Past terrestrial hydroclimate sensitivity controlled by Earth system feedbacks

Geologic evidence indicates major reorganizations of subtropical hydroclimate during the mid-Piacenzian Warm Period (3.3–3.0 Ma), including widespread lakes and reduced dust across North Africa and more mesic vegetation in South and East Asia, implying large increases in P–E despite mid-Pliocene pCO2 (~400 ppm) comparable to today. While elevated CO2 can increase precipitation via tropospheric moistening and enhanced land–sea contrasts, evaporation also rises, leaving terrestrial water balance responses uncertain. Previous model results and future projections often follow a wet-gets-wetter, dry-gets-drier paradigm with sensitivity to land warming patterns, land–sea contrast, SST patterns, and land-surface feedbacks. However, future projections under mid-range scenarios show minimal P–E increases in the Sahel and East Asia by late 21st century, contrasting with mid-Pliocene proxy evidence. The study asks why mid-Pliocene subtropical continents were wetter and whether CO2 forcing alone or longer-term Earth system feedbacks (vegetation and ice sheets) controlled this state. It evaluates competing hypotheses (El Niño-like mean state, weakened Hadley circulation) and investigates mechanisms using PlioMIP2 multi-model simulations and targeted CESM2 experiments.

The paper reviews prior findings that CO2-driven warming tends to produce a wet-gets-wetter, dry-gets-drier response modulated by land–sea thermal contrast, SST patterns, and land feedbacks. Two leading hypotheses for wetter subtropics in past warm periods are: (1) an El Niño-like Pacific mean state that strengthens and shifts the subtropical jet equatorward, enhancing transient eddy moisture convergence; and (2) a weaker Hadley circulation due to reduced meridional SST gradients, decreasing zonal mean moisture divergence from the subtropics. Advances in Earth system modeling and boundary conditions (PlioMIP2) have altered simulated Pliocene states, including SST patterns, polar warmth, AMOC, and precipitation. Prior equilibrium simulations with only mid-Pliocene CO2 failed to reproduce moist subtropical continents, suggesting additional boundary condition effects beyond direct radiative forcing are important.

• Multi-model ensemble: Used 13 PlioMIP2 coupled GCMs (last 100 years averaged) to compute mid-Pliocene anomalies relative to each model’s preindustrial control. Evaluated annual and boreal summer (June–September) δ(P–E), significance relative to model spread, and consistency across regions (Sahel 10°–20°N, 10°W–25°E; East Asia 20°–30°N, 80°E–100°E). Fourier decomposition of 600 hPa stationary wave streamfunction to identify dominant zonal wavenumber components.
• Proxy compilation: Built on existing Pliocene hydroclimate compilations and databases, identifying initially 64 records and, after age-model filtering requiring multiple control points, retained 62 records spanning 0–67°N and 23°W–172°E. Proxy types include sedimentology, palynology, floral/faunal indicators, and stable isotopes. Proxies were categorized as wetter, drier, or no change relative to late Quaternary/modern based on original interpretations.
• Proxy–model comparison: Converted model δ(P–E) to categorical outcomes (wetter/drier/no change) using percentage thresholds (1%–100% in 1% steps) relative to PI at proxy sites. Agreement assessed with Gwet’s AC statistic (robust to category imbalance), with significance estimated per established methods.
• Targeted CESM2 experiments: Performed sensitivity simulations to decompose responses to (i) elevated CO2 (F_CO2: 400 ppm), (ii) vegetation and ice sheet changes (F_vegice: biome redistribution, deglaciated areas per PRISM4), and (iii) geography/topography changes (F_geotop). Designed simulations E400, Eo400, Eo280, alongside published Eoi400 (full forcing) and E280 (PI), allowing calculation of R(F_all) and separation into R(F_vegice), R(F_geotop), R(F_CO2), and assessment of state dependence. Each run ≥300 years; last 100 years used for climatologies.
• Moisture budget diagnostics (MBD): Decomposed δ(P–E) into contributions from seasonal moisture tendency, zonal mean circulation changes, stationary wave dynamics, changes in tropospheric moisture content, interaction terms, and a residual capturing transient eddies and unresolved terms. Implemented with monthly climatologies due to data availability; also related stationary wave responses to surface temperature patterns via thermal wind diagnostics.

