and the drying simulated by the PlioMIP2 multi-111 model ensemble.

112
To understand the nature of and influences on Pliocene hydroclimate in the 113 US Southwest, we first must understand the (dis)agreement between existing 114 4 proxy reconstructions and paleoclimate model output. The "PMIP triangle" de-115 scribes three broad causes of model-data discrepancies: uncertainties in proxy 116 data, climate model boundary conditions, and climate model physics (Haywood 117 et al., 2013b(Haywood 117 et al., , 2016. We ask the following questions. Given unresolved orbital 118 variability in proxy data, does the mismatch between proxies and models stem 119 from differences in the time periods captured by each? Or do PlioMIP2 mod-

127
2.1. Regional setting 128 We focus on a 93,000 km 2 area of the Great Basin covering much of East-129 ern California and some parts of Western Nevada (see Supplement Section 1.1).

130
This "South Great Basin" area, shown in Figure 1, encloses a group of contigu-131 ous watersheds including Owens, China,Searles,Panamint,and Death Valleys,132 which, under wetter (i.e., Last Glacial Maximum) conditions, formed an inter-133 connected system of lakes and rivers (Reheis et al., 2014;Knott et al., 2019). At 134 present, the South Great Basin experiences an arid desert and steppe climate, 135 bordered to the northwest by the warm temperate climate of the Sierra Nevada.

136
Most precipitation arrives in the wintertime ( Figure S1). The ensemble of ten 137 PlioMIP2 climate models used in this study predicts a 3.4 • increase in tempera-   152 We assemble a compilation of South Great Basin lakes that existed between  (Table S1). For each lake, we gather wet scenario includes perennial lakes at their highstand as well as "ephemeral" 175 lakes, which meet condition 1 but not condition 2, and "potential" lakes. Poten-176 tial lakes include those for which there are no Pliocene outcrops or drill cores, 177 but geophysical data indicate a deep basin fill of probable Pliocene age (e.g.,

178
Owens Lake). Potential lakes also include those with outcrop evidence of lacus-179 trine conditions, but poor dating (e.g., Mono Lake). The difference between the 180 wet and dry scenarios is thus a measure of uncertainty, reflecting uncertainty in 181 dating and lake extent, as well as limitations of interpretation and extrapolation 182 based on available data. Further details of dating and lake area estimation for 183 each lake are given in Table S1. To directly compare between the proxy compilation and climate model data, we update the proxy-system model described in Ibarra et al. (2018). This model assumes a simplified steady-state balance between volumetric fluxes of water into and out of a system of pluvial lakes: This water balance can be expressed in terms of basin area and lake area, as follows: where P is precipitation, k run is the fraction of P converted to runoff, A B is the area of a terminal basin, A L is lake area, and E L is lake evaporation. Rearranging equation 1, 6 We use the Priestley-Taylor equation (Priestley and Taylor, 1972) to determine E L . The latent heat flux over the surface of a lake, expressed as LE L , where L is the latent heat of evaporation and E L is the rate of evaporation, is determined as follows: The constant α is empirically determined. R N is the net downward radiation flux at the surface, which can be expressed as the sum of the surface radiation fluxes R s,i − R s,o + R l,i − R l,o , where the subscripts s, l, i, and o denote shortwave, longwave, incoming, and outgoing radiation, respectively. ∆ is the temperature-dependent slope of the saturation vapor pressure curve in kPa K −1 : where T is temperature. γ is the psychrometric constant, which is elevationdependent and which we calculate following Allen et al. (1998): where c p is the specific heat of water at constant pressure, is the ratio of the molecular weights of water vapor and dry air, λ is the latent heat of vaporization of water, and atmospheric pressure p depends on elevation z. Pressure p is approximated by: where p 0 and T 0 are reference pressure and temperature, R is the ideal gas 186 constant, g is gravitational acceleration, M is the molar mass of air, and Γ is the 187 environmental lapse rate 0.0065 K m −1 .

Parameter uncertainty 246
Each model parameter introduces some uncertainty to our estimate of A L /A B .

247
To quantify the combined uncertainty due to these parameters, we estimate 248 the uncertainty in each parameter, then propagate these uncertainties using a States, E P scales with temperature at a rate of ∼1.8-2.8% across models, with a 294 multi-model mean value of approximately ∼2.25% (Scheff and Frierson (2014) 295 Figure 11). We use these ranges-0.9-1.6% and 1.8%-2.8% for R N vs. T and 296 E P vs. T respectively-to represent uncertainty.   3.6%, and 18.7%, respectively. This range of areas suggests a slightly to much 301 wetter water balance than the modern, where perennial lakes cover 0.5% of the 302 basin and perennial and ephemeral lakes combined cover 1.5% of basin area. 303 We note that our dry scenario does not unequivocally represent an absolute min-304 imum; since some lakes are reconstructed from cores taken at a single location, 305 it is possible that they could have recorded lakes with areas smaller than those 306 included in our dry estimate. This is nevertheless unlikely given the presence of 307 a number of perennial 3.2 Ma lakes in areas that are presently dry. In fact, our 308 dry scenario is likely conservative, since 1) we omit Owens Lake despite Owens it assumes that all lake highstands occurred simultaneously. Further details of 315 dating, lake area estimation, and paleoenvironment for each lake can be found 316 in Table S1.  Table S2). The multi-model in ω that would bring models into closer agreement with proxy reconstructions.