• PlioMIP2 models robustly simulate positive δ(P–E) across the Sahel and subtropical Eurasia, strongest in boreal summer (June–September), consistent with proxy indicators. Models that are wetter in the Sahel tend to be wetter in subtropical East Asia, indicating common drivers.
• Proxy–model agreement (Gwet’s AC) is significant for annual means across a wide range of thresholds and is driven by summer δ(P–E). Models such as IPSL-CM6A-LR and EC-Earth3.3 show expansive inland wetting; NorESM-L and GISS-E2-1-G show muted/mixed responses.
• CESM2 forcing decomposition: Vegetation and ice sheet changes (F_vegice) explain ~78% (2.2 mm day^-1) of the regional mean δ(P–E) under full forcing (F_all = 2.8 mm day^-1) in the Sahel and subtropical East Asia. Contributions from CO2 alone (F_CO2) and geography/topography (F_geotop) are small, and only F_vegice reproduces proxy–model agreement comparable to F_all.
• Thermodynamic and dynamic contributions: Positive δ(P–E) primarily arises from increased tropospheric moisture content (δ(P–E)Q) and stationary wave dynamics (δ(P–E)ψ). Contributions from zonal mean circulation changes (δ(P–E)v) and interaction terms are negligible. Intermodel spread in δ(P–E) scales with δ(P–E)ψ and δ(P–E)Q.
• Stationary wave structure: Zonal wavenumber-1 stationary wave accounts for ~67% of wave energy (0–90°N), featuring cyclones over W. Europe and N. Africa and an anticyclone over the N. Pacific. Associated rotational winds create moisture transport corridors into North Africa (from tropical Atlantic) and into South Asia/East Asia (from Indian and Pacific Oceans). This pattern links to surface warming via thermal wind and is reproduced under F_vegice.
• Moistening shares: Globally, F_vegice accounts for ~50% of ΔTs (2.7 °C of warming under F_all) and 58% (0.45 g/kg) of tropospheric moistening (100–1000 hPa). In 20°–30°N, F_vegice explains 61% (0.56 g/kg) of moistening. F_CO2 produces 45% of global ΔTs (2.4 °C) but only 31% of global moistening and 26% in 20°–30°N; F_geotop contributions are minor (slightly larger in northern subtropics at ~13% of moistening, 0.11 g/kg).
• Mechanism distinction: Changes are not driven by an ENSO-like response or a weakened Hadley circulation; δ(P–E)RES and δ(P–E)v are small, storm tracks and subtropical jet weaken rather than shift equatorward, and the ENSO precipitation fingerprint (dry N. Africa, wet SE Asia) is absent. The response reflects a pattern effect from continental greening and ice loss that induces zonally heterogeneous warming and stationary waves.
• Model skill vs sensitivity: Ability to reproduce proxy hydroclimate scales with Earth System Sensitivity (ESS, quantified by global mean warming) rather than Equilibrium Climate Sensitivity (ECS).

The study demonstrates that wetter mid-Pliocene conditions over the Sahel and subtropical East Asia were not an immediate result of elevated CO2 radiative forcing but predominantly reflected long-term Earth system feedbacks—loss of high-latitude ice sheets and widespread continental greening. These boundary changes increased tropospheric humidity and, crucially, generated a zonal wavenumber-1 stationary wave linked to the spatial pattern of surface warming, enhancing inland moisture transport and convergence. This mechanism contrasts with prior hypotheses emphasizing ENSO-like SST states or a weakened Hadley circulation and differs from an ITCZ-shift framework that focuses on zonal mean energy budgets. The findings reconcile proxy evidence of moist Pliocene subtropics with muted subtropical P–E changes in near-future projections by highlighting differing timescales: near-future climates reflect short-term CO2 responses and internal variability, whereas Pliocene hydroclimate captures slow feedbacks from vegetation and ice sheets. The scaling of proxy–model agreement with ESS underscores the importance of incorporating Earth system feedbacks to understand paleohydrological and, by implication, long-term future hydroclimate responses.

The paper shows that mid-Pliocene subtropical hydroclimate sensitivity was controlled by Earth system feedbacks, especially vegetation changes and ice sheet retreat, which amplified tropospheric moistening and induced stationary wave patterns that delivered moisture inland to the Sahel and subtropical East Asia. CO2 forcing alone and tectonic/geographic changes had comparatively minor effects on terrestrial P–E in these regions. Model skill in reproducing proxy hydroclimate patterns scales with Earth System Sensitivity rather than ECS, emphasizing the need to represent slow feedbacks. Future research should: (1) integrate prognostic vegetation and dynamic ice sheets more broadly in paleoclimate and long-term projection frameworks; (2) improve proxy quantification and spatial coverage to constrain seasonal hydroclimate; (3) obtain higher-frequency model outputs to better resolve moisture budget residuals and transient eddies; and (4) explore regional differences (e.g., North Pacific, western North America) where other mechanisms, including eddy transports and ENSO-like SSTs, may be more influential.

• Proxy data are largely qualitative or semi-quantitative and compare average mid-Pliocene conditions to modern/late Quaternary; many records lack seasonal resolution and may not capture orbital variability. The final filtered proxy set (62 records) balances spatial coverage and dating precision but retains uncertainties.
• PlioMIP2 experiments target a specific Pliocene interval (MIS KM5c) with present-day orbital parameters by design; orbital effects are not included, whereas most proxies do not resolve orbital-scale changes.
• Moisture budget residual terms combining transient eddies, high-frequency variability, and topographic pressure effects were not explicitly calculated due to lack of high-frequency outputs; they are inferred as residuals.
• Vegetation and ice sheet feedbacks are not fully prognostic in most PlioMIP2 models; many simulations prescribe vegetation, potentially limiting feedback realism.
• The decomposition assumes approximate decorrelation between CO2 and geographic/topographic forcings; while tested under moderate forcing, some state dependence and interaction effects may persist.