376
It is therefore unlikely that a change in the range of ω values across the South 377 Great Basin would narrow the discrepancy between A L /A B indicated by our 378 proxy compilation and by climate models. 379 Figure 3: Contours of A L /A B for the South Great Basin from our proxy compilation on corresponding Pliocene-pre-industrial changes in surface temperature and precipitation. Red, blue, and purple contours represent A L /A B from our dry, intermediate, and wet scenarios respectively. Thick solid lines contour the mean, thin solid lines contour the 25th and 75th percentiles, and dotted lines contour the 5th and 95th percentiles. Colored dots mark (∆T, ∆P) for each PlioMIP2 climate model, and the black star marks the multi-model mean.

380
The considerable differences between our dry, intermediate, and wet scenar- and all fall well below the 90% confidence interval of the intermediate scenario.

389
This indicates that uncertainties in the interpretation of proxy data, though sub-390 stantial, are not solely responsible for model-data discrepancy in this region.

391
Our analysis therefore suggests that even during the narrow KM5c interglacial 392 time-slice, the water balance in the South Great basin was at least slightly wetter 393 than models predict, and potentially much wetter.  Figure S5), and also because the regional cli-400 mate simulated by PlioMIP2 models is too dry (Figure 3). This inconsistency 401 is notable not only from the perspective of model evaluation, but also because 402 these boundary conditions can in turn impact the regional modeled climate.

403
Second, we speculate that errors in more remote boundary conditions may forcing similar to present-day (Haywood et al., 2013a). In this time slice, cli-418 mate models simulate drier conditions than indicated by proxy data. These 419 proxy data, in turn, may be interpreted to suggest a water balance either simi-420 lar to or much wetter than the present-day. It is therefore not obvious whether

430
In the American Southwest, we compared mid-Pliocene warm period cli-431 mate model output to lacustrine proxy records from the same period, with an 432 emphasis on quantifying the uncertainties in the interpretation of proxy records. 433 We assembled a refined compilation of proxy-recorded lakes, calculating areas 434 for a conservative "dry", an "intermediate," and a generous "wet" scenario. We 435 used a proxy-system modeling framework to directly compare between basin-436 normalized lake areas from our proxy compilations, and paleoclimate model-  Stepanek, C., Tan, N., Zhang, Q., Zhang, Z., Wainer, I., and Williams, C. J. R.

Defining the South Great Basin boundary
The South Great Basin perimeter is defined by the outer boundaries of inwardly-draining basins from the HydroBASINS database (Lehner and Grill, 2013; available from hydrosheds.org). At the northeastern boundary of the basin, we include a sub-region of an adjacent basin, since in this area a recent USGS groundwater model (Brooks et al., 2014) predicts subsurface water to flow toward Clayton Valley, where one of our Pliocene proxy sites is located (see Figure 1 in main text).

Finding modern lake areas
We draw on published maps, crowdsourced Open Street Map data (Open Street Map, 2021; Open Street Map data copyrighted OpenStreetMap contributors and available from openstreetmap.org), and Google satellite imagery (Google, 2021; Google Satellite Imagery ©2021 TerraMetrics, Map data ©2021Google) to produce GIS shapefiles of modern perennial and seasonal lakes in the South Great Basin. The two largest modern perennial lakes are Mono Lake and Owens Lake. Although Owens Lake is dry today, it existed perennially until the 1920s, when the city of Los Angeles diverted water for human use (Reheis, 1997;Smith and Street-Perrott, 1983). There are also a number of smaller perennial lakes on the western side of the South Great Basin. Smith (1984) mapped present-day lakes and seasonal playas; we use their Figure 1 as an initial reference for the locations of seasonal playas, then trace more precise shapes using Open Street Map and Google satellite data. In addition to those mapped in Smith (1984), we include several additional playas known to contain water on a seasonal basis.
We use output from 10 models participating in the Paleoclimate Model Intercomparison Project, Version 2 (PlioMIP2, Haywood et al. 2020). The modeling groups have archived monthly output for precipitation and surface temperature at the PlioMIP2 data repository at the University of Leeds. To find Pliocene minus pre-industrial anomalies in precipitation and surface temperature, we took the difference between the "Eoi400" and "E280" runs for each model. For additional detail regarding PlioMIP2 boundary conditions, experimental design, and results, see Dowsett et al. (2016), Haywood et al. (2016), andHaywood et al. (2020). For a summary of and references to the models used in this study, see Table S2